National Conference

On
Recent Trends in Manufacturing
Science and Technology

E-proceedings
RTMST-2013

Date: 18th and 19th April, 2013

Venue: NITTTR, Kolkata

Organized by

Department of Mechanical Engineering

National Institute of Technical Teachers’
Training & Research
(Sponsored by MHRD, Govt. of India)
Block- FC, Sector-III, Salt Lake City,
Kolkata- 700106, West Benga
 Preface
Modern manufacturing encompasses an ever increasing variety of machines and
process which becomes a challenge to manufacturing engineers. Selection of
most economic combination of processes to make a product of high quality at
the right price has become more important than ever. The recent trends in
manufacturing indicate that the role of innovation cannot be ignored to create a
product with high productivity with less cost. Design, production and
development of new techniques are the three major engineering activities those
are responsible to produce a product with high quality. On the other hand,
development of new materials and alloys with complex shapes for specific uses
eventually led to the growth of various non-traditional machining processes.
Moreover, sometimes the combination of material properties and the job
dimension is such that it makes the use of conventional processes impossible.
The newer manufacturing processes have made possible to produce jobs which
were previously impossible with conventional processes. However the recent
emphasis on miniaturization has led to the development of micro
manufacturing.

The aim of the National Conference on Recent Trends in Manufacturing

Science and Technology is to bring together academicians, researchers,
technologist, and industrialist to a common platform for presentation and
discussion on various aspects of modern manufacturing processes and systems.

We gratefully acknowledge the reviewers of the papers, members of Advisory

Committee and Organizing Committee for their valuableadvice and cooperation.
We are also thankful to our research scholars and staff members of the
Mechanical Engineering Department for their help at various stages of
organizing the conference. Finally, we convey our sincere thanks to the authors
for their valuable cooperation and contribution of research papers to the
conference.

Steering Committee

RTMST 2013
 अशोक ठाकु र भारत सरकार
स चब मानव संसाधन वकास मं ालय
उ चतर श ा वभाग
ASHOK THAKUR
SECRETARY
Government of India
Ministry of Human Resource Development
Department of Higher Education

16th April 2013

MESSAGE

I feel happy to note that National Institute of Technical Teachers’ Training and
Research, Kolkata is organizing a National Conference on “Recent Trends in
Manufacturing Science and Technology” on 18th – 19th April 2013 at the Institute
premises in Salt Lake, Kolkata.

This conference proposes to address the various developments that have taken
place over the last 60 to 70 years in manufacturing sector and their importance in
developing a comprehensive manufacturing plan for a country like ours where 65% of
the population is below the age of 35.

Although management and marketing play a positive role in enhancing the

prosperity of a country, we tend to undermine the role of manufacturing which creates
more job than any other sector. It is, therefore, very relevant for the National
Institutes of repute to leverage the manufacturing sector for creating jobs for our
youth.

I wish the conference a grand success and fervently hope that the deliberations
in the conference can be used as necessary inputs for developing a nationwide
manufacturing strategy that will create gainful employment for the young people of the
country.

Sd/-
(Ashok Thakur)
 अ ण कु मार साहा राष ् य तकनीक श क श ण एवं अनुसंधान संसथान,
् कोलकाता
बलॉक-एफ.सी.,
् से क्टर-3, सॉलट् लेक सट , कोलकाता-700106
अ य
Arun Kumar Saha National Institute of Technical Teachers’
CHAIRMAN Training & Research, Kolkata
Board of Governors Block-FC, Sector-III, Salt Lake, Kolkata-700 106
Tel: (033) 2337-0937 Fax: (033) 2337 6331

16th March 2013

MESSAGE

I am pleased to learn that a National Conference on “Recent Trends in

Manufacturing Science & Technology” is being organized by National Institute of
Technical Teachers’ Training and Research, Kolkata during 18th – 19th April 2013.

The various themes of the Conference are of utmost importance in the present
scenario of technical education.

Although a tremendous amount of advancement has taken place in manufacturing

technology, it appears that these developments and their importance to enhance the
prosperity of the nation have not been appropriately addressed till now.

I hope that the conference will, inter alia, discuss the evolution of
manufacturing science and its relevance on enhancing the quality of life of mankind.

I convey my sincere greetings and wish a grand success for the Conference.

Sd/-
(Arun Kumar Saha)
Chairman
Board of Governors
NITTTR, Kolkata
 ोफेसर एस. के. भ ाचाय, राष ् य तकनीक श क श ण एवं अनुसध
ं ान संसथान
् ,कोलकाता
बलॉक-एफ.सी
् ,.सेक्टर ,3- सॉलट् लेक सट ,कोलकाता700106-
नदे शक
Prof. S K Bhattacharyya National Institute of Technical Teachers’
DIRECTOR Training & Research, Kolkata
Block-FC, Sector-III, Salt Lake, Kolkata-700 106
Tel: (033) 2337-0937 Fax: (033) 2337 6331

16th March 2013

MESSAGE
NITTTR Kolkata is one of the four institutes which were established around 1965
by the Ministry of Higher Education (MHRD) of the Government of India. One of thae
objectives of setting up this Institute is to find out and implement ways and means to
improve the technical education system of the country in general and the eastern and north
eastern states of the country in particular.

In order to meet the said objective, the Institute organises periodically but on a
regular basis, conferences, seminars, workshops etc. so that the academicians and
practitioners exchange their views on the relevance and utility of any major development
that takes place in teaching learning system and/or technological field. This National
Conference on “Recent Trends in Manufacturing Science & Technology” (RTMST-2013) is
one of those events that the Institute has decided to conduct in the current year that is
2013- 2014 AD.

There are several reasons for selecting advancement in manufacturing system as

the theme of the conference. Firstly it is the tremendous technological growth in the
industry sector since the first industrial revolution that resulted in rapid improvement in
our living standards. Secondly, Cost, Quality, Delivery and Flexibility have been identified
as the competitive priorities in a market place. The ability of a firm to compete on the
basis of these competitive priorities depends however on its manufacturing set up. Lastly,
manufacturing sector has failed so far to act as an engine of growth in the Indian economy
and as a source of new employment. The net result is that this sector employs only 9 % of
India’s working population and contributes about 16 % of India’s GDP.

With changing global realities, the manufacturing sector will however need to be the
bulwark of employment creation in order to realise India’s long touted demographic
dividend.

I wish all success to the conference (RTMST-2013) and hope that the results of the
conference will prove beneficial for the skill and employment growth in our Country.

Sd/-
(Prof. S K Bhattacharyya)
Director
NITTTR, Kolkata
 Organizing committee
Chief Patron
Arun Kumar Saha
Chairman, BOG, NITTTR, Kolkata

Patron
Prof. Sanjay Kumar Bhattacharyya
Director, NITTTR, Kolkata

Steering Committee

 Dr. SamiranMandal, Professor & Head, Mechanical Engineering, NITTTR

 Dr. Dipankar Bose, Professor, Mechanical Engineering, NITTTR
 Nirmal Kumar Mandal, Associate Professor, Mechanical Engineering, NITTTR
 Sukanta Kumar Naskar, Associate Professor, Curriculum Development Centre,
NITTTR

Advisory Committee

 Dr. U. S. Dixit, Professor, ME Department, IIT, Guwahati

 Dr. M. K. Tiwari, Professor, Industrial Engineering& Management, IIT, Kharagpur
 Dr. Sadhan K Ghosh, Professor & Head, ME Department, Jadavpur University
 Dr. Siddhartha Ray, Professor, ME Department, Heritage Institute of Technology
 Dr. BijanSarkar, Professor, Production Engineering Department, Jadavpur University
 Dr. Bijoy Bhattacharyya, Professor, Production Engineering Department, Jadavpur
University
 Dr. AsishBandoyopadhyay, Professor, ME Department, Jadavpur University
 Dr. IndrajitBasak, Professor, ME Department, NIT, Durgapur
 Dr. S. C. Saha, Professor, Mechanical Engineering Department, NIT, Agartala
 Dr. B. K. Basak, Executive Director, Foundry Division, WESMAN Engineering Co.
Ltd.
 Dr. SubhasisBhaumik,, Associate Professor, Applied Mechanics & Aerospace
Engineering, BESU, Shibpur
 Dr. AmitRoychowdhury, Associate Professor, ME Department, BESU
 Syamal Mukherjee, Ex. General Manager, WESMAN Engineering Co. Ltd. & Training
Officer ,TCWCC, NITTTR, Kolkata
Members (NITTTR, Kolkata)
 Dr.R.Srinivasan, Professor & Head, Education
 Dr. SekharChakraborty, Professor & Head, Educational Planning & Management
 Dr. RanjanDasgupta, Professor & Head, Computer Science & Engineering
 Dr. UrmilaKar, Professor & Head, Curriculum Development Centre
 Dr.Anuradha De, Professor & Head, Science
 Dr. JagatJyotiMandal, Professor & Head, Civil Engineering
 Dr. Uday Chand Kumar, Professor & Head, CD & R Technology
 Dr. SubrataChattopadyaya, Associate Professor, Electrical Engineering
 Dr. S. K. Mandal, Associate Professor, Electrical Engineering
 Dr. Rajashri Roy, Associate Professor, Education
 Dr. Samir Roy, Associate Professor, Computer Science & Engineering
 Dr. SailendraNathMandal, Associate Professor, Science
 Dr. ShantunuBhanja, Associate Professor, Civil Engineering
 Dr. HabibaHussain, Assistant Professor, Education
 Dr. Sagarika Pal¸ Assistant Professor, Electrical Engineering
 Shri Rajeev Chatterjee¸ Assistant Professor, Computer Science & Engineering
 Ms. SheelaYadavRai, Assistant Professor , CD & R Technology
 Ms. MithuDey, Assistant Professor, Civil Engineering
 A. K. Chakraborty¸ Senior Administrative Officer
 Y Das, Production Officer, Learning Resource Centre
 Ms. ManishaChakraborty, Librarian
 List of Reviewers:

Name of the Reviewer

 Mr. Indrajit Basak

 Mr. Biswanath Doloi

 Mr. S Das

 Mr. Asish Banerjee

 Mr. Pradip Pal

 Mr. Subhasis Bhaumik

 Mr. Sukanta Naskar

 Mr. Souren Mitra

 Mr. Siddhartha Ray

 Dr. Dipankar Bose

 Mr. N. K. Mandal
 Contents
Preface

Secretary’s message

Chairman’s message

Director’s message

Organizing committee

List of reviewers

A. Recent trends in Material Processing Technologies Pages

1. Laser Welding of Polymers – Temperature

Field Modelling and Experimental Validation
Bappa Acherjee, Arunanshu Shekhar Kuar, Souren Mitra, Dipten Misra 1-10

2. Artificial neural network based monitoring of weld quality

in pulsed metal inert gas welding using wavelet packets of
current signal
Kamal Pal , Surjya K. Pal 11-26

3. Experimental Investigation Of Tig Welding Of

Austenitic Stainless Steel With Change In
Composition Of Filler Material
Nabendu Ghosh, Pradip Kumar Pal, Goutam Nandi and Ramesh Rudrapati 27-35

4. Additive Manufacturing of Casting Die with

Ideal Shapes of Gating System through Rapid
Tooling 36-45
Divyansh Patel, Abhilasha Singh, Puneet Tandon

5. Application Of Artificial Neural Networks For

Parametric Modelling Of Mig-Co2 Welding Process
Akash Kumar, Sudipto Chaki and Dipankar Bose 46-55

6. Prediction Of Process Parameters For Tig Welding

Using Back-Propagation Neural Networks
Soham Das, Dipankar Bose, Sudipto Chaki 56-65
B. Advancement of Machining Technologies

7. Mathematical Modeling Of Machining Parameters

For MRR In EDM Using Regression Analysis
Sandip Kunar, Abdul Halim Khan, Dr. D. C. Roy 65-74

8. Experimental Investigation and Optimization of

Controllable Parameters in CNC Wire-EDM
Kingshuk Mandal, Subrata Mahata, Dipankar Bose 75-82

9. Multi-Objective Optimization of Electrochemical

Deburring Process Using Grey-based Taguchi Method
M. Singha, S. Sarkar, S. Mitra and A. S. Kuar 83-91

10. An Investigation On Temperature Distribution Of An

Erw Pipe During Tig Welding
Tathagata Bhattacharya , Asish Bandyopadhyay, Pradip Kumar Pal 92-103

11. An Experimental Investigation on Dimensional Accuracy

and Burr Formation in Slot Milling Operation
A. Ghosh, A. RawanI, G. Maity, S. Majee and Santanu Das 104-113

12. Determining The Effect Of Grinding Parameters On

Surface Roughness In Traverse Cut Cylindrical
Grinding Process Using Taguchi Method
Ramesh Rudrapati, Pradip Kr. Pal Asish Bandyopadhyay 114-124

13. Machining of Cylindrical Surface by Electric Discharge

Machine
Sanasam Sunderlal Singh, P. K. Patowari 125-131

14. Modelling and Optimization of Thermal Energy of

CNC High Speed End Milling of EN 362
Indranil Chakraborty and N.K. Mandal 132-142

15. Initiation Time for Electrical Discharge through

Liquid Medium
Debasish Das, Debasish Nandi, Indrajit Basak 143-157
C. Development in Machine Tools and Tooling

16. Surface Modification of Graphite by EDM using

Powder Metallurgical Green Compact Tools
Ngangkham Devarani, P. K. Patowari 158-165

17. An Experimental Investigation of Optimum Flank

Wear of Carbide Inserts for Dry Turning of EN19
Tool Steel
Saurav Nandi, Nirmal Kumar Mandal 166-172

D. Innovation in Manufacturing automation

18. Adaptation of Compliant Multi Bar Mechanism for

Generation of Compliant Segment Motion
Dr. Debasish Dutta, Subhankar Das 173-187

19. Automatic Shape Recognition By Using Robot Vision

Technique
Sudip Kumar Mondal, Mandakinee Bandyopadhyay, Amit Kumar Shah 188-203

20. Obstacle Avoiding Wall Following Robot

Sagarika Pal, Arjita Das, Sanjeev Kumar 204-217

21. Hydroforming Setup And Its Experimental Results

Dwaipayan De, Asish Bandyopadhyay 218-226

E. Current trends in Micro Manufacturing

22. Parametric Optimization of Laser Microdrilling of

SIC-30BN Nanocomposite
N. Roy, A. S. Kuar, S. Mitra, B. Acherjee 227-235

23. Effect of Machining Parameters on the Surface

Roughness of Mg/Y2O3 Nano-composite
Sanjib Gogoi, Jitul Saikia, Naumeet Meena and Ashis Mallick 236-247

24. Micro-Hole Drilling On Thin Sheet Metals By

MICRO-EDM
Trisha Biswas, Amit Kumar Singh, P. K. Patowari 248-258
F. Improvement in Manufacturing Management

25. A Policy Making Framework For Resilient Supply

Chain Systems
Anupam Haldar, Amitava Ray, Debamalya Banerjee, Surojit Ghosh 259-268

26. Development of A New Shakeout Machine By Applying

Value Engineering: A Case Study
Aditya, Samiran Mandal 269-282

27. Image Based Quality Checking Of Industrial

Product Using Applied Robotics
Avirup Das, Rohit Kumar, Biplab Roy and Ganesh Ram 283-294

28. Reliability Analysis and Maintainability Aspects of a

Paint-Manufacturing Plant
Anujit Chanda and Sukanta Kumar Naskar 295-305

29. Optimal Precision Measuring Instrument Selection

Using Analytical Hierarchy Process
Dr. S. K. Deb, Mr. M. Bhuyan 306-315
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18th and 19th April, 2013, NITTTR, Kolkata

Laser Welding of Polymers – Temperature Field

Modeling and Experimental Validation

Bappa ACHERJEE
Department of Production Engineering, BIT Mesra, Deoghar Campus, Deoghar 815 142, India
E mail: a.bappa@yahoo.com, bappa.rana@gmail.com

Arunanshu Shekhar KUAR

Department of Production Engineering, Jadavpur University, Kolkata – 700 032, India
E mail: askuar@rediffmail.com

Souren MITRA
Department of Production Engineering, Jadavpur University, Kolkata – 700 032, India
E mail: souren.mitra@gmail.com

Dipten MISRA
School of Laser Science & Engineering, Jadavpur University, Kolkata – 700 032, India
E mail: dipten@gmail.com
________________________________________________________________________

Abstract: This paper presents a three-dimensional transient thermal model of laser

transmission welding of polymers with a moving laser beam. The configuration of the
contour laser welding process with lap joint geometry is used in this study. The main
objectives are to predict the transient temperature field as well as the weld dimensions as
a function of process parameters. All the major physical phenomena associated with the
laser transmission welding process, such as, heat radiation, thermal conduction and
convection heat losses are taken into account in the model development. The results
obtained for different welding conditions are compared to the experimental results. A
reasonable consistency is observed between the predicted and measured outputs.

Keywords: laser transmission welding, numerical simulation, finite element analysis,

temperature field, weld dimensions
________________________________________________________________________

1
 RTMST
1. Introduction
In laser transmission welding (LTW), a laser beam is directed at two overlapping
thermoplastic parts. The first part contacted by the laser beam should be primarily
transparent, and the second part should absorb the laser radiation. As the laser beam
strikes the surface of the transparent part, a fraction of the incident light is reflected and
the remaining light energy enters the bulk of the material. While the majority of the light
entering the transparent part is transmitted through the material, with the possibility of
significant scattering, a portion of the energy is absorbed. The laser light that is
transmitted through the transparent material strikes the surface of the absorbing material,
where a fraction of the energy is reflected and the remaining energy enters the bulk of the
material. Depending on the thickness and absorption coefficient of the absorbing material,
major portion of that energy is absorbed over a depth of that material, and a small portion
of the laser energy may transmit completely through both parts. [1, 2].

The laser energy in the near-infrared spectrum that is absorbed in the opaque
thermoplastic causes vibration of electron bonds [3]. This bond vibration causes localized
heating of the material. Using the laws of heat transfer, this heat is conducted within and
between both parts and transferred to the surroundings through convection and radiation.
As both of these components heat up the polymer will start to expand. The expansion of
the polymer is a result of molecular excitement and needs to be contained. If no
containment of the polymer is maintained during heating no weld will occur because the
polymers will not cross-link across the joint interface. Pressure must be applied to the
joint to both hold the part in position to ensure good enough contact with the melt for the
transparent material to be heated by conduction and restrict the heating polymer from
expanding. This restriction of expansion will allow the polymer to cross-link across the
joint interface and a weld to occur. When heated to temperatures above the melting point,
or melting range, the thermoplastic enters a liquid form. By intermixing or diffusion the
polymer chains may combine. After cooling and re-solidification a firm connection is
formed at the weld seam [4]. Principle of laser transmission welding is illustrated in Fig.
1.

Fig. 1. Principle of contour laser transmission welding

2
 RTMST

If properly performed and if the materials combination is suited for laser welding, the
laser transmission welding generates a strong and cosmetically perfect, almost invisible,
seam with little thermal load to the product. Furthermore, laser welding of polymers is
flexible, gentle and environmentally friendly [5].

The finite element method (FEM) is known to be a versatile tool for material processing
simulation. A number of attempts have been made to simulate the laser transmission
welding process using FEM. However, most of them neglected the effect of weld pressure
and/or thermal expansion during modeling the process. Becker et al. [6] studied the
heating phase of laser transmission welding of polypropylene using finite element
method. Ilie et al. [7] presented a study on effects of laser beam scattering phenomena in
a semi-transparent polymer induced by their compositions in the LTW process. Mayboudi
et al. [8] developed a two-dimensional heat transfer model of LTW of unreinforced nylon
6 in a T-like joint geometry. Van de Ven and Erdman [9] set a two-dimensional heat
transfer model to study the LTW process of polyvinyl chloride. A three-dimensional
thermal model of LTW solved with FEM is presented by Mayboudi et al. [10] for PA6
with lap joint geometry with a stationary laser beam. Coelho et al. [11] studied the beam
spot influence in high speed laser lap welding of thermoplastic films. Acherjee et al. [12]
presented a computational model for contour laser transmission plastic to metal welding
with a moving heat source. A further study by Acherjee et al. [13] investigated the effect
of carbon black on temperature field distribution during LTW of thermoplastics.

In this wok, an attempt has been made to develop a 3-D FE model to simulate laser
transmission welding process. ANSYS® parametric design language is used to create the
model. The developed model is used to study the evolution of temperature field during
the welding as well as the formation of weld bead. The results obtained for different
welding conditions are compared to the experimental results and reasonable agreement is
observed.

2. Development of the Model

The configuration of the contour laser welding process with lap joint geometry is used in
this study. The isometric view of the sample that is used for the present study is shown in
Fig. 2. Using the presence of loading and geometrical symmetries, the model is reduced
to cut the simulation time. This idealization of the model is adequate to represent the
problem to achieve study goals.

3
 RTMST

Fig. 2. Schematic of the sample with symmetry boundary conditions

(dimensions are in mm)
The following three-dimensional heat conduction equation is considered to model heat
transfer in laser transmission welding process that defines the temperature distribution
within the body. It is based on the energy conservation law, which balances the rate of the
internally generated heat within the body, body’s capacity to store this heat, and the rate
of thermal conduction to the boundaries:

T   (1)
 .c  ( k T )  qv
t

where, ρ is the material density (kg/m3), c is the specific heat (J/kg-K), T is the
temperature (K), t is the time (s), k is the thermal conductivity (W/m-K), qv is the rate of
internal heat generation (W/m3), and  is the gradient operator.

An appropriate volumetric heat generation term with Gaussian expression is employed to

model the laser beam absorption in the plastic parts and heat generation at the irradiated
zone. For the case of Gaussian laser flux distribution the rate of internal heat generation,
qv, is determined as:

 0 ; for transparent part

q(x, y, z,t)   (2)
(1Ra )KIa exp(Kza ) ; for absorbing part

where, Ra is reflectivity of the absorbing material, K is absorption coefficient (m-1) of the

absorbing material, za is the depth (m) within the absorbing materials, and Ia is the laser
intensity (W/m2) after passing through the transparent material, which can be further
expressed as:

4
 RTMST

Tt P  r2  (3)
Ia  exp  2 
 r02  r0 

where, Tt is the transmissivity of the transparent polymer part, P is laser power (W), r0,
the laser beam radius (radius of Gauss function curve), is the radius at which the intensity
has decreased to 1/e2 or 0.135 of its maximum value on the axis, and r is the radial
distance of any point on the surface of the material; r = √(xs2+ys2), where xs and ys are the
Cartesian coordinates of that point.

The heat loss from the material surfaces due to convection and radiation is realized by
imposing convection surface load with a temperature dependant convection-radiation
combined heat transfer coefficient. The convection and radiation boundary conditions are
combined into a single combined boundary condition as:
 
k(T)Tn  hr (Ts T0 ) (4)

where, hr is the combined heat transfer coefficient that can be expressed as:

hr  h   (Ts  T0 )(Ts2  T02 ) (5)

The values of hr are calculated at different temperatures and stored in a look-up table as
temperature dependant material property data. Convective heat transfer coefficient, h, is
assumed to be 5 W/m2-K [8]. Emissivity of polycarbonate is taken as 0.95[14]. Further
details about the FE modeling and implementation of moving heat source are described in
[13].

Polycarbonate is considered as work material for developing and validating the model.
Temperature dependent material properties of polycarbonate are used in the finite element
model.

A subroutine is implemented in ANSYS parametric design language for moving the laser
beam and changing the boundary conditions, accordingly. The transient heat transfer
model is then solved using the Newton-Raphson iteration method based on the difference
between the externally applied forces and the internally calculated forces. The load moves
with time. Thus, a nodal result history is obtained as a function of time.

3. Boundary Condition Sensitivity Analysis

The influence of thermal boundary conditions on the temperature field distribution is

investigated herein. Radiative and convective heat losses from material’s surfaces are
considered as thermal boundary conditions. The convection and radiation boundary
conditions are combined into a single combined boundary condition. The values of
combined heat transfer coefficient are calculated at different temperatures using Eq. 4. It
is noteworthy to mention that the combined heat transfer coefficient is a function of
convective heat transfer coefficient (h) and emissivity of the material (ε). The boundary
condition sensitivity on temperature field distribution is tested for three different

5
 RTMST

Fig. 3 Influence of thermal boundary conditions on the simulated temperature (power = 5 W,

welding speed = 20 mm/s, beam diameter = 1.5 mm and carbon black content in absorbing
polycarbonate = 0.2 % wt.)
conditions, as: (a) h = 0 W/m2.K, ε = 0, (b) h = 5 W/m2.K, ε = 0 and (c) h = 5 W/m2.K, ε =
1. It can be observed from Fig. 3 that for the thermal boundary conditions of (a), (b) and
(c), the maximum temperature achieved at weld interface is almost same, and the
difference between the temperature drops is about 6 %. It implies that the effect of heat
conduction is predominant on temperature field distribution. For a very short welding
time, the energy loss through radiative and convective heat losses from the small heated
zone is very small.

4. Experimental Validation
A Coherent FAP-diode laser system is used for experimental work. The maximum optical
power and system wavelength are 30 watt and 809.40 nm, respectively. The FAP-system
optical radiation is delivered via an SMA 905 connector, with a multi-mode, single core
fiber optic transport of 800 µm diameter. The output power of the laser system is
calibrated using a Gentec-EO SOLO-2 laser power & energy meter of 1 mW resolution.

The workpieces are moved via a CNC X-Y table coordinated with the motion system and
computer interface. A laser based Micro-Epsilon opto NCDT 1300-50 displacement
sensor is used to cross-check the accuracy of the travelling speed of CNC X-Y table
carrying a quite heavy hydraulic clamp pressure system. These trials give acceptable
results except initial acceleration (jerk) before reaching to uniform velocity.

An Olympus STM 6 measuring microscope is used for measuring weld seam widths. The
STM6 microscope offers high performance three axis measurements of parts, with sub-
micron precision (0.1 µm). Inbuilt LED illuminator is used for reflected coaxial
illumination during measurements. Fig. 4 shows the Olympus STM 6 measuring
microscope that is used for measuring weld seam widths.

6
 RTMST

Fig. 4. Olympus STM 6 measuring microscope used for measuring weld seam widths

Experiments are conducted under different parametric conditions and the weld width
responses obtained from the developed FE model are compared with the experimentally
measured results. Fig. 5 shows the comparison of results that are obtained by laser
transmission contour welding of polycarbonates via experiments and FE modeling. The
red region at the modeled weld interface denotes fusion zone where temperature exceeds
the glass transition temperature (150 °C) of polycarbonate. A reasonable consistency is
observed between the predicted and measured weld widths. The marginal difference
between the predicted and measured widths is because of the assumptions made in the
simulations, such as uniform material properties, constant optical properties of the
materials, perfect Gaussian distribution of the laser beam intensity at weld interface, and
the perfect contact between mating parts.

Welding parameters:

Power: 5 W
Welding speed: 20 mm/s
Beam diameter: 1 mm
Clamp pressure: 1 MPa Cross-sectional view

Carbon black content of

absorbing polymer: 0.1
%wt.

Weld width Top view (weld interface)

Top view
Predicted: 0.94 mm
Measured:0.8562 mm

7
 RTMST

Welding parameters:

Power: 12 W
Welding speed: 15 mm/s
Beam diameter: 1 mm
Clamp pressure: 1 MPa Cross-sectional view

Carbon black content of

absorbing polymer: 0.1
%wt.

Weld width
Top view Top view (weld interface)
Predicted: 1.14 mm
Measured: 1.0246 mm
WW = weld width Experimentally Predicted half weld width
WW/2 = half weld width measured weld width (half of the model using
symmetry)

Fig. 5. Comparison of weld width results predicted by the FE model and measured
experimentally at different operational conditions

5. Results and Discussion

The developed model is capable of predicting the three-dimensional transient temperature

field as outputs. The objectives of this work is to predict the transient temperature field as
well as the weld dimensions as a function of process parameters viz. laser power, welding
speed, beam diameter and welding pressure. Fig. 6 shows the transient temperature
distribution during laser transmission welding of polycarbonates (absorbing polymer
containing 0.2 % wt. carbon black) for laser power = 5 W, welding speed = 20 mm/s, and
beam diameter of 1.5 mm. It is seen from this figure that the temperature reaches to a
maximum of 314 °C during heating, and, the heat generated at irradiation zone is
gradually transferred to surrounding materials by effect of thermal conduction.
Polycarbonate has a glass transition temperature of about 145-150 °C, so it softens
gradually above this point. Glass transition temperature is the reversible transition
in amorphous materials from a solid state into a molten or rubber-like state. However, the
welding condition to get an acceptable weld is that the temperature at the weld interface
should be above the melting temperature of the polycarbonate (250 °C) and the
temperature of weld zone must not exceed the decomposition temperature of the
polycarbonate (500 °C). The region in the contours plot temperature above 250 °C
represents the weld zone. It can be noticed that the maximum temperature occurs within
the absorbing material. This phenomenon indicates volumetric absorption of the laser
energy within the absorbing material. It is further observed that the isotherms are dense in
the leading side of the advancing laser source than those in the trailing side.

8
 RTMST

Fig. 6. 3D modeled contours obtained for body temperatures

Fig. 7 shows the sectional view of temperature distribution at X-Z plane (for absorbing
polycarbonate part containing 0.1 % wt. carbon black). Given the melting temperature of
polycarbonate, the region in the contours plot temperature above 250 °C, shows the depth
of penetration in both the polycarbonate parts (DT and DA) from weld interface. The
weld half width (WW/2) can also be estimated from Fig. 7.

Fig. 7. Temperature distribution at X-Z plane (WW = weld width, DA = depth of penetration in
absorbing part, DT = depth of penetration in transparent part)

6. Conclusion
In this research work, a three-dimensional FE model is developed to simulate the laser
transmission welding of thermoplastics with a moving laser beam. Polycarbonate is
considered as work material for developing and validating the model. The commercial
finite element code ANSYS® multiphysics is employed to obtain the numerical results.
Heating and cooling phase of welding is successfully modeled. It is observed from the
sensitivity analysis that the effect of heat conduction is predominant on temperature
distribution. For a very short welding time, the energy loss through radiative and
convective heat losses from the small heated zone is negligible. The results obtained for
9
 RTMST
different welding conditions are compared to the experimental results and reasonable
agreement is observed which validates the accuracy of the model. The developed model
can further be used for parametric study and process optimization.

REFERENCES

[1] R. Prabhakaran, M. Kontopoulou, G. Zak, P.J. Bates, B.K. Baylis. Contour laser –
Laser-transmission welding of glass reinforced nylon 6, Journal of Thermoplastic
Composite Materials, Vol. 19, 2006, pp. 427-439.
[2] H. Potente, J. Korte, Becker, F. Laser transmission welding of thermoplastics: analysis
of heating phase. Journal of Reinforced Plastics and Composites, Vo. 18, Issue 10, 1999,
pp. 914-920.
[3] C. Bonten, C. Tüchert. Welding of plastics-Introduction into heating by radiation,
Journal of Reinforced Plastics and Composites, Vol.21, Issue 8, 2002, pp.699-710.
[4] F.G. Bachmann, U.A. Russek. Laser welding of polymers using high power diode
lasers, in Proceedings of the ‘Photonics West Conference’, San Jose, CA, USA; Proc.
SPIE, Vol.4637B, 2002.
[5] B. Acherjee, A.S. Kuar, S. Mitra, D. Misra. Laser transmission welding: A novel
technique in plastic joining, in: Welding: Process, Quality and Applications, Ed. Richard
J. Klein, Nova Science Publishers, Inc.: New York, 2011, pp. 366-387.
[6] F. Becker, H. Potente, A step towards understanding the heating phase of laser
transmission welding in polymers, Polymer Engineering and Science, Vo. 42, Issue 2,
2002, pp. 365-374.
[7] M. Ilie, J.C. Kneip, S. Mattei, A. Nichici. Effects of laser beam scattering on through-
transmission welding of polymers, in Proceedings of the ‘24th International Congress on
Applications of Lasers & Electro-Optics (ICALEO- 2005)’, Miami, Florida, USA., 2005.
[8] L.S. Mayboudi, A.M. Birk, G. Zak, P.J., Bates. A 2-D thermal model for laser
transmission welding of thermoplastics, in Proceedings of the ‘24th International
Congress on Applications of Lasers & Electro-Optics (ICALEO-2005)’, Miami, Florida,
USA., 2005.
[9] J.D. Van de Ven, A.G. Erdman. Laser transmission welding of thermoplastics—Part I:
Temperature and pressure modeling, Journal of Manufacturing Science and Engineering,
Vol. 129, 2007, pp. 849-858.
[10] L.S. Mayboudi, A.M. Birk, G. Zak, P.J., Bates. Laser transmission welding of a lap-
joint: Thermal imaging observations and three–dimensional finite element modeling,
Journal of Heat Transfer, Vol. 129, 2007, 1177-1186.
[11] J.M.P. Coelho, M.A. Abreu, F.C. Rodrigues. Modeling the spot shape influence on
high-speed transmission lap welding of thermoplastic films, Optics and Laser in
Engineering, Vol. 46, 2008, pp. 55-61.
[12] B. Acherjee, A.S. Kuar, S. Mitra, D. Misra. Finite element simulation of laser
transmission welding of dissimilar materials between polyvinylidene fluoride and
titanium, International Journal of Engineering, Science and Technology, Vol. 2, Issue 4,
2010, 176-186.
[13] B. Acherjee, A.S. Kuar, S. Mitra, D. Misra. Effect of carbon black on temperature
field and weld profile during laser transmission welding of polymers: A FEM study,
Optics and Laser Technology, Vol. 44, Issue 3, 2012, pp. 514-521.
[14] M. Mitchell. Design and microfabrication of a molded polycarbonate continuous
flow polymerase chain reaction device, Master thesis, Louisiana State University, 2000

10
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18th and 19th April, 2013, NITTTR, Kolkata

Artificial neural network based monitoring of weld quality

in pulsed metal inert gas welding using wavelet packets of
current signal
Kamal PAL1 , Surjya K. PAL2
1
Associate Professor, School of Mechanical Engineering, KIIT University, Bhubaneswar
Email: kpal5676@gmail.com
2
Associate Professor, Mechanical Engineering Department, IIT Kharagpur, Kharagpur
Email: skpal@mech.iitkgp.ernet.in

Abstract: Pulsed metal inert gas welding (P-MIGW) is often used to improve weld quality using an
advanced spray metal transfer with less heat input to the weld. It is a nonlinear process with various
uncertainties like contamination. Thus, the necessity for online welding process monitoring has
increased in modern manufacturing environment. Though, statistical regression methods are widely
used to develop mathematical models in arc welding, it found to be inadequate to predict some specific
weld quality features like joint strength. So, various intelligent tools like soft computing techniques are
developed with better predictability. Different sensor based features may also be useful to improve the
process monitoring. This paper addresses different statistical features as well as time-frequency
wavelet packet coefficients of sensors’ signals like voltage, current for the prediction of weld quality
features in P-MIGW. The voltage and welding current signals have been acquired during welding
experiments as per response surface design of experiment technique. Initially, a comparison has been
made between regression models and back propagation neural network (BPNN) models for weld
quality characteristics as a function of process parameters like peak voltage, pulse frequency, welding
speed, torch angle etc. The mathematical regression models were found to be inadequate. Therefore,
BPNN model has been retrained using different sensor based features to improve in weld quality
prediction capability. Wavelet packets of welding current were found to be an important indicator of
butt joint strength.

Keywords: Weld Joint strength, hardness variation, Pulse parameters, Arc power, Wavelet packet coefficient

11
 RTMST
1. Introduction

Pulsed metal inert gas welding (P-MIGW) is often used in today’s manufacturing industries because of uniform

metal transfer without any spatter at realistic cost than constant voltage arc welding processes. It is a superior

spray metal transfer with reduced heat input to the weld. It is a variation of constant voltage arc welding

which engage cycling of the arc voltage from a peak value to a root value at a particular frequency [1]. Current

pulsing is applied to obtain uniform finer grain in fusion zone (FZ) which may produce high joint strength

because of uniform hardness in FZ and heat affected zone (HAZ) interface. The weld quality is the best for one

droplet of molten metal transfer at the end of wire electrode per one pulse in pulsed gas metal arc welding (P-

GMAW) [2]. The weld quality characteristics primarily depend on bead geometry [3] and weld microstructure

[4], which in turn indicate the joint strength and welded plate distortions. These bead features are affected by

the metal transfer modes and arc stability in GMAW. Therefore, it is necessary to establish a relation between

weld quality features with the process parameters. The conventional mathematical regression tools like

response surface methodology (RSM) focus mainly on the mean of the performance characteristic, whereas

the Taguchi method considers the variance to develop the model in arc welding. Various numerical and

analytical thermal models like finite element method also found to be useful to build up the weld distortion

model. However, arc welding processes, being highly dynamic and nonlinear with various types of

uncertainties like environmental conditions, it is really difficult to design reliable welds. This multivariate

environment indicates the necessity for an intelligent system which can characterize and monitor the process

in a better way.

The soft computing tools like artificial neural network (ANN) provide an alternative approach for predictive

learning and modeling of weld quality without any mathematical model. These evolutionary algorithms

consider the uncertainty features of the welding processes, which may not be expressed by mathematical

equations. Thus, they are better compared to conventional statistical and analytical techniques. These tools

12
 RTMST
can handle a large number of data to generate the model and optimize it with a short time span. These tools

are also adaptable for incremental learning, enabling the models to be improved incrementally as new data

become available. In recent years, weld quality can be monitored in real time with the application of adaptive

sensor integrated control systems [5-6].

In the present work, two major weld quality features namely, joint strength (σt) and hardness variation in the

FZ-HAZ interface (∆Hw-h) have been modeled using ANN as well as RSM to improve process monitoring for

achieve a desired weld bead in P-MIGW. The models are compared and validated by using the data from

validation experiments. The welding torch angle (αt), welding speed (S), and wire feed rate (F) along with three

major pulse parameters, such as peak voltage (Vp), pulse frequency (fp) and pulse on-time (tp) were considered

for model development. Various time domain statistical features as well as time-frequency wavelet features of

current and voltage signals have also been used in ANN models for further improvement of the predictions.

2. Experimental Procedure

In this work, a constant voltage P-MIGW machine (FRONIOUS make) was used. The experiments were carried

out on 6 mm mild steel plates using same type 1.2 mm diameter electrode wire. Pure argon was used as the

shielding gas. Initially, fifty three butt welding experiments have been carried out using half fractional central

composite response surface methodology. The butt weld joints made of one pair of plates were tack welded at

the two ends before final welding. A Hall-effect current sensor (LEM, model LT 500S) and voltage sensor were

used to acquire the actual process behavior. The signals were acquired using one A/D cards (National

Instruments, USB-6210) to an Intel Pentium-4 PCs using LabVIEW 7.1 data acquisition interface at a sampling

frequency of 40 kHz. The schematic representation of the experimental set-up is shown in Fig.1.

13
 RTMST

Fig. 1. Schematic diagram of experimental set-up

The butt welded joint specimens have been prepared to measure hardness variation and tensile test according

to ASTM standard. The tensile characteristics have been measured by universal tensile testing machine

(INSTRON, 8862) with attached software (Instron Wave Matrix). The weld bead was cut crosswise by an

abrasive cutter (Buehler Delta Abrasimet, 10-2155). The weld fusion zone and HAZ have been identified under

microscopes after cutting, grinding and polishing (using Buehler Ecomet® 3000), and etching with 2% nital

solution. The micro-hardness was measured with the help of micro-hardness tester (UHL VMHT, VMH 001) at

500 gf in various weld zones. The butt weld joint samples have been prepared for the tensile test according to

ASTM (E 8) standard. The hardness variation at the FZ-HAW interface was considered with the ultimate joint

tensile strength as these parameters primarily indicate the butt weld joint quality.

The acquired voltage and current signals were post-processed in the time domain to obtain their root mean

square (RMS) values Vrms and Irms, respectively. Various time domain statistical values like mean, RMS, standard

deviation and kurtosis of sensors’ signals were found to be correlated with different weld quality

14
 RTMST
characteristics in butt welding. The RMS value of arc power was found to be strongly correlated with joint

tensile strength (7-8). So, RMS value of current and voltage were further used to develop the process

modeling. These sensors’ signals have also been further analyzed in time-frequency wavelet mode to measure

the higher level wavelet packet coefficients which were further used to improve the prediction capability of

the models.

3. Time-frequency Wavelet Analysis

The wavelet transform is localized in both time and frequency instead of only frequency as in case of Fourier

transform. Wavelet analysis is an efficient windowing technique with variable sized windows. It allows the use

of long time intervals for low frequency information and shorter regions for high frequency information. Thus,

it can analyze the localized high frequency area of a larger signal.

The wavelet is a mathematical function used to divide a continuous mode time function (or signal) into

different scale components. The wavelet transform represents a signal by wavelets, which contains translated

and scaled wave functions of a finite length wave, called as mother wavelet. Generally, a wavelet is expressed

mathematically as:

1 t b 
Wa ,b (t )  W 
a  a 
(1)

where, b indicates location parameter and a stands for scaling parameter [9].

The wavelet transfer function may be expressed as:

1 t b
C ( a, b)   f (t ). W .dt
t a  a 
(2)

15
 RTMST
For every (a, b) interval, it shows a wavelet transform coefficient which represents the degree of similarity of

the scaled wavelet to the function at the position of t = (b/a). A critical sampling indicates the resolution of

discrete wavelet transform (DWT) in both time and frequency. It shows minimum number of wavelet packet

coefficients sampled from continuous wavelet transform (CWT), which ensure about all the information

present in the original signal. In critical sampling a  2 j and b  k .2 j , where integer j and k represent the

discrete translation and discrete dilations, respectively. Then C (a, b) can be expressed as C ( j, k ) .

C ( j, k )   f (t ).2 j / 2.W (2 j t  k ).dt   f (t ).W j ,k (t )

t t

(3)

In the wavelet packet analysis, the original signal W (0,0) has been decomposed into two separate frequency

band parts as low frequency and high frequency component by passing the signal through a high-pass and low-

pass filter, respectively in the first level of decomposition. This decomposition process will be continued up to

the number of level of decomposition under consideration. In this work, original signal W (0,0) have been

decomposed into three levels, which is represented by tree structure as shown in Fig. 2. 1-D wavelet packet

decomposition has been carried out using MATLAB 7.8. The signal features has found to be insignificant at

higher levels beyond this third level of decomposition. The RMS values of the third level wavelet packet

function W ( j , k ) j[0,3];k[0,7] are termed as C ( j, k ) j[0,3];k[0,7 ] , which have been processed related to the

frequency band under consideration.

In this work, these wavelet coefficients corresponding to third level of decomposition were assumed as

C (3,0) for frequency band of 0Hz to 2500Hz, C (3,1) for 2500 to 5000 and so on up to C (3,7) for 17500Hz to

20000Hz, where sampling rate was 40 kHz. The Daubechies family of wavelets ( dbM ) has been considered,

where M indicates the order of mother wavelet function.

16
 RTMST
Sensitivity analysis has been carried out to identify significant wavelet coefficients of voltage and current

signals. The most significant wavelet packet coefficients for voltage [V(3,0), V(3,4), V(3,6) and current signals

[(I(3,1), I(3,3), I(3,4)] of the welding experiments (53 experiments for the model development and 7 more

experiments for validation of models) along with RMS value of voltage and current signals and weld quality

features are shown in Table 1.

Table 1: Process parameters with corresponding sensors’ outputs and joint features
Process parameters RMS V & I Significant wavelet packet coefficients Joint features
Sl I(3,1) I(3,3) I(3,4) V(3,0 V(3,4) V(3,6) ∆Hw-h
F S αt Vp fp tp Vr Ir σt
no
1 7 5.8 -15 30.4 144 4.6 24.3 216 0.054 0.029 0.015 0.93 0.013 0.028 85 353.1
. 2 5 0 57 4 6
2 8 5.8 -15 35.6 144 4.6 26.1 259 0.051 0.026 0.014 0.99 0.007 0.019 23 513.7
. 5 7 5 43 9 1
3 8 5.8 -15 30.4 144 3.4 20.3 239 0.092 0.054 0.025 2.04 0.020 0.042 83 340.2
. 8 4 9 09 0 1
4 8 5.8 -15 35.6 105 3.4 21.9 240 0.065 0.034 0.020 1.79 0.019 0.042 94 280.8
. 0 2 9 56 8 5
5 7 5.8 -15 35.6 144 3.4 24.1 231 0.052 0.027 0.016 1.28 0.008 0.020 68 483.2
. 9 1 7 67 8 6
6 7 5.8 -15 35.6 105 4.6 23.9 231 0.047 0.026 0.016 1.38 0.008 0.021 30 507.6
. 7 3 5 66 0 5
7 8 5.8 -15 30.4 105 4.6 20.6 236 0.080 0.043 0.022 1.79 0.021 0.043 93 336.3
. 1 6 9 36 3 9
8 7 5.8 -15 30.4 105 3.4 19.4 217 0.081 0.045 0.024 1.70 0.018 0.040 87 383.7
. 9 7 5 76 7 0
9 8 8.8 -15 30.4 144 4.6 22.9 236 0.072 0.039 0.019 1.54 0.020 0.043 83 346.7
. 0 0 3 09 4 5
10 8 8.8 -15 35.6 144 3.4 23.7 246 0.058 0.029 0.017 1.45 0.016 0.034 94 363.9
. 1 1 8 47 3 7
11 7 8.8 -15 35.6 105 3.4 21.9 221 0.052 0.028 0.019 1.51 0.017 0.037 85 368.3
. 0 3 1 64 1 1
12 8 8.8 -15 30.4 105 3.4 18.2 237 0.096 0.055 0.028 1.89 0.019 0.039 98 130.7
. 8 5 1 36 4 2
13 7 8.8 -15 30.4 105 4.6 20.6 213 0.080 0.045 0.023 1.70 0.017 0.037 73 301.8
. 2 0 8 46 7 4
14 8 8.8 -15 35.6 105 4.6 23.5 253 0.052 0.028 0.017 1.47 0.015 0.033 28 491.7
. 8 3 8 98 3 4
15 7 8.8 -15 30.4 144 3.4 20.6 216 0.076 0.043 0.022 1.61 0.017 0.037 85 223.6
. 2 9 6 08 3 1
16 7 8.8 -15 35.6 144 4.6 26.4 238 0.050 0.027 0.013 0.96 0.007 0.021 54 437.4
. 5 9 9 80 4 8
17 7 5.8 15 35.6 105 3.4 23.3 220 0.070 0.039 0.021 1.90 0.022 0.046 81 373.7
. 5 6 2 27 2 8
18 8 5.8 15 35.6 105 4.6 24.3 244 0.059 0.032 0.018 1.65 0.021 0.045 70 416.2
. 7 7 3 34 1 6
19 7 5.8 15 30.4 144 3.4 21.4 219 0.083 0.049 0.023 1.94 0.020 0.041 51 446.3
. 5 3 5 27 9 0
20 7 5.8 15 30.4 105 4.6 22.1 215 0.076 0.045 0.021 1.87 0.021 0.043 74 315.1
. 3 3 9 49 9 3
21 8 5.8 15 35.6 144 3.4 25 243 0.076 0.041 0.020 1.93 0.023 0.048 31 506.7
. 8 1 3 91 0 5
22 7 5.8 15 35.6 144 4.6 27 238 0.050 0.026 0.013 0.90 0.007 0.021 60 384.9
. 8 6 3 24 0 2
23 8 5.8 15 30.4 105 3.4 19.9 240 0.099 0.061 0.028 2.14 0.020 0.041 117 137.8
. 9 9 4 55 2 5
24 8 5.8 15 30.4 144 4.6 23.8 237 0.081 0.043 0.020 1.95 0.024 0.049 61 347.7
. 6 1 7 82 7 3
25 8 8.8 15 30.4 144 3.4 20.3 239 0.098 0.057 0.027 2.09 0.020 0.042 65 401.7
. 2 1 0 89 7 5
26 7 8.8 15 35.6 144 3.4 25.4 232 0.054 0.027 0.016 1.21 0.007 0.020 43 399.1
. 3 3 5 27 3 6
27 7 8.8 15 30.4 105 3.4 18.7 219 0.091 0.054 0.026 1.81 0.018 0.039 85 352.5
. 3 1 9 45 6 0
28 7 8.8 15 30.4 144 4.6 24.5 218 0.061 0.033 0.016 2.67 0.024 0.050 90 288
. 7 0 7 71 6 8
29 7 8.8 15 35.6 105 4.6 23.8 236 0.049 0.026 0.016 1.29 0.006 0.020 77 380
. 2 7 4 65 8 9

17
 RTMST
30 8 8.8 15 35.6 144 4.6 25.8 264 0.052 0.027 0.014 1.02 0.007 0.020 21 520
. 7 7 4 51 1 7
31 8 8.8 15 35.6 105 3.4 22.6 240 0.068 0.037 0.021 1.71 0.020 0.043 84 354.4
. 4 0 6 72 2 4
32 8 8.8 15 30.4 105 4.6 20.6 241 0.078 0.044 0.022 1.71 0.020 0.043 83 244.7
. 5 4 8 09 4 2
33 8 7.7 -35 33 124 4 22.5 242 0.058 0.032 0.018 1.32 0.017 0.038 36 410.3
3
0.085 7
0.048 1
0.023 01
1.89 1
0.022 2
0.046
34 8 7.7 35 33 124 4 23.4 242 68 400.3
8
0.062 0
0.033 2
0.018 81
1.54 0
0.020 2
0.044
35 8 7.7 0 33 124 4 22.7 228 65 404
0
0.052 8
0.028 5
0.017 85
1.26 8
0.014 8
0.033
36 8 9.9 0 33 124 4 22.4 229 95 368.9
5
0.056 7
0.030 0
0.017 14
1.36 8
0.019 5
0.042
37 8 7.7 0 33 124 4 22.6 227 59 421.4
2
0.075 4
0.045 6
0.020 52
2.18 8
0.023 2
0.047
38 8 4.6 0 33 124 4 23.9 231 80 225
7
0.062 0
0.033 2
0.018 52
1.53 1
0.020 0
0.044
39 8 7.7 0 33 124 4 22.9 225 62 399.9
2
0.063 4
0.033 6
0.018 10
1.57 3
0.021 3
0.046
40 8 7.7 0 33 124 4 22.9 226 65 411.3
2
0.071 7
0.040 4
0.019 20
1.93 1
0.022 1
0.047
41 8 7.7 0 33 124 4 23.8 226 71 403.2
7
0.054 9
0.028 6
0.013 49
0.89 3
0.009 8
0.023
42 8 7.7 0 33 172 4 25.8 233 54 452.9
6
0.089 1
0.054 5
0.026 91
2.31 9
0.036 4
0.053
43 8 7.7 0 27 124 4 20.4 226 131 47.59
2
0.050 6
0.025 3
0.017 36
1.49 2
0.009 2
0.021
44 8 7.7 0 39 124 4 27.8 252 82 373.5
4
0.068 5
0.037 7
0.022 78
1.73 6
0.018 3
0.038
45 8 7.7 0 33 80 4 19.5 226 105 130.4
1
0.095 9
0.055 7
0.027 44
0.22 2
0.021 7
0.044
46 8 7.7 0 33 124 2.6 19.9 227 95 187
4
0.048 2
0.026 5
0.012 42
0.03 2
0.006 0
0.019
47 8 7.7 0 33 124 5.4 24.4 234 25 499
5
0.069 0
0.037 9
0.019 91
0.24 3
0.021 5
0.047
48 8 7.7 0 33 124 4 23.7 227 66 397.8
6 6
0.055 7
0.031 2
0.016 08
0.14 8
0.015 1
0.034
49 7.7 0 33 124 4 23.8 204 87 272.1
. 9 5 2 07 6 7
50 9 7.7 0 33 124 4 23.2 251 0.073 0.039 0.020 0.23 0.021 0.045 79 337.2
. 0 4 6 19 3 6
51 8 7.7 0 33 124 4 22.4 231 0.058 0.032 0.018 0.21 0.018 0.041 68 419.1
4
0.067 7
0.035 8
0.018 20
0.22 6
0.021 7
0.043
52 8 7.7 0 33 124 4 23 227 73 409.3
1
0.065 2
0.034 3
0.018 61
0.22 6
0.020 4
0.042
53 8 7.7 0 33 124 4 23.1 225 67 416.3
7 8 7 33 7 7
Validation experiments Validation experiments
54 7 7.2 0 33 84 4.2 19.2 221 0.083 0.050 0.018 0.19 0.022 0.041 113 136
. 6 4 82 4 4
55 8 7.6 -25 28 121 3.8 18.8 236 0.099 0.051 0.021 0.21 0.028 0.049 95 283.2
. 1 5 91 9 3
56 8 8 -15 36 136 4.5 25.2 239 0.052 0.027 0.012 0.23 0.021 0.030 53 451.3
. 3 8 81 9 5
57 7 6.8 -25 32 92 4 22.4 217 0.064 0.031 0.017 0.20 0.023 0.041 71 323
. 1 6 08 7 6
58 8 7.6 -15 34 124 3.2 23.7 220 0.061 0.034 0.013 0.22 0.031 0.047 54 387.1
7 0.053 4
0.042 7
0.011 25
0.14 3
0.017 3
0.021
59 7.2 25 38 112 3.6 24.3 219 47 417.2
. 4 2 63 0 8
60 8 7.6 15 35 100 4.6 25.5 241 0.056 0.030 0.011 0.19 0.023 0.030 71 332.1
. 4 4 42 4 9

Fig. 2. Wavelet packet tree for third level decomposition of acquired sensors’ signals

4. Development of RSM and ANN model

Response surface is a functional mapping of multiple process parameters to a single output feature. In the

present research, second order polynomial response surface models are developed using first 53 sets of data

to correlate six input process parameters: S, F, αt, Vp fp, and tp with the each weld quality feature. The MINITAB

(release 13.31, Minitab Inc. 2002) software was used for the model development and further statistical

analysis to check the adequacy of the model. The hardness variation from WZ to HAZ (∆Hw-h) and joint ultimate

18
 RTMST
tensile strength (σt) were modeled as joint quality features as shown in equation 4 and 5. The adequacy of the

models was tested with 95% confidence level using the analysis of variance (ANOVA) technique. When the

calculated value of ‘t’ corresponding to a coefficient exceeds the standard tabulated value, the coefficient may

be considered as significant. The significant regression coefficients were recalculated to develop the final

model. Finally, the adequacy of the models was tested with 95% confidence level using the analysis of variance

(ANOVA) technique.

H w h  101.5  0.3 t  16.7 S  113.3F  31.9VP  215t p  1.5 S 2  3.3F 2  0.9VP 2 

7t 2p  0.2 t F  0.7 t t p  2.2SF  0.2SVP  0.1Sf p  St p  1.8 FVP  0.3Ff p  12.1Ft p 
3.8VP t p  0.4 f p t p
(4)

 t  697.2  0.4 t  135.2S  595 F  177.1VP  2.8 f p  545.9t p  5 S 2  17.9 F 2  3.9VP 2 

0.7 t S  0.7 t F  0.2 tVP  2.7 t t p  15.3SF  1.9SVP  0.2 Sf p  10 St p  8.4 FVP 
2.1Ff p  56.8 Ft p  0.1VP f p  8.1VP t p  1.6 f p t p
(5)

The multi-layered feed-forward network with back propagation gradient descent learning algorithm is widely

used in welding process modeling. The feed forward network constitutes an input layer, an output layer and

any number of hidden layers. Each layer is comprised of a variable number of nodes as neurons. In the present

work, a code for multi-neuron, multi-layered back-propagation (BPNN) model is used for mapping the P-MIGW

process parameters to weld quality characteristics.

In this work single hidden layer with log-sigmoidal transfer function (f) has been used in all the layers

considering the non-linearity of the process behavior. The nodes of each layer are interconnected to the

19
 RTMST
preceding and subsequent layer nodes with synaptic weights. In the forward pass, the weighted inputs (I) are

summed up to determine the output (O) of the neuron as per equation 6. The weight of each preceding nodes

(wji) multiplied by corresponding inputs (yi) whose summed up value indicate the weighted input of jth neuron.

O  f ( I )  f ( w ji y i ) (6)

Training algorithms change the inter-neuron weights in such a way as to reduce a desired error function (E)

relating the target values (Ti) to the actual output (Oi) values (equation 7).

1 N (7)
E  (Ti  Oi )2
N 1

Each synaptic weight is modified from Wold to Wnew according to an error correction rule (equation 3) based on

the gradient descent technique to minimize the mean square error (MSE) between actual pth output (Opk) and

desired pth output (Tpk) to the total number of training pattern (N) during the backward pass as per equation 8.

The learning rate (η) is to be adjusted to reduce MSE. The momentum coefficient (α) has also been used to

maintain the stability of η with adequate learning according to delta rule.

E
Wnew  Wold  
Wi

(8)

2
1 N P (9)
MSE   k k
 Tp  O p
2N k  1 p  1 

20
 RTMST
The performance of the model depends on the network parameters like number of neurons in hidden layer (j),

no of hidden layer (h), learning rate of the synaptic weights (η) and momentum coefficient (α). There is no

significant improvement of MSE in testing with the consideration of more number of hidden layers in the

present case. Thus, single hidden layer is considered, whereas j, η and α were varied from 1 to 30, 0.1 to 0.9

and 0.1 to 0.9, respectively. Single hidden layer was found to be sufficient to reduce MSE in training as well as

testing with less number of iterations (i.e. less computational time). Several trials were made to finally obtain

the optimal architecture, which can provide the minimum MSE in testing. The optimum architecture was

found by varying the number of neurons in the hidden layer along with the variation of η and α. This

evaluation was carried out by the determination of MSE in testing (MSE_TEST) based on the absolute

prediction error value of the weld quality characteristics.

The RMS value of arc current and voltage were used with process parameters in BPNN models to improve the

prediction capability as RMS arc power was found to be correlated with joint mechanical properties. Various

combinations of sensors’ signal features were used along with six process parameters to investigate the weld

quality prediction as shown in Table 2. The process parameters were only been considered in Strategy #1. The

RMS value of welding signals (current and voltage) was used in Strategy #2. Wavelet packet coefficients of

current and voltage signal with process parameters were used in Strategy #3 and #4, respectively. Finally, the

best wavelet packet coefficients of the voltage and current signals were considered in Strategy #5.The number

of neurons in the input layer of the developed BPNN models depends on the strategy under consideration.

21
 RTMST
Table 2. Weld quality prediction strategies

Sl no Features used in the BPNN model Number of input nodes

1 Six process parameters 6
2 Six process parameters with RMS welding 8

3 Six parameters with RMS values of wavelet 14

packet coefficients of current signal

4 Six parameters with RMS values of wavelet 14

packet coefficients of voltage signal

5 Six parameters with RMS values of 16

significant wavelet packet coefficients of

current and voltage signals

5. Result and Discussions

The ANOVA results are shown in Table 3 and Table 4 for two weld quality feature models. The acceptance of

these models mainly depends on P-value, F-value and R2 value. P-value indicates the probability of

significance. It is calculated based on the F-ratio. The P-value is then compared with the assumed confidence

level (in this case 95 %). If the P-value is less than 0.05, then the model may be accepted. The F-value of the

model has to be higher than the tabulated F- value at 95 % confidence level at respective DOF of both the

regression model and residual error. Thus, these criteria were found to be fulfilled, i.e. the regression models

were acceptable. However, the R2 value i.e. the coefficient of correlation indicates the closeness of the

predicted output values with the actual experimental responses. Its’ value lies in between 0 to 1. Higher R2

value indicates better model. The R2 values for hardness variation ( H w  h ) and tensile strength (  t ) were

found to be less than 0.85. Thus, the mean absolute prediction errors for seven validation experiments were

found to be more than 35% (Table 6). Therefore, it may be concluded that the second order regression

22
 RTMST
equations are not so adequate to represent the relationship between process parameters with respect to

these two weld quality characteristics.

Table 3. ANOVA table for hardness variation ( H w  h ) model

Source DOF SS MS F- value P-value

Equation 4 26 24465.8 940.99 5.89 0.000

Residual error 26 4153.5 159.75

Total 52

F0.05,26,26 = 1.94

Table 4. ANOVA table for joint tensile strength (  t ) model

Source DOF SS MS F- value P-value

Equation 5 26 450447 17324.9 4.10 0.000

Residual error 26 109801 4223.1

Total 52 560248

F0.05,26,26 = 1.94

The simulation result of various strategies using BPNN code was compared according to prediction error (Table

5). The prediction capability of the various neural network structures were compared with mean square error

in testing (MSE_TEST). The 6-30-2 architecture with η and α as 0.4, and 0.4, respectively proved the best

data fitting for the prediction of joint tensile strength and hardness variation without considering sensor

features. This optimum architecture provided the minimum MSE in training (MSE_TRAIN) and testing

(MSE_TEST) as 0.007358 and 0.009317, respectively.

23
 RTMST

Table 5. Prediction performance of BPNN models for weld quality features using various strategies

Process Strategy Best Optimum Optimum MSE_TRAIN MSE_TEST

outputs no network  

σt, ∆Hw-h 1 6-30-2 0.4 0.4 0.007358 0.009317

2 8-29-2 0.5 0.4 0.009733 0.007742

3 14-4-2 0.5 0.1 0.002864 0.005781

4 14-4-2 0.6 0.5 0.003861 0.007341

5 16-12-2 0.5 0.5 0.004917 0.006355

The absolute prediction error was found to be 14.69% and 27.18% for joint tensile strength and hardness

variation, respectively. However, joint strength prediction error was considerably reduced to 7.44%

considering best sensor based strategy (strategy #3) as shown in Table 6. The 14-4-2 network structure with

learning rate and momentum coefficient of 0.5 and 0.1, respectively, was found as the best network

parameters for the prediction of joint strength and hardness variation at the interface of weld fusion zone to

unaffected base plate. The minimum MSE in training (MSE_TRAIN) and testing (MSE_TEST) using this best

network were 0.002864 and 0.005781, respectively. Therefore, it may be concluded that wavelet feature of

current signal is an important indicator to monitor joint strength.

24
 RTMST

Table 6. Comparison of prediction performance of various modeling techniques

Weld quality Mean absolute prediction error (%) using

features
RSM BPNN Sensor based BPNN (Strategy #3)

∆Hw-h 36.90 14.69 14.86

σt 37.45 27.18 7.44

The actual joint tensile strength of the seven validation experiments with predicted values using RSM, BPNN

and best sensor based BPNN (Strategy #3) are shown in Fig 3. The predicted value of individual tests using

sensor based BPNN technique were almost close to experimental values. Therefore, sensor based features are

highly useful to predict weld quality features in a better way.

550

500

450
Joint tensile strength, MPa

400

350

300

250
Experimental
200
RSM
BPNN
150
Sensor based BPNN
100
1 2 3 4 5 6 7
Validation experiment number

Fig. 3. Prediction of joint tensile strength using various modeling techniques

25
 RTMST

6. Conclusion

The arc sensors’ signals are strongly correlated with joint mechanical properties in P-MIGW process. The

mathematical regression model is inadequate to predict weld joint strength. However, it is significantly

improved using wavelet packet coefficients of current signal in back-propagation ANN technique. The average

prediction error was reduced from 37% using RSM model to 21% using BPNN models which again further

improved to 11% using wavelet features of current signal with six process parameters which is highly

competitive with earlier works. Thus, BPNN models are better than response surface regression models in

terms of prediction capability. The current signal wavelet values are highly useful to improve joint strength

monitoring capability.

References

[1]. I.E. French, and M.R. Bosworth, “A comparison of pulsed and conventional welding with basic flux cored
and metal cored welding wires”, Welding Journal, Vol.74, Issue 6, 1995, pp. 197s–205s.
[2]. K. Stanzel, “Pulsed GMAW cuts cycle time by 600 percent”, Welding Design and Fabrication, 2001, pp.
85–87.
[3]. E. Karadeniz, , U. Ozsarac, and C. Yildiz, “The effect of process parameters on penetration in gas metal arc
welding processes”, Material Design, Vol. 28, Issue 2, 2007, pp. 649–656.
[4]. G. Powell, and G. Herfurth, “Charpy V-notch properties and microstructures of narrow gap ferritic welds
of a quenched and tempered steel plate”, Metallur. Mater. Trans. A, Vol. 29, Issue 11, 1998, pp. 2775-2784.
[5]. Y.M. Zhang, E. Liguo, and B.L. Walcott, “Robust control of pulsed gas metal arc welding”, J. Dyn. Syst.
,Meas. Contr., ASME, Vol. 124, 2000, pp. 1–9.
[6]. H.C.D. Miranda, A. Scotti and V.A. Ferraresi, “Identification and control of metal transfer in pulsed
GMAW using optical sensor”, Sci. Tech. Weld. Join., Vol. 12, Issue 3, 2007, pp. 249-257.
[7]. K. Pal, S. Bhattacharya, and S.K. Pal, “Investigation on arc sound and metal transfer modes for on-line
monitoring in pulsed gas metal arc welding”, Journal of Materials Processing Technology, Vol.210, Issue 10,
2010, pp. 1397-1410.
[8]. K. Pal and S.K. Pal, “Monitoring of weld penetration using arc acoustics in pulsed MIG welding”, Materials
and Manufacturing Processes, Vol. 26, 2011, pp. 684-693.
[9]. K.P. Soman and K.I. Ramachandran, “Insight into wavelets from theory to practice”, Prentice-hall Ind. Pvt.
Ltd., 2nd edition., 2006.
___________________

26
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

EXPERIMENTAL INVESTIGATION OF TIG WELDING OF AUSTENITIC

STAINLESS STEEL WITH CHANGE IN COMPOSITION OF FILLER
MATERIAL

Nabendu GHOSH1, Pradip Kumar PAL2, Goutam NANDI3 and Ramesh RUDRAPATI4
1
PhD Student, Mechanical Engineering Department, Jadavpur University, Kolkata-700032, India
Email: nabendu2003_ghosh@yahoo.co.in
2
Professor, Mechanical Engineering Department, Jadavpur University, Kolkata-700032, India
Email: pradippal54@yahoo.com
3
Professor, Mechanical Engineering Department, Jadavpur University, Kolkata-700032, India
Email: gnandi57@yahoo.com
4
PhD Student, Mechanical Engineering Department, Jadavpur University, Kolkata-700032, India
Email: rameshrudrapati@gmail.com

Abstract: Weld quality mainly depends on features of bead geometry, mechanical-metallurgical characteristics
of the weld as well as on various aspects of weld chemistry and these features are expected to be greatly
influenced by various input parameters like current, voltage, electrode stick-out, gas flow rate, edge
preparation, position of welding, welding speed and many more. Therefore, preparation of a satisfactory good
quality weld seems to be a challenging job. Among other factors, type / composition of filler metal may
influence the quality of weldment. From the literature survey it is found that work has not been done to a
good extent in so far as effect of filler metal on the mechanical –metallurgical characteristics of weld is
concerned. In the present work some of the input parameters have been considered to identify the effect of
these parameters on quality of weld, with variation in composition of filler material. These parameters are
current and gas flow rate in TIG welding of austenitic stainless steel of 3mm thickness. Responses considered
are: ultimate strength and breaking strength of the butt welded joints . The influence of change in filler
material composition on weld quality has been emphasized in the present work .The objective of the present
work is to investigate the influence of three different filler materials on quality of weldment in TIG welding of
austenitic stainless steel. Not only change of filler material has been taken into consideration but also effects
of current and gas flow rate are also considered as mentioned earlier, Visual inspection and X-ray radiographic
tests have been conducted after making butt welded joints with three different filler materials. Next tensile
test specimens have been prepared from the welded joints and these have been tested on tensile testing

27
 RTMST
machine. The results of tensile tests are reported and presented graphically in this paper. Finally, based on the
results of the present investigation; some useful conclusions have been drawn. Process optimization has also
been done by Grey – based Taguchi method.

Keywords: TIG; Filler Material; Tensile Testing; X-ray Radiographic; Microstructure; Taguchi – Grey
Relation Analysis

1. Introduction

Tungsten Inert Gas Shielded Arc welding (or TIG) is a special type of arc welding, which is used under certain
conditions to obtain better quality of weldment, by using an inert gas or a mixture of inert gases. Weld quality
mainly depends on features of bead geometry, mechanical-metallurgical characteristics of the weld as well as
on various aspects of weld chemistry and these features are expected to be greatly influenced by various input
parameters like current, voltage, electrode stick-out, gas flow rate, edge preparation, position of welding,
welding speed and many more. Moreover, the cumulative effect of various input parameters determines the
extent of joint strength that should meet the functional aspects of the weld in practical field of application.
Therefore, preparation of a satisfactory good quality weld seems to be a challenging job. Among other factors,
the selection of filler metal may influence the quality of weldment. The welding investigators have always
been in search for better quality of weldment. Juang and Tarng [1] adopted a modified Taguchi method to
analyze the effect of TIG welding process parameters on the weld pool geometry of S 304 stainless steel
weldment. Starling et al. [2] developed a model and studied the effect of process parameters on the bead
shape in a narrow gap – GTAW process with magnetic arc oscillation. Durgutlu Ahmet [3] examined
microstructure, penetration and mechanical properties for tungsten inert gas welding of 316L austenitic
stainless steel. Seshank et al. [4] used ANN and Taguchi method to analyze the effect of pulsed current GTAW
process parameters on bead geometry of aluminium bead-on plate weldment. Several applications of neural
networks to the welding process have been reported in the literature [5-7]. From the literature survey, it is
found that work has not been done to a good extent in so far as effect of filler metal on the mechanical –
metallurgical characteristics of weld is concerned. Extensive research is needed in this context. A data of
knowledge – base may thus be generated which may be utilized by the practicing engineers and technicians to
produce good quality weld more precisely, reliably and predicatively.

2. Experimental Procedure, Inspection And Testing

In the present work, butt welding of AISI 304 stainless steels has been done based on L9 Taguchi’s Orthogonal
Array Design of experiment. Each of the two pieces welded, under varied conditions has the size; 100 mm x 50
mm x 3mm. In this set of experiments, gas flow rate, welding current and welding filler rod have been
considered as process variables. Welding process parameters and their levels is shown in Table 1. Butt welded
joints being done under varied input parameters, visual inspection and X-ray radiographic test of all welded
specimens has been made. After visual inspections and X-ray radiographic test, tensile test specimens have
been prepared from the welded joints, by cutting/machining. During cutting/machining of the tensile test
specimens, small cut- outs have been taken. These cut pieces have then been ground, polished and etched for
studying microstructures under Versamet metallurgical microscope.

28
 RTMST
Table 1. Welding process parameters and their levels for L9 Taguchi’s Orthogonal Array Design of experiment

Process Parameter Unit Level 1 Level 2 Level 3

Current (C) A 80 90 100
Flow Rate (Fw) l/min 10 11 12
Filler Rod (F) ER 304L ER 316 L ER 310

3. Results of Visual Inspection and Discussion

The Table 2 shows the defects found while visually inspecting the welded samples obtained under
varied conditions of current, gas flow rate and with different filler rod materials.

Table 2. The observed results of visual inspection

Exp. Identification Observation

No.
1 S-2/Fw-10/C-80/ F-304L Uneven deposition, Undercut
2 S-17/Fw-10/C-90/ F-316L No defect
3 S-18/Fw-10/C-100/ F-310 Undercut, uneven deposition
4 S-19/Fw-11/C-80/ F-316L Blow holes, undercut
5 S-20/Fw-11/C-90/ F-310 Undercut, small spatter
6 S-7/Fw-11/C-100/ F-304L No defect
7 S-21/Fw-12/C-80/ F-310 No defects
8 S-10/Fw-12/C-90/ F-304L Undercut, blow holes
9 S-22/Fw-12/C-100/ F-316L Undercut

It is found from this table that for certain welding conditions; no defect has been observed. These are for the
Exp. Nos. 2, 6, and 7. Blow holes and undercut, small spatter, uneven penetrations are the types of defects
which are found in other samples. The combined effect of the levels of the gas flow rate, current and the type
of filler material has possibly enchanced / reduced the possibility of the reasons which may lead to lack of
penetration, undercut, uneven deposition found in several welded samples. Porosity and blow holes may have
possibly resulted from gas getting entrapped in solidifying metal. Larger arc, faster level of travel speed, too
low or too high arc currents, incorrect welding technique (stringer beads or weaved beads etc.), damp filler
rod, unclean job surface are the general reasons behind porosity and blow holes. Undercuts found under
certain conditions in the present study may have been caused by improper joint geometry for some of the
samples and unfavourable combination of current and gas flow rate. Spatter found is caused possibly due to
damp filler rod or arc blow or bubble of gas being entrapped in the molten globule of metal expanding with
great violence and projecting small drops of metal outside the arc seam. Lack of penetration may be related to
improper setting of the current and gas flow rate combination, wrongly held electrode / filler rod, faster arc
travel speed, too small root gap, insufficient heat input.

29
 RTMST
4. Some Typical Results of X-ray Radiographic Test

X- Ray radiographic tests have been conducted. The detailed results (not given here) indicate the following. For
Samples corresponding to Exp. No. 2, 6 and 7 no significant defects are found. For Samples corresponding to
Exp. No. 3, 5 and 9 lack of penetration is observed. Other are found to be more or less fine. Among nine X-ray
radiographic films, two representative ones are shown in Fig. 1 and Fig. 2 respectively.

Fig. 1. Radiography film of Exp.No.5 Fig. 2. Radiography film of Exp. No.7 –No Defect
Lack of Penetration (C= 80A, Fw= 12 l/min, F= ER 310)
(C=90Amps, Fw=11 l/min and F= ER 310)

5. Results of Tensile Test and Discussion

The tensile test specimens prepared corresponding to L9 Taguchi Orthogonal Array experiments have been
tested for tensile strengths and the results obtained are given in Table 3.

Table 3. Tensile Test Result for L9 Taguchi’s Orthogonal Array Design of experiment

Exp. Ultimate Breaking Ultimate Breaking

Load Load Tensile Tensile
(KN) (KN) Strength Strength
(Mpa) (Mpa)
1 32.5 31.2 570.2 547.4
2 35 34.2 614 600
3 29.5 28.6 517.5 501.8
4 30 29 526.3 508.8
5 25 24 438.6 421.1
6 35 34.2 614 600
7 36 35 631.6 614
8 34 33.6 596.5 587.7
9 28.5 26 500 456.1

30
 RTMST
The Table 3 indicates that for many of the welded samples test results are satisfactory. The best result
is obtained for the sample corresponding to Exp. No. 7(Corresponding to current 80 A, flow rate 12
l/min and ER 310 filler material). For this sample, ultimate tensile strength = 631.6MPa and breaking
strength= 614 MPa. The worst result in tensile test has been obtained for the sample corresponding to
Exp. No. 5 (corresponding to current 90 Amp, gas flow rate 11 l/min and filler material ER 310). This
sample breaks at 421.1 MPa and its ultimate tensile strength is 438.6 MPa. However, the best
combination of parameter setting plus filler material has been determined later by Taguchi-cum Grey
relation analysis. Load Vs. elongation curves indicate that in the most of the samples ductile behavior
is observed excepting in exp. no.05. Load Vs. elongation curves – Exp. No. 05 and Exp. No.07 are shown
in Figs. 3 and 4 respectively. The result of visual inspection, X-ray radiographic test and tensile testing
are found to be more or less consistent.

Fig. 3. Load vs elongation curve –Exp.No.05 Fig. 4. Load vs elongation curve –Exp.No.07
(C=90Amps, Fw=11 l/min and F= ER310) (C=80Amps, Fw=12 l/min and F= ER310)

6. Results of Microstructural Studies

Study of microstructures has been made for all the welded samples and the photographs are taken in weld
and HAZ regions, for each of the samples. Only some typical micro structural views in weld region are shown
in Figs. 5 and 6 respectively.

Fig. 5. Sample corresponding to Exp. No.05 - Fig.6. Sample corresponding to Exp. No.07 –
Weld Weld

31
 RTMST

Microstructure of weld metal corresponding to Exp. No.07 (Fig. 6) Shows dispersed chromium carbide
throughout the structure. In the weld metal microstructure of Exp. No.05 (Fig. 5) Chromium carbide is found
in the austenite ferrite matrix in a more or less equi-axed manner. The results of microstructural studies are in
consistent with the results of visual inspection and X-ray radiography. More detailed analysis of
microstructures both in weld HAZ regions are to be reported separately.

7. Optimization By Using Grey-based Taguchi Method For L9 Taguchi Orthogonal

Array Design Of Experiment

Responses (ultimate tensile strength and breaking tensile strength) of L9 Taguchi orthogonal array design of
experiment should be high, so normalizing the experimental data is to be done according to Taguchi’s larger -
the- better (LB) criterion by using the equation (1)

( ) ( )
( ) = ( ) ( ) (1)

where xi (k) is the value after the grey relational generation, min yi (k) is the smallest value of yi (k) for the kth
response, and max yi (k) is the largest value of yi (k) for the kth response. Experimental data are collected as
per L9 Taguchi orthogonal array design matrix discussed earlier. Normalization of experimental data is shown
in Table 4.

Table 4. Normalization of experimental data

Exp. No. Ultimate Tensile Strength (Mpa) Breaking Tensile Strength(Mpa)

Ideal
Sequence 1.0000 1.0000
1 0.6810 0.6540
2. 0.9090 0.9270
3 0.4090 0.4180
4 0.4540 0.4540
5 0.0000 0.0000
6 0.9090 0.9270
7 1.0000 1.0000
8 0.8180 0.8630
9 0.3180 0.1810

Grey relation coefficients are shown in Table5. Grey relation coefficient ξi(k) has been calculated by using
equation 2.

32
 RTMST

(k) = (2)
( )

Where Δ =∥ - ( ) ∥difference of the absolute value x0(k) and xi(k); θ is the distinguishing coefficient 0 ≤
θ ≤ 1; Δmin = the smallest value of Δ0i ; and Δmax = largest value of Δ0i. andΔ0,i(k) is the absolute value of
difference between X0 (k) and Xi (k ).

Usual steps have been followed while applying Grey relation analysis.

Table5. The grey relation coefficient ξi(k)

Exp. No. Ultimate Tensile Strength (Mpa) Breaking Tensile Strength(Mpa)

1 0.6105 0.5910
2. 0.8460 0.8726
3 0.4582 0.4621
4 0.4780 0.4780
5 0.3333 0.3333
6 0.8460 0.8460
7 1.0000 1.0000
8 0.7849 0.6361
9 0.3790 0.3576

After averaging the grey relation coefficients, the grey relational grade can be computed as:

= ∑ ( ) (3)

The overall grey relation grade is shown in Table 6.

Table 6. Calculation of overall grey relation grade

Exp. No. Grey Relation Grade

1 0.6007
2 0.8593
3 0.4601
4 0.4780
5 0.3333
6 0.8593
7 1.0000
8 0.7590

33
 RTMST
9 0.4010

Fig. 7 Mean effects plot for overall grey relation grade

With the help of mean effect plots for S/N ratio (Fig. 7), optimum parametric combination has been
determined. The optimal factor setting becomes Fw-12 l/min, C-80 A, F- ER 304L. This has been validated by
confirmatory test.

8. Conclusions

The following conclusions are drawn in the context of TIG welding of AISI 304 stainless steel done under varied
conditions with several filler materials.

 Results of visual inspection indicate that undercut and blow holes have been found in few samples;
uneven deposition, and excessive penetration have also been found in some samples. In many cases
no defects have been detected as well.
 Considering all the filler materials used in the study and all the levels of current and gas flow rate –
the best joint strength is obtained for the sample corresponding to Exp. No. 7[C=80Amp, Fw=12
l/min and Filler material ER 310].
 The worst result in tensile test has been obtained for the sample corresponding to Exp. No. 5
(corresponding to current 90 Amp, gas flow rate 11 l/min and filler material ER 310).
 From Mean effect plots for S/N ratio optimum parametric combination has been determined. The
optimal factor setting becomes Fw-12 l/min, C-80 A, F- 304L, validated by confirmatory test.
 The result of visual inspection, X-ray radiographic test and tensile testing are found to be more or
less consistent.

9. References

[1]. Juang, S.C. and Tarng,, Y.S., Process parameters seletion for optimizing the weld pool geometry in the
tungsten inert gas welding of stainless steel, J Mater Process Technol., 2002;122:33-37.
[2]. CMD starling, PV Marques and PJ Modenesi, Stastistical modeling of narrow –gap GTA welding with
magnetic arc oscillation, J Mater Process Technol, 1995;51:37-49.

34
 RTMST
[3]. Durgutlu Ahmet, Experimental investigation of the effect of hydrogen in argon as a shielding gas on TIG
welding of austenitic stainless steel, Materials and Design , 2004; 25: 19-23.
[4]. Seshank, K., Rao, SRK., Singh, Y., Rao, KP., Prediction of bead geometry in pulsed current gas tungsten arc
welding of aluminium using artificial neural networks, in Proceedings of international conference on
information and knowledge engineering, IKE 03, June 23–26, 2003, Las Vegas [NV], USA 149–53.
[5]. Senthil Kumar, T., Balasubramanian, V. and Sanavullah, M.Y, Influences of pulsed current tungsten inert
gas welding on the tensile properties of AA 6061 aluminium alloy,Materials and Design 2007; 28:2080-2092.
[6]. Juang, S.C., Trang, Y.S. and. Lii. H.R , A compression between the back-propagation and counter-
propagation networks in the modeling of the TIG welding process, J Mater Process Technol., 1998; 75: 54-62.
[7]. Tarng YS., Juang S.C., Chang C.H. , The use of grey-based taguchi methods to determine submerged arc
welding process parameters in hardfacing, .J Mater Process Technol., 2002; 128:1-6.

35
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Additive Manufacturing of Casting Die with Ideal Shapes of

Gating System through Rapid Tooling

Divyansh PATEL1, Abhilasha SINGH2, Puneet TANDON3

1
Mechanical Engineering Discipline, PDPM IIITDM,
Dumna Airport Road, Jabalpur - 482005 (M.P.) India,
Mob.: +91-8989763171, Email: divyanshpatel.008@gmail.com
2
Department of Mechanical Engineering, SGSITS, Indore,
23, Park Road, Indore- 452003 (M.P.) India,
Mob.: +91-9425809542, Email: abhilasha0325@gmail.com
3
Professor, Mechanical Engineering Discipline,
PDPM IIITDM, Dumna Airport Road, Jabalpur – 482005
(M.P.) India, Mob.: +91-9425324240,
Email: ptandon@iiitdmj.ac.in

Abstract:
Producing tooling directly from CAD models to manufacture industrial products is regarded as an
important method of reducing the cost and time to market the new products. A comprehensive review
of examples of rapid tooling indicates a major shift in tooling practice. This paper describes the role of
Additive Manufacturing (AM) technology in increasing the speed of tooling development. The
proposed work presents a new design and manufacturing paradigm based on direct Rapid Tooling (RT)
concept to increase casting yield by improving the geometry and design of gating elements which
plays very important role in the production of sound casting. Ideal shapes of gating system (i.e. runner,
riser and sprue) are obtained by using the additive manufacturing technique with help of solid based
rapid prototyping machine. RT shortens the process by eliminating several steps of the cycle of
traditional casting process and leads to design optimization. Casting yield and production time for unit
product are the output parameters which are optimized by techniques of RT.

36
 RTMST
Keywords: Casting yield; gating system elements; FDM: Fused Deposition Modeling

1. Introduction:
1.1 Casting dies (molds):

Mold is a sort of container which when poured with molten metal produces a casting of the shape of
the mold after solidification.
Gating system: The term refers to all passageways through which the molten metal passes to enter the
mold or die cavity. Since the way in which liquid metal enters the mold has a decided influence upon
the quality and soundness of a casting, the different passages for the molten metal are carefully
designed and produced.

1.2 Rapid prototyping and tooling:

RP was introduced as a designers’ visualization tool, the cost involved does not justify its usage
restricted to visualization alone, therefore RP has been extended to manufacturing of tools also, efforts
are more focused on rapid development of tooling for casting, molding and sheet-metal forming
industries. Rapid tooling is the use of additive manufacturing methods to produce either prototype
parts, or tools to transform materials into functional parts which can be used as a tool. In
Manufacturing, RP means to create physical prototypes directly from digital data i.e. CAD designs.
Surfaces of the CAD model are tessellated and STL file is exported to the software which allows
selection of part deposition orientation than tessellated model is sliced before firing to rapid
prototyping machine.

This experimental study shows an engineering application of rapid prototyping (RP) technology in
mold cavity or die manufacturing. Rapid prototyping (RP) is a new kind of manufacturing technology
integrated with multiple studies. In the past two decades, the development of CAD improved the
product modeling and relevant digitization to a great extent. As a digital manufacturing technology
directly driven by CAD, RP is also developing greatly and is widely used in product development and
manufacturing day by day. Rapid prototyping (RP) initially focused on polymers. These were later
replaced / supplemented by ceramics, metals and composites. Composites are used in RP not only to
make desired product but also to facilitate the process. For example in Fused Deposition Modeling
(FDM), a blend of various polymers may be used, in which different polymers play the role of
tackifier, plasticizer, surfactant, etc.

1.3 Literature review:

A rapid wax injection tool of a gearbox shift fork was designed, simulated, and manufactured using
rapid prototyping and rapid tooling technology to save time and cost of producing wax models used for
the investment casting process. The criteria of such comparison were based upon parameters such as
time, cost, and other related characteristics, which resulted in saving of 50% in time and 60% in cost
[1]. The gating and riser design plays an important role in the quality and cost of a metal casting. Due
to the lack of existing theoretical procedures to follow, the design process is normally carried out on a
trial-and-error basis [2]. The capture of design and fabrication knowledge in the form of episodes
provides easy access to vital information. The Rapid Foundry Tooling System has been shown to
provide unique and useful capabilities in support of the design of sand castings and molds for these

37
 RTMST
castings [3]. For design of shapes of gating system i.e. runner, riser, sprue etc. and principles of gating
system a text book [4] is referred. Sand casting of LM6 was experimentally explained by Simulation
driven feeder size, shape and neck optimization. Design of casting and dimensions of gate, sprue and
riser are scaled and modification in shape of gating elements were done in present work. Greg
Krikorian [6] presented an experimental comparison of rapid prototyping and tooling options. J.C.
Ferreira [7] concluded that with new processing technique integrating Rapid Tooling time is reduced
from the design of cores to core-boxes production. D. King et al. [8] wrote about selective laser
sintering (SLS) rapid prototyping process has been adapted to produce metal-based prototypes that can
be used in rapid tooling applications. Clinton L. Atwood [10] studied scope o investment casting.
Junyong Park et al. invented an efficiently delivering molten metal to regions of a casting where
shrinkage is expected [11]. C. W. Lee et al. [12] says about the advantages derived include significant
amounts of cost and time savings, relatively accurate final castings (average dimensional error 1.5%
for the direct RP method and 0.9% for the indirect RP method) with reasonable surface quality and the
complete elimination of hard tooling required in conventional investment casting (IC) process.

1.4 Objective:

The objective of the work is to optimize the casting yield by eliminating several steps of conventional
casting process and modifying the designs of gating system elements. Present work describes an
experimental study of manufacturing of a Mold cavity/die using the Rapid prototyping machine by
eliminating several steps of a Casting process. Using Rapid Tooling mold cavities or patterns of
intricate geometries can easily be produced which is very difficult and time taking task for
conventional casting process.

A casting process involves following steps:

1. First to start with a pattern (a replica of the Object to be made with some modifications like
shrinkage allowance, draft allowance, machining allowance etc.)
2. Calculations of gating elements (volume, shape, allowances etc.)
3. Sand preparation.
4. Creating hollow cavities with the help of patterns.
5. Pouring the liquid material into the cavity and allow to solidify.
6. Cleaning and finishing.
7. Heat treatment to relieve the residual stresses.
8. Inspection and removal of defects.

Rapid tooling eliminates 1st, 2nd and 3rd Steps and generates directly the mold cavities by additive
manufacturing layer by layer material addition process. In addition, with some small modification in
shapes of the gating system elements by improved CAD model of the die, casting yield has been
increased by 20 %.

2. Design of Gating System:

Gating system is composed of

1. Pouring cup and basins
2. Sprue (receiving the poured melt)
3. Runner (a channel through which the melt is supplied to the gates)
4. Gates (through which the molten metal enters the mold cavity)
5. Risers ( feeds the melt to fill the shrinkage during solidification)
38
 RTMST

The metal solidified in the sprue, runner, gates and risers should be minimum because these elements
are removed from the final casting; the gating system should provide for the maximum yield. Figure 1
illustrates the gating system of step casting pattern designed in CATIA. Actual model was
dimensioned to analyze the mechanical properties of step casted by sand casting with aluminum.
Dimensions of gating system are calculated on the basis of step casting volume and shrinkage
allowances for aluminum, solidification rates etc. are tabulated in table 1. All the dimensions like
Volume of riser, Area of gate, area of bottom and top of the sprue are calculated according to the
volume of part to be manufactured and based on principles of gating, Principles of risering and
principles of mold making.

Fig. 1. Original pattern of step casting

Table 1: Data regarding original step casting [5]

S.N. Parameter Specification

1. Gating Ratio 1:2:2
2. Filling time 28.3 min
3. Casting Volume 750000 mm3
4. Casting Weight 6 kg
5. Casting Surface Area 6000 mm2
6. Gating system volume 550000 mm3
7. Riser diameter:50 mm, height: 50 mm , Volume: 108131 mm3
8. Casting Yield- 58.3%

3. Design and CAD model improvements

3.1 Design improvements in Gating system:

3.1.1 Sprue: Sprue is the passage placed vertically in moulding box through which molten material
reaches to the runner. When material is allowed to flow from the sprue following problems may occur:
 Liquid material passes down the sprue it loses its pressure head but gains velocity.
 High turbulence occurs.
 A vortex tends to form in the sprue.
 The negative pressure effect may be generated.
39
 RTMST

To overcome these problems only the shape of sprue is responsible. An ideal shape for sprue is
‘tapered parabola’ but in practice generally ‘Inverted cone’ type sprues are used as shown in figure 2.
So using modeling software sprue with tapered parabolic shape with exact dimensions was created.
The sprue is modified by fitting a parabolic curve in CATIA by fixing the upper and lower cross
sectional areas same as in original model.

3.1.2 Runner: A runner is a placed horizontally which connects sprue with gates or casting as shown
in figure 2 (top view). A good runner and gate should have:
 Abrupt changes in section and sharp corners which create turbulence and gas entrapment.
 A suitable relationship must exist between the cross sectional area of several ingates, gates and
runner and runner and sprue.
 Ingates should be properly located to feed the casting.

Fig. 2. Dimensions of pattern; front view (Left), top view (Right)

In general practice runner shape is created rectangular cross section, Ideal shape of a runner is
trapezoidal. In Rapid Tooling there is no need of dealing with patterns and a moulding box with
trapezoidal cross section can be manufactured easily so that the rectangular shaped runner was
replaced by trapezoidal cross section.

3.1.3 Riser: A riser or feeder head is generally manufactured on the mould cavity to compensate
Liquid and solidification shrinkage so while designing a riser a designer should emphasis on reducing
the surface area so that solidification time for riser can be reduced. In general practice Cylindrical riser
as shown in figure 1 and 2 is easy to produce but it decrease casting yield due to the reason that it
requires more amount of material to provide compensation. A sphere has minimum surface area so a
spherical riser is designed while modeling the mold cavity.

3.2 Improvements in CAD model:

The enabling technology for use of rapid prototyping processes is CAD solid modeling. The CAD
solid model captures the geometry, which can be easily processed in any RP machine. After design
modifications model for which the die had to be generated is dimensioned after scaling of the original
model to manufacture a prototype for mock experiments. Modified CAD model of pattern and die is
shown in figure 3 and figure 4 respectively.

40
 RTMST

Fig. 3. Modified pattern for generating mold cavity

4. Implementation:

4.1 Preparation of model:

Pattern is converted into die by dividing it into two halves and inverted into negative die by using the
CATIA the model is saved in STL format. Figure 4 shows the inverted die of the pattern model with
improvements in shapes of riser, runner and sprue. Model preparation begins with validation of the
input model to ensure it is a solid; if it is not, it must be repaired. (Models are corrupted either by
designer misunderstandings or inadequate CAD post-processing, as described below.) The valid model
is then scaled and oriented with respect to the build chamber, taking into account build direction.

Fig. 4. Modeling of the mold in CATIA

4.2 Firing of mold cavity in Rapid Prototyping machine FDM Fortus M140:
41
 RTMST

To be more precise, CAD post-processors actually approximate the vendors’ internal CAD geometric
forms (e.g., B-splines) with a simplified mathematical form (triangles), which in turn is expressed in a
specified data format: STL. Unfortunately, this approximating operation, if not done precisely,
sometimes introduces undesirable geometric anomalies, such as holes or overlapping portions in the
boundary surface. After completing the modeling the model is saved in STL format and transferred to
INSIGHT software first allows the operator to make a proper orientation of the part then estimates
time required, volume of material to be defused (ABS M30) and volume of support material (SR 20)
needed. Then software inputs the machine to fire the mold cavity. Figure 5 shows the mold cavity
manufactured by rapid tooling.

Table 2: Volume of materials required and Expected time

Nos. Parameters Part 1 Part 2

1. Volume of material of the part 6.486 inch3 6.313 inch3

2. Volume of support material 0.198 inch3 0.198 inch3

3. Time to complete the part 2 hours 11 minutes 2 hours 8 minutes

 Machine used: FDM 140 Fortus

 Material of part: ABS M 30 and Support material: SR 20

Fig. 5. Prototype of mold cavity part1 (left) and Part 2 (right) manufactured by FDM Fortus 140

4.3 Experiment:

The material chosen for the experiment was wax because of its low melting temperature. Waxes are a
class of chemical that are plastic near ambient temperatures. Characteristically, it melts above 45°C to
give a low viscosity liquid. Wax is insoluble in water but soluble in organic and nonpolar solvents.
Wax is first molten in a pot by heating and poured into the assembled mold cavity from the top
opening of sprue. After filling the mold cavity it was left for solidification until wax is completely
solidified. It takes 15 – 20 minutes to complete the solidification in the mold cavity because the

42
 RTMST
prototype is made by a polymer and thermal conductivity of ABS M30 is very low. After solidifying
the mold is disassembled and shaped wax is removed carefully from the cavity. Figure 5 shows the
disassembled die after solidifying the wax inside. Result is drawn after removing gating system
elements sprue, riser and gate from the step (target product), and weighed all the parts separately.

Fig. 5. Wax casting on mold cavity

5. Result and Discussions:

As a result of using Rapid Product Development Technologies manufacturing of moulding box with
Ideal shapes but with same areas of ingate, Sprue top and bottom, riser-casting interface, the volume of
gating elements found reduced as compared with original one. Figure 4 is showing the manufactured
mold cavity. As we know,
Weight of casting
Casting Yield = x 100
( Weight of casting + Weight of gating system)
Table 3: Weights of gating system and casting

S. No. Elements Weight (grams)

1 Step casting 33.586
2 Runner 0.890
3 Riser 3.111
4 Sprue 4.670
5 Total 43.261

.
Casting Yield = x 100 % = 79.4 %
( . . )
43
 RTMST
6. Conclusions:
With new processing technique integrating Rapid Tooling time is reduced from the design of molding
boxes production. It takes a few hours to fabricate a casting die with its CAD model by RT technology.
When using conventional techniques, it will take some weeks. Thus the manufacturing period of new
casting tools is greatly reduced. RT adds of another advantage of precise modeling by applying reverse
engineering (RE) and simulation increases the efficiency of the process. Based upon the study we can
say that efficiency of casting found to be increased by 15 to 20% using Rapid Tooling. Manufacturing
of molding box using Rapid Tooling is good advancement in the field of production engineering.
Demerit of this technology is only that it is not suitable for mass manufacturing, but it can increase the
capability of mass manufacturing.
1. Sudden or right angle changes in direction are reduced from the gating system.
2. Reduction in turbulence, mold shift or mismatch and gas pick-up.
3. Regulated flow rate of the molten material through the gating system.
4. Increased casting yield by 15-20%.

References:

[1] RAHMATI Sadegh, REZAEI Mohamad Reza, AKBARI Javad T, “Design and Manufacture of a Wax
Injection Tool for Investment Casting Using Rapid Tooling”, TSINGHUA SCIENCE AND
TECHNOLOGY, ISSN 1007-0214, 2009.

[2] Jean Kor, Xiang Chen, and Henry Hu, “Multi-Objective Optimal Gating and Riser Design for Metal-
Casting”, IEEE Multi-conference on Systems and ControlSaint Petersburg, Russia 978-1-4244-4603, 2012.

[3] William H. VerDuin, Dr. Yoh-Han Pao, 1993, “The rapid foundry tooling system : Edge computer aided
design system”, IEEE , CH3306.

[5] Shailendra Sharma, Durgesh Joshi, “Simulation driven feeder size, shape and neck optimization for
LM6 material”, M.Tech Thesis, Shri Govindram Seksaria Institute of Technology and Science, 2011.

[6] Greg Krikorian, “A practical comparision of Rapid prototyping and tooling”, San Leandro, CA 94070,
ISBN#0- 7803-3274-1.

[7] J.C. Ferreira, “Manufacturing core-boxes for foundry with rapid tooling technology”, Journal of
Materials Processing Technology, 1118–1123, 2004.

[8] D. King, T. Tansey, “Alternative materials for rapid tooling”, Journal of Materials Processing
Technology 121, 313–317, 2002

[9] A. Freitas & R. Soares, P. G. Martinho, A. S. Pouzada “Rapid Prototyping and Tooling in the
manufacture of mould elements for large parts”, University College Ghent, 2012.

[10] Friedrich B. Prinz, Panel Chair, Clinton L. Atwood, Richard F. Aubin, Joseph J. Beaman, Robert L.
Brown, Paul S. Fussell, Allan J. Lightman, Emanuel Sachs, Lee E. Weiss, Michael J. Wozny, “JTEC/WTEC
Panel Report on Rapid Prototyping in Europe and Japan”, ISBN 1-883712-44-0, 1997.

[11] Junyong Park, Robert Parolini, Conrad Schaeffer, “Sensor aided direct gating for metal casting”, US
8,387,678 B1,2013.

44
 RTMST
[12] C. W. Lee, C. K. Chua, C. M. Cheah, L. H. Tan, C. Feng, “Rapid investment casting: direct and
indirect approaches via fused deposition modeling”, International Journal of advance manufacturing
technology, 23: 93–101, 2004.

45
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

APPLICATION OF ARTIFICIAL NEURAL NETWORKS

FOR PARAMETRIC MODELLING OF MIG-CO2 WELDING
PROCESS
Akash KUMAR1, Sudipto CHAKI2 and Dipankar BOSE3

1 Mechanical Engineering Department

National Institutes of Technical Teachers Training and Research, Salt Lake City,
Kolkata-700106
aakash.khandekar21@gmail.com

2 Automobile Engineering Department

MCKV Institute of Engineering, Liluah,
Howrah-711204
Sudipto_chaki@yahoo.co.in

3 Mechanical Engineering Department

National Institutes of Technical Teachers’ Training and Research, Salt Lake City,
Kolkata-700106
Dipan88@yahoo.co.in

Abstract - Artificial Neural Networks are particularly useful for online control of process parameters
of welding processes. Several researches have been conducted so far to study its ability to model the
complex relationship between process parameters of Metal Inert Gas (MIG) welding. Present paper
have investigated prediction capability of three back propagation neural Network (BPNN) training
algorithms namely Bayesian Regularization (BR), Levenberg Marquardt (LM), Gradient Descent
Method (GDM) for prediction of MIG-CO2 welding (MIG welding with CO2 as assist gas) process
parameters. Among those techniques BPNN with BR has a unique capability to model small noisy
experimental dataset. But its application in the field of MIG welding has not been observed so far.
Present paper has considered welding current, voltage and welding speed as input parameters and
Depth of Penetration (DP), Material deposition rate (MDR) and Width of HAZ as output parameters
for ANN modelling. Altogether 30 networks have been tested by varying hidden layer neurons from 4
to 13. 3-4-3 network of BPNN with BR produces best prediction capability with mean % absolute
error of DP- 0.859%, MDR-9.89%, and width of HAZ- 1.671%. The result is undoubtedly superior
compared to mean % absolute error obtained by BPNN with GDM (DP-32.78%, MDR-24.32% and

46
 RTMST
width of HAZ-30.75%). Therefore, 3-4-3 network employing BPNN with BR as training algorithm can
be successfully used for process parameter modelling and online implementation of MIG-CO2
welding.

Keywords – MIG-CO2 Welding, Artificial Neural network, Bayesian Regularization, Levenberg

Marquardt, and Gradient Descent Momentum.

1. Introduction
Metal Inert Gas (MIG) welding process is a fabrication method widely used in every industry large or
small. It is an efficient, economical and dependable means of joining metal. This is the only process
which has been tried in the space. MIG-CO2 is variation of MIG Welding in which identical equipment
is used but the inert gas is replaced by CO2. In this process, joining of metal is obtained by heating
them with an arc established between a continuous filler metal, consumable electrode and the work.

However weld properties have an apparently very complex relationship between the operational input
parameters such as welding current, welding voltage, welding speed, wire feeding rate, gas flow rate,
stick out distance, electrode wire diameter, travel speed, gun angle, alternating shielding gas, globular
transfer, work piece plate thickness, gas mixture etc and the corresponding output(s) such as weld
strength, tensile strength, impact strength, weld hardness, elongation, effect on bead geometry etc.
Quality of welding depends on variation of these welding input parameters.

Many researchers in this field have been motivated to develop various mathematical tools for
prediction and optimisation of weld properties. Nowadays artificial neural networks (ANN), a
biological inspired soft computing technique, have emerged as a vital tool for prediction and modelling
of process parameters. ANN has unique ability to approximate the highly complex non-linear
relationship between input-output variable of a given process to any arbitrary degree of accuracy and it
helps in online process monitoring and control of a manufacturing process.

Hakan Ates [1] employed ANN to predict the effect of the gas mixture as input process parameter on
the output parameters such as tensile strength, impact strength, elongation, weld hardness. Seeraj and
Kannan [2] have carried out ANN and multiple regression analysis to model and predict bead
geometry of the MIG welding for input process parameters such as Welding Current, Welding Speed,
Gun Angle and Contact tip. Weld quality also has been predicted through a Neuro-XL predictor and
ANN [3] where welding current, wire diameter and wire feed rate have been used as input parameters.
Recently, Campbell et al [4] have employed ANN for prediction of weld bead geometry by altering
shielding gas as input process parameter.

Significantly no other work of parametric modelling found for prediction of process parameter of
MIG-CO2 with BPNN model of ANN is observed. So In the present work a model of ANN is
employed on an experimental dataset for modelling of depth of penetration, material deposition rate
and width of HAZ as output parameters and welding speed, welding current and voltage has been
taken as input parameters. Back Propagation Neural Network (BPNN) is a feed forward ANN model
widely used for process modelling. In present work three variant of BPNN such as BPNN with
Levenberg Marquardt (LM) algorithm, BPNN with Gradient Descent Momentum (GDM) algorithm
and BPNN with Bayesian Regularization (BR) algorithm are employed for training an experimental
dataset obtained from MIG-CO2 and further compared for prediction capability.

47
 RTMST

2. Experiment
In present investigation, experimental dataset has been obtained from a literature[**]. It has been
found that, experiment has been carried out using CPT-400 MIG welding set up as shown in fig.1 with
CO2 as assist gas. The controllable input process parameters considered are Welding current (A),
Welding voltage (V), and Welding speed (mm/min) and the output are considered Depth of penetration
(DP), Material deposition rate (MDR), and Width of HAZ. The material used for work piece forged
carbon steel of grade “EN-3A” designated by British standard-970 due to its good weld ability. Out of
25 experiment dataset, 20 sets have been considered for ANN training and testing. Dataset is furnished
following in Table 1.
Table.1 Experimental dataset
Input Parameters Output Parameters
Depth Material Width
Curren Welding of deposition of HAZ
Sl no. Voltage
t speed penetration rate zone
(V)
(A) (m/min) (mm) (gm/min) (mm)
A V S DP MDR HAZ
1 140 24 0.165 5.3 1.525 5.524
2 140 25 0.179 5.1 1.213 5.312
3 140 26 0.193 4.9 1.031 5.130
4 140 27 0.206 4.7 0.873 4.972
5 140 28 0.220 4.6 0.796 4.835
6 150 24 0.179 5.2 1.365 5.464
7 150 25 0.193 5 1.886 5.285
8 150 26 0.206 4.9 1.031 5.130
9 150 27 0.220 4.8 0.896 4.995
10 150 28 0.165 6.7 2.907 6.906
11 160 24 0.193 5.2 1.313 5.412
12 160 25 0.206 5 1.162 5.261
13 160 26 0.220 4.9 1.032 5.131
14 160 27 0.165 6.9 3.104 7.103
15 160 28 0.179 6.5 2.690 6.799
16 170 24 0.206 5.1 1.268 5.367
17 170 25 0.220 5 1.143 5.242
18 170 26 0.165 6.9 3.368 7.267
19 170 27 0.179 6.7 2.867 6.966
20 170 28 0.193 6.5 2.709 6.708
Testing Data
21 180 24 0.220 5.1 1.529 5.328
22 180 25 0.165 6.9 3.290 7.399
23 180 26 0.179 6.8 3.004 7.103
24 180 27 0.193 6.5 2.790 6.849
25 180 28 0.206 6.4 2.520 6.629

48
 RTMST

Fig.1 MIG-CO2 Welding setup

3. Artificial Neural Network (ANN)

ANN is an emerging as a vital tool for parametric modelling and due to its inherent prediction
capability. Basically ANN is inspired from the structure of the human brain. They are massively
parallel adaptive networks consisting of simple non-linear computing elements called neurons. Each
neuron receives signals as input which passes through weighted pathway in order to generate a linear
weighted aggregation of the impinging signals. It can be then certainly transformed through an
activation function to generate the output signal of the neurons. The activation function may be binary
threshold, Sigmoidal, Gaussian etc. One main advantage of Artificial Neural Network (ANN) is their
ability to model complex nonlinear relationship between multiple input variables and the required
output. It helps in online process monitoring and control of a manufacturing process.

In this investigation a model has been used which is become famous now days in parametric modelling
with ANN is Back propagation Neural Network (BPNN) with single hidden layer which is also known
as “Universal function Approximator” due to its inherent capability to approximate the underlying
function of a given dataset to an arbitrary degree of accuracy. In the employment of the BPNN
algorithm, iterations of training involves some steps so that a particular training data is fed through the
network in a forward direction and produces results at the output layer. Now error is calculated at the
output nodes based on known target information, and the necessary changes to the weights that lead
into the output layer are determined based upon this error calculation. The changes to the weights that
lead to the preceding network layers are determined as a function of the properties of the neurons to
which they directly connect (weight changes are calculated, layer by layer, as a function of the errors
determined for all subsequent layers, working backward toward the input layer) until all necessary
weight changes are calculated for the entire network. The calculated weight changes are then
implemented throughout the network, the next iteration begins, and the entire procedure is repeated
using the next training pattern.

Researchers often use Traditional Gradient descent BPNN with momentum in modelling of the
experimental studies; however the term momentum does not always seem to speed up training. As it is
more or less application dependent and is found to be too slow for many practical problems, gradually
Gauss Newton and Levenberg Marquardt (LM) algorithm emerged as decade faster than Traditional

49
 RTMST
BPNN as they can achieve second order approximation speed. But for small and noisy dataset,
performance of LM during generalization is not satisfactory. It is observed that Back propagation with
Bayesian Regularization (BR) technology can bridge the gap in performance. Bayesian Regularization
(BR) is more efficient for modelling and prediction of small noisy dataset. But so far application of BR
was not observed in MIG-CO2 and field of research. For the present problem GDM, LM, and BR all
three are employed with Back Propagation (BP) and employed for ANN trainings testing as it deals
with relatively small numbers of experimental dataset.

3. Training and Testing with BPNN algorithms

In BPNN modelling Welding current (A), Welding voltage (V) and Welding speed (S) constitutes the
input nodes of ANN while Depth of penetration (DP), Material deposition rate (MDR), and Width of
HAZ constitutes the output nodes. Therefore both input and output layer contain 3 neurons. The
neurons of the input and output layer constitutes input (X) and output(Y) vector of the system and are
given as follows:
X  [A V S] , Y  [DP MDR HAZ] (1)

Architecture of present ANN is given in Fig. 1. The number of hidden layer is used here as design
parameter of the network. The number of hidden layer neuron in the hidden layer is varied from 4 to
13 at an increment of one neuron. Therefore altogether 30 architectures have been trained and tested
with three BPNN algorithms and best network is selected and used for further analysis. All the input,
hidden and output nodes are interconnected by weights, all the weights are initialised by generating
random numbers before training. For all the cases activation function of the hidden layer and output
layer is sigmoid and linear respectively.

In order to increase accuracy and speed of the network, input and output dataset are normalized
between 0 and 1 before actual application in the network and designated as, normalized input (Xnor)
and output (Ynor) respectively. After forward pass the obtained network output (O) is compared with
desired output value (Ynor) and computed as Mean square errors (MSE), which is minimised over
1 Q
iterations. Mean square errors (MSE) are obtained as follows, E  MSE   (Ynor i  O i ) 2 (2)
Q i1
where Q is total number of training data. If computed MSE satisfies the convergence criterion,
training stops. Otherwise network weights are updated during back propagation and employed in next
iteration.

The process continues till some convergence criterion is met. Batch training is carried out with BPNN
with GDM, LM, and BR. Each network is trained over 300 epochs in this purpose. After completion of
training, 5 sets of unknown test data are fed to the trained network to test its prediction capability.
Accuracy in prediction is measured by comparing predicted output with corresponding known
experimental output and given by test MSE. Along with that, sum of the square error (SSE) and mean
absolute error (MAE) during testing are also calculated for each case. All the training and testing is
done by MATLAB 9.0 in windows 8, 4.3 GHz, 4 GB RAM with 500GB HD.

50
 RTMST
3. Training and Testing parameters
In present investigation, the training is performed with single hidden layer Back propagation (BP)
where input and output layer of the Network has 3 and 3 numbers of nodes respectively. So number of
neurons in the Hidden layer is used here as design parameter of the network. The ANN architecture for
the present problem is shown in Fig. 2.The number of hidden layer neuron in the hidden layer is varied
from 4 to 13. Altogether 10 different architectures are trained and tested with single hidden layer
network, and best network is selected and used for the later part of the program. All the input, hidden
and output nodes are interconnected by weights, all the weights are initialised by generating random
numbers before training. For all the cases activation function of the hidden layer and output layer is
sigmoid and linear respectively.

Welding Speed Material deposition rate

Welding current Depth of Penetration

Voltage Width of HAZ

Fig. 2 ANN architecture for present problem

Here all the data in the input and output are normalized first then fed as input and output to the
Network. MSE is considered as criterion of performance of the network during training and testing.
Before training, random initialization for weight is used. BPNN with GDM, LM, and BR are used here
for batch training. Each network is trained over 300 epochs in this purpose. Then the trained network
is tested for the 5 sets of unknown test data. All the performance indices such as mean square error
(MSE), sum of the square error (SSE) and mean absolute error (MAE) are calculated for each case.
The testing is also done with training or input data to show performance of the network with known
value. All the training and testing is done by MATLAB 9.0 in windows 8, 4.3 GHz, 4 GB RAM with
500GB HD.

4. Results and discussion

On training all dataset using BPNN with BR by varying the number of hidden layer neurons from 4 to
13, it has been observed that training and testing MSE does not varying with hidden layer neurons and
remain fixed at 0.0296 from network 3-5-3 to 3-13-3. A network with lesser number of hidden layer
neurons incurs lesser training time. As the networks having hidden layer neurons from 5 to 13
produces same MSE, the smallest network i.e. 3-5-3 network is considered here as best network and
used for further analysis. Variation of train and test MSE with hidden layer neurons is shown in Fig. 3.

51
 RTMST

Fig.3 Variation of train and test MSE with hidden layer neurons during application of BPNN with BR

Table.2 Training and testing performance using BPNN with LM

Sl. No. ANN Training Training Testing Testing Testing
Networks MSE time MSE SSE MAE
1 3-4-3 0.0003 10.487 0.0021 0.0313 0.0332
2 3-5-3 8.4645 10.641 0.0157 0.2357 0.0779
3 3-6-3 9.3935 10.465 0.0121 0.1812 0.0650
4 3-7-3 2.1412 10.774 0.0027 0.0409 0.0246
5 3-8-3 2.2413 10.946 0.384 5.772 0.3060
6 3-9-3 7.7800 11.365 0.0156 0.234 0.0874
7 3-10-3 6.5117 11.277 0.0213 0.3195 0.0951
8 3-11-3 4.6328 8.2633 0.0310 0.4650 0.1296
9 3-12-3 1.3227 3.446 0.0080 0.1202 0.0647
10 3-13-3 2.96e-24 3.1417 0.0036 0.0533 0.0405

During training and testing using BPNN with LM by varying the number of hidden layer neurons from
4 to 13 it has been observed that 3-4-3 network has best training and predication capability with train
MSE of 0.0003 and test MSE of 0.0021. Variation of train and test MSE with hidden layer neurons is
shown in Table 2 and Fig. 4.

52
 RTMST
Fig.4 Variation of train and test MSE with hidden layer neurons during application of BPNN with LM

Table.3 Training and testing performance using BPNN with GDM

ANN Training Training Testing Testing Testing

Sl. No.
Network MSE time MSE SSE MAE
1 3-4-3 0.0602 27.122 0.156 1.2148 0.0244
2 3-5-3 0.0412 14.925 0.215 7.7232 0.5843
3 3-6-3 0.0726 15.035 0.209 5.2875 0.4842
4 3-7-3 0.147 14.602 0.257 2.1189 0.3329
5 3-8-3 0.135 15.425 0.101 2.9444 0.3332
6 3-9-3 0.0912 14.101 0.182 9.0426 0.6301
7 3-10-3 0.0917 20.049 0.340 3.6857 0.3406
8 3-11-3 0.135 21.701 2.568 4.6251 0.4742
9 3-12-3 0.139 13.888 0.521 7.3408 0.5965
10 3-13-3 0.159 2.269 0.294 6.0017 0.5380

Fig.5 Variation of train and test MSE with hidden layer neurons during application of BPNN with GDM

Results of training and testing of all dataset with Gradient Descent Momentum Back Propagation
Neural Network by varying the number of hidden layer neurons from 4 to 13 are provided in Table 3.
It has been observed that best testing MSE is produced by 3-8-3 network with test MSE 0.101. But
corresponding training MSE (0.135) is not the minimum one. But as prediction capability of the
network is considered as main criteria for selection of the network 3-8-3 network is considered as best
network. Training and testing performance of networks using BPNN with GDM are provided in Fig.5.

Table.4 Selection of best Networks corresponding to least error

Max % Min % Mean %

Training Best
Parameters Absolute Absolute Absolute
Algorithms Network
Error Error Error
BPNN with DP 1.846 0.059 0.859
Bayesian 3–4–3 MDR 33.547 0.305 9.899
regularization HAZ 1.645 0.442 1.671
53
 RTMST
BPNN with DP 20.216 1.089 6.660
Levenberg 3 – 7– 3 MDR 32.344 5.111 15.241
Marquardt HAZ 4.404 0.690 2.088
BPNN with DP 69.747 7.313 32.277
Gradient Descent 3 – 8 – 3 MDR 60.052 5.693 24.320
Momentum HAZ 59.996 15.029 30.750

Fig. 3-5 reveals an interesting fact that, only in Fig. 3 testing/prediction performance of BPNN with
BR is far better than training performance. In Fig. 4 and Fig. 5 testing performance is far inferior to
training performance. It may be due to over fitting of data and tendency of the solution point to stick to
the local minima. Where as better prediction performance of BPNN with BR indicate its ability to
model small noisy dataset. Comparing prediction performance of all 30 networks 3-5-3 network
trained and tested using BPNN with BR is considered as best network.

Prediction capability is further assessed through calculating absolute % errors between the de-
normalised predicted outputs by the best networks and the actual experimental output (or desired
output). Table 4 indicates that, 3-4-3 architecture with Bayesian regularization BPNN for the present
problem can reduce the Minimum Absolute % Error up to 0.059 % and Mean % Absolute Error up to
0.859% and employed for subsequently simulation.

3-4-3 network of BPNN with BR produces best prediction capability with mean % absolute error of
DP- 0.859%, MDR-9.89%, and width of HAZ- 1.671%. The result is undoubtedly superior compared
to mean % absolute error obtained by BPNN with GDM (DP-32.78%, MDR-24.32% and width of
HAZ-30.75%). Moreover, 3-4-3 architecture using BPNN with BR can reduce the minimum absolute
% error up to 0.059 % and mean % absolute Error up to 0.859%. Therefore, 3-4-3 network employing
BPNN with BR as training algorithm can be successfully used for process parameter modelling and
online implementation of MIG-CO2 welding.

6. Conclusion

In the above study with the experimental dataset of MIG-CO2 welding process, parametric modelling
is done with the application of three different BPNN models namely Bayesian Regularization (BR),
Levenberg Marquardt (LM), and Gradient Descent momentum (GDM).

It is concluded that 3-4-3 network architecture trained and tested using BPNN with BR is the best
network for prediction of MIG-CO2 welding process parameters with least mean % absolute error and
can be employed for online process monitoring.

References

[1] H. Ates, Prediction of gas metal arc welding parameters based on artificial neural networks, 2006,
http://dx.doi.org/10.1016/j.matdes.2006.06.013
[2] P. Sreeraj, T. Kannan. Modelling and Prediction of Stainless Steel Clad Bead Geometry Deposited by
GMAW Using Regression and Artificial Neural Network Models, Advances in Mechanical Engineering,
Article ID 237379, June 2012.
[3] P. D. Patel, S.P. Patel.Prediction of weld strength of metal active gas (MAG) welding using Artificial Neural
Network, IJERA, Vol.1, Issue 1, pp. 36-44.

54
 RTMST
[4] S. W. Campbell, A. M. Galloway, N. A. Mcherson. Artificial Neural Network Prediction of Weld Geometry
Performed Using GMAW with Alternating Shielding Gases, Welding Journal, Vol. 91,2012, pp.174-181
[5] S. Pal, S. K. Pal, A. K. Sumantaray. Artificial Neural Network Modeling of weld joint strength prediction of
a pulsed metal inert gas arc signal, Journal of Material Processing Technology, Vol.202, 2008, pp. 464-474.
[6] S. Pal, S. K. Pal, A. K. Sumantaray. Determination of optimal pulse metal inert gas welding parameter with
a neuro GA- technique, Journal of Material Processing Technology, Vol.25, 1998, pp. 606-615.
[7] K.M. Tay, C. Butler. Modeling and Optimizing of a MIG welding Process – A case study using
experimental design and Neural Networks, Quality and Reliability Engineering International, Vol. 13,
1997, pp.61-70.
[8] K. Pal, S. Bhattacharya, S. K. Pal. Optimization of weld deposition efficiency in pulsed MIG welding using
hybrid Neuro-based techniques, International Journal of Computer Integrated Manufacturing, Vol.
24, 2011, pp. 198-210.
[9] M. Wu, G. Zhang, J. Huang, S. Yu, S. Ling. Edge detection for aluminium alloy MIG welding pool based
on pulse coupled neural network, Seventh International Conference on Natural Computation, IEEE , Vol.
978-1-4244-9953-3/11, 2011.
[10] K. Manikya Kanti, P. Srinivasa Rao. Prediction of bead geometry in pulsed GMA welding using back
propagation neural network, Journal of Materials Processing Technology, Vol. 200, Issue 1–3, 2008, pp.
300-305.
[11] D.S. Nagesh, G.L. Datta. Prediction of weld bead geometry and penetration in shielded metal-arc
welding using artificial neural network, Journal of Materials Processing Technology, Vol. 123, Issue
2, 2002, pp. 303-312.

55
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Prediction of Process Parameters for TIG Welding Using Back-

Propagation Neural Networks

Soham DAS
Mechanical Engineering Department
National Institutes of Technical Teachers’ Training and Research, Saltlake city, Kolkata
Sohamengg10@gmail.com
Dipankar BOSE
Mechanical Engineering Department
National Institutes of Technical Teachers’ Training and Research, Saltlake city, Kolkata
Sudipto CHAKI
Department of Automobile Engineering,
MCKV Institute of Engineering,
243, G.T. Road (N), Liluah, Howrah-711 204, India
e-mail: sudipto_chaki@yahoo.co.in

ABSTRACT

It has been already established that, process parameters of Tungsten inert gas (TIG) welding involve
complex mathematical relationship with measured outputs such as weld penetration, welding strength
and different parameters of weld geometry. The present paper is an attempt to model and predict
welding parameters of Tungsten Inert Gas welding on Aluminium plates using back propagation
neural network approaches. In the present work, the input parameters selected are welding speed, wire
speed, current, cleaning(%), gap while the output parameters are front width(FW), back height(BH)
and back width(BW). Three different back-propagation training algorithms such as Baysian
Regularization (BR), Levenberg Marquardt (LM) and Gradient Descent Momentum (GDM) have been
employed for modeling. Though BPNN with BR has an unique capability to model noisy experimental
dataset with certain accuracy, the algorithm has been employed before 72 training data and 36 testing
data for the present problem have been retrieved from a published literature. Altogether 30 networks
has been trained and tested by varying the hidden layer neurons from 6 to 15. Results ensure best
prediction capability of BPNN with BR with least mean absolute percentage prediction error of 4.53%
for FW, 14.93 % for BH and 7.98 % for BW. That prediction performance clearly outperforms BPNN
with GDM which yields high prediction error of 61.60% for FW, 74.20% for BH and 62.19% for BW.

56
 RTMST
The accuracy in prediction performance therefore indicate efficacy of BPNN with BR for prediction of
TIG welding parameters during online process control.

Keywords: - TIG welding, Back Propagation Neural networks, Bayesian Regularization, Levenberg
Marquardt Algorithm, Gradient Descent Momentum method.

1. Introduction

Tungsten inert gas (TIG) welding is an arc welding process which has been opted by the
manufacturing industries from long time ago due to its inherent capability of producing clean weld. It
basically uses a non consumable tungsten electrode to produce the weld in which the weld area is
protected from atmospheric contamination by an inert shielding gas either Argon or Helium and a filler
material is normally used. Due to its high utility, numbers of research work have been carried out
mainly focusing on increase in welding joint strength i.e. to increase the mechanical properties of the
weld pool between the plates. As inputs and outputs of TIG welding process bears a complex
mathematical relationship, conventional regression techniques does not produces accurate
mathematical modeling of the process. An artificial neural network (ANN) is a soft computing based
methodology that mimics the working of human brain. It is an effective tool for complex process
modeling and can be successfully employed for modeling of TIG welding process. Back propagation
neural networks (BPNN), a type of feed forward neural networks, are widely used for modeling of
various engineering processes. Application of ANN in TIG welding was introduced by Juang et. al. [1]
where a comparison has been established between BPNN with gradient descent momentum (GDM)
and counter propagation networks for modeling the TIG welding process. Further in the field of TIG
welding most of the research works have taken place by comparing the process parameters using
different soft computing tools simultaneously. A remarkable contribution is found from the work of
Nagesh et.al.[2] in which the optimisation of TIG welding process variables was carried out using GA
and the bead geometry was predicted by using ANN. Dutta et. al. [3] has employed conventional
regression analysis and neural network based approaches for modeling TIG welding process.
Afterwards Lakshminarayanan et. al. [4] has compared performance of RSM and ANN in predicting
the tensile strength of friction stir welding of Aluminium alloy. Sathiya et. al. [5] employed ANN with
GA for optimisation of laser beam welding process for austenitic stainless steel. In the previous
literatures BPNN with gradient descent momentum (GDM) and Levenberg Marquardt (LM) algorithm
has been used for modeling and prediction of TIG welding process parameters. But BPNN with
bayesian regularization (BR) [6] has an inherent capability of modeling small and noisy dataset with
greater accuracy. So far no work has been observed where BPNN with BR has been used modeling of
TIG welding process.

In the present work, three different BPNN training algorithms such as BPNN with BR, BPNN with
LM and BPNN with GDM are employed for modeling of TIG welding parameters. Altogether 30
different architectures have been trained and their mean percentage absolute errors during
testing/prediction have been discussed in detail. The paper is organised as follows. Section 2 indicates
working principles of three BPNN approaches. Section 3 deals with training and testing of
experimental dataset and section 4 include results and discussion.

57
 RTMST
2. Working of BPNN models

Artificial neural networks are inspired from the structure of the human brain. They are massively
parallel adaptive networks consists of simple nonlinear computing elements called neurons. Each
neuron receives signals as input which passes through a weighted pathway in order to generate a linear
weighted aggregation of the impinging signals. It can be then certainly transformed through an
activation function to generate the output signal of the neuron. Because of this capability it has a great
potential to model the relationships between the input and output variables for non linear, coupled,
multi variable systems. Back-propagation neural networks (BPNN) with single hidden layer neuron is
known as “Universal Function Approximator” [7] due to its inherent capability to approximate the
underlying function of a given data set to an arbitrary degree of accuracy. This network is composed of
many interconnected neurons that are often grouped into input, hidden and output layers.

2.1. Dataset for BPNN modeling

Experimental dataset for present paper has been taken from a published literature [1]. The experiments
have been conducted with welding speed (S), wire speed (WS), current (V), cleaning (%) (C) and
gap(G) as input parameters while front width (FW), back height (BH) and back width (BW) are
considered as the output parameters. Available 108 data have been randomly divided into training and
testing data. The first 72 data set are considered as training data and rest 36 data set are as testing or
simulation data. Each dataset contains 5 input and 3 output data which indicate nodes of input and
output layer. The neurons of the input layer constitutes input vector X of the system and the neurons
of the output layer generates the corresponding output vector Y of the system. Input (X) and output (Y)
vectors are given as follows:

X = [S WS C G V] Y = [FW BH BW ] (1)

Architecture of present ANN is given in Fig. 1. The number of hidden layer is used here as design
parameter of the network. BPNN algorithms works according to the steps mentioned below.

58
 RTMST

Speed(cm/min)
Front width
Wire (mm)

Cleaning
Back height
Gap

Current
Back width
(mm)

Hidden layer

Fig. 1. Architecture of BPNN

2.1. Normalization of data

In order to increase accuracy and speed of the network, input and output dataset are normalized
between 0 and 1 before actual application in the network. Normalized input (Xnor) and output (Ynor)
are computed as follows

X  X min
X nor  , Ynor  Y (2)
X max  X min Ymax

where, Xmax and Xmin are the maximum and minimum real values of input and Ymax are maximum real
values of output.

2.2 Forward Pass Computation

Normalized input data is multiplied with corresponding initial and subsequently updated weights to
produce activation values of hidden layer neurons. For the jth neuron of the hidden layer, activation
value (Mj) is calculated as,
T
M j  w j x , for j= 1, 2,………, HLN, (3)

where weight vector wj denotes the association between inputs and jth node of the hidden layer and can
be defined as, W j   W W W W W  , n is the number of hidden layer neurons. Mj is
 j1 j2 j3 j4 j5 

59
 RTMST
further converted into output signal Oj of the jth hidden neuron through sigmoidal activation function
and Zj is given by,

1 (4)
Z j  f (M j ) 
M j
1 e

Output signals (Z) of all hidden layer neurons are computed in this way and are employed as input to
output layers neurons. All Z’s are further multiplied with corresponding weights between hidden and
output layer and converted into output signal (O) of the network through linear activation function.

2.3 Back propagation of error and updating weights

In BPNN the difference between network output O and desired output value Y is computed and
designated as error. The mean of the squared errors obtained during feeding of each training data in
forward pass, is minimized over iterations through updating weights. Mean square errors (MSE) are
1 Q
obtained as follows, E  MSE   (Ynor i  O i ) 2 (5)
Q i1
where Q is total number of training data. If computed MSE satisfies the convergence criterion,
training stops. Otherwise network weights are updated during back propagation and employed in next
iteration. Three BPNN algorithms adopted in the present work updates weights as follows,

BPNN with GDM:

It is the traditional BPNN algorithm [8] updates weights through first order approximation of errors. If
w k is the weight at k th iteration and w k denotes the weight gradient, weight in k+1 th iteration can
be updated in the following way:

  E k  
w k 1  w k  w k  w k   LR      MC  w k 1 
k 
(6)
  w  
Where LR is learning rate and MC is momentum coefficient for BPNN.

BPNN with LM:

LM algorithm has capability to model high nonlinearity in dataset and to achieve second order
approximation speed even without computing the Hessian matrix [9]. Weights in k+1th iteration will be
updated as
w  w  w  w  H  I  J E
k 1 k k k 1 T (7)

Where I is the unit matrix, μ  0 is the scaling coefficient, E is the squared error and J is the Jacobean
of error function.

BPNN with BR:

In order to improve generalization (or prediction) capability, BR minimizes Ф, which is a linear
combination of sum of the squared errors (SSE) and sum of the squared weights (SSW) in the
following way:
Φ  β  SSE  α  SSW (8)
2
where SSE is given by, SSE   Ynor  O  and value of regularization parameters α and β are
determined by the user. In any iteration, Ф is used to compute the Hessian matrix (H) in the following
way:
60
 RTMST
2 T
H   Φ  2β  J J  2α  I (9)
where, J is the Jacobean of the training set errors and I is the Identity Matrix. Weights are updated for
next iteration using the Hessian (H) in the same way as done by the traditional BP with LM algorithm.
In the process, regularization parameters α and β are also updated for next iteration.

2.4. Convergence and testing of ANN

The process continues till some convergence criterion is met. In order to test generalization capability
[8], the trained network is further fed with a 36 sets of test input dataset that was not used (i.e.
unknown to ANN) during training. The ANN output so produced is compared with corresponding
known experimental output to determine the prediction error. Accuracy in prediction is measured by
computation of test MSE. The testing phase thus indicates the efficacy in prediction capability of the
particular ANN architecture and algorithm.

3. Training and Testing with BPNN Algorithms

Single hidden layer BPNN with three algorithms namely, BPNN with GDM, BPNN with LM and
BPNN with BR have been used for training and testing of TIG welding dataset. The number of hidden
layer is used as design parameter of the network. During training with every BPNN algorithms the
number of neurons in hidden layer is varied from 6 to 15 at an increment of one neuron. Therefore
altogether 30 architectures have been trained and tested with three BPNN algorithms. During every
training and testing MSE is considered as criterion of performance of networks. The activation
functions for the hidden and output layers are sigmoidal and linear respectively. Each network is
trained over 300 epochs during batch mode BPNN training. All the training and simulation is done by
MATLAB 8 in a Pentium Dual core, 3.07GHz and 1GB PC.

4. Results and Discussions

Prediction capability of the network is considered as the criteria for selection of best ANN. Prediction
performance in terms of test MSE along with the training time has been provided in Table 1. Variation
of test MSE with hidden layer neurons for three different BPNN algorithms under study have been
plotted in Fig.2. From Table 1 and Fig. 2, it can be observed that during testing phase of ANN trained
using BPNN with BR, testing MSE does not varying with hidden layer neurons and remain fixed at
0.0048 from network 5-6-3 to 5-15-3. A network with lesser number of hidden layer neurons incurs
lesser training time. As the networks having hidden layer neurons from 6 to 15 produces same MSE,
the smallest network i.e. 5-6-3 network is considered here as best network and used for further
analysis.

Similarly, during testing of ANN trained using BPNN with LM the maximum error obtained is 3.2891
for 15-th hidden layer and the minimum mse is 0.0086 for 8-th hidden neuron. So the best network in
case of BPNN with LM is 5-8-3.

61
 RTMST
The Networks trained using BPNN with GDM indicates that, 5-7-3 network will produce the minimum
test MSE of 0.0759 and considered as best network. Above analysis indicates that 5-6-3 networks
trained and tested with BPNN with BR produces best predication accuracy among all tested networks.
Table 1: Training and testing performance of BPNN models

Training &Testing performance

Hidden
SL. BPNN with BR BPNN with LM BPNN with GDM
Layer
No Train Train Train
Neurons Test
MSE Test MSE MSE MSE Test MSE
MSE
1 6 2.67 0.0048 3.04 0.0078 2.64 0.0910
2 7 1.74 0.0048 2.02 0.0126 1.43 0.0759
3 8 2.24 0.0048 2.63 0.0086 0.723 0.1039
4 9 1.86 0.0048 1.91 0.0160 5.43 0.1415
5 10 1.85 0.0048 3.18 0.0565 2.35 0.1387
6 11 3.88 0.0048 2.19 0.0247 4.35 0.1489
7 12 1.97 0.0048 3.42 5.8255 3.68 0.2043
8 13 2.48 0.0048 3.51 0.1313 3.11 0.1848
9 14 2.80 0.0048 3.21 0.0747 4.94 0.1548
10 15 15.22 0.0048 4.30 3.2891 4.19 0.2652

Fig.2. comparison of prediction Performance of different ANN models with variation of hidden layer neurons

Fig.3 indicates comparison of training time for different networks. Training time observed for best
networks of BPNN with BR, LM and GDM are 2.99s, 3 s and 1.82 s respectively. Though training
time for BPNN with BR is high, its prediction accuracy is maximum among all tested networks. The
62
 RTMST
larger training time in that specific case may be due to complexity in the networks incurred during
evaluation of Hessian matrix.

Fig.3. Comparison of training time for different ANN models with variation of hidden layer neurons

Table 2: comparison of mean % absolute error for best networks for BPNN with BR, LM, GDM.

Training Best Max. % abs error Min. % abs error Mean % abs error
Algorithm network FW BH BW FW BH BW FW BH BW
BPNN
5-6-3 16.84 42.39 25.54 0.037 0.452 0.339 4.53 14.90 7.98
with BR
BPNN
with LM 5-8-3 20.04 94.26 41.68 0.233 1.168 0.176 4.94 23.47 10.20
BPNN with
5-7-3 213.96 245.19 219.49 4.39 6.13 1.17 61.60 74.20 62.19
GDM

Now prediction performance of best three networks i.e. 5-6-3 network for BPNN with BR, 5-8-3
network for BPNN with LM and 5-7-3 network for BPNN with GDM is further compared in terms of
absolute % error. The predicted outputs by the best networks are de-normalised and compared with
actual experimental output (or desired output) to produce absolute % error in prediction.
The absolute percentage error is calculated as

Experimental output - Predicted output of ANN (11)

Absolute % error in Prediction   100
Experimental output

The values regarding the percentage absolute error have been shown in Table 2 which clearly indicates
that in case of testing BPNN with BR the mean absolute % error for FW is 4.53, BH is 14.93 and BW
is 7.98, while in case of testing using BPNN with LM, FW is 4.94, BH is 23.47and BW is 10.20and in

63
 RTMST
case of testing using BPNN with GDM, FW is 61.60, BH is 74.20 and BW is 62.19. While comparing
these values it has been found that the mean percentage absolute error is minimum in case of testing 5-
6-3 network using BPNN with BR. Therefore accuracy in prediction performance indicate efficacy of
BPNN with BR for prediction of TIG welding parameters during online process control.

5. Conclusion
In the paper process parameters of TIG welding have been modeled and predicted by means of three
ANN training algorithms such as BPNN with BR, BPNN with LM and BPNN with GDM. Prediction
performance indicates that, 5-6-3 network during training and testing with BPNN with BR produces
best prediction capability with minimum test MSE of 0.0048 among 30 different tested networks.
Percentage absolute error obtained through comparing output of best networks for each BPNN
algorithms with experimental output indicates 5-6-3 network using BPNN with BR produces least
mean absolute % error (FW- 4.53%, BH- 14.90%, and BW- 7.98%). It is quite low compared to the
error obtained during testing with LM and GDM (FW- 61.60%, BH-74.20% and BW-62.19%). So it
can be concluded that for ANN modeling of TIG welding parameters, the prediction accuracy can be
best obtained in case of training and testing BPNN with BR.

References

[1]. S.C. Juang, Y.S. Tarng, H.R. Lii. A comparission between the back- propagation and counter-
propagation networks in the modeling of the TIG welding process, Journal of Materials Processing
Technology, 1996, pp. 54-62.
[2]. K.K.Aggarwal, Y. Singh, P. Chandra, M. Puri. Bayesian Regularization in a Neural Network Modelto
Estimate Lines of Code Using Function Points, Journal of Computer Science, 2005, pp.505-509.
[3] D.S. Nagesh, G.L. Datta. Genetic algorithm for optimization of welding variables for height to width ratio
and application of ANN for prediction of bead geometry for TIG welding process, Applied Soft
Computing, Vol. 10, Issue 3, 2010, pp. 897-907.
[4] P. Dutta, D. K. Pratihar. Modeling of TIG welding process using conventional regression analysis and
neural network-based approaches, Journal of Materials Processing Technology, Vol. 184, Issues 1–3,
2007, pp. 56-68.
[5] A.K. Lakshminarayanan, V. Balasubramanian. Comparison of RSM with ANN in predicting tensile
strength of friction stir welded AA7039 aluminium alloy joints, Transactions of Nonferrous Metals
Society of China, Vol. 19, 2007, pp. 9-18.
[6] F.D. Foresee, M.T. Hagan. Gauss-Newton Approximation to Bayesian Learning, Proceedings of
International joint Conference on Neural Networks, 1997, pp.1930-1935.
[7] K. Hornik, M. Tinchcombe, H. White. Multilayer Feed Forward Networks Are Universal Approximators,
IEEE Transaction on Neural Networks, Vol.2, 1989, pp.359-366.
[8] S. Haykin, Neural Networks: A Comprehensive Foundation, Pearson Education Inc, India, 2006.
[9] M.T .Hagan, M.B. Menhaj. Training feed forward networks with the Marquardt Algorithm. IEEE
Transactions on Neural Networks, Vol.5, Issue 6, 1994, pp. 989–993.

64
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18th and 19th April, 2013, NITTTR, Kolkata

Mathematical modeling of machining parameters for MRR in

EDM using regression analysis

Sandip KUNAR1, Abdul Halim KHAN2, Dr.D.C.ROY3

1
Department of Production Engineering, Jadavpur University, Kolkata-700032, India
1&2
Department of Mechanical Engineering, Jalpaiguri Government Engineering College, Jalpaiguri-
735102,
sandip.sandip.kunar@gmail.com(Corresponding author)

Abstract

Electrical Discharge Machining (EDM) is one of the earliest non-traditional machining processes used
for finishing components used in manufacturing as well as automotive industry, aerospace, medical
surgeries etc. EDM process is based on thermo-electric energy between the work piece and an
electrode. A pulse discharge occurs in the small gap between work piece and electrode and removes
the unwanted material from the parent metal through the simultaneous processes of melting and
vaporizing. The electrode and the work piece must have electrical conductivity in order to generate the
spark. The adequate selection of manufacturing conditions for machining of conductive materials
becomes one of the most important aspects to take into consideration in the die-sinking EDM. Material
Removal Rate (MRR) in EDM process is considered to be one of the most significant factors because
of its vital effect on the industrial economy. In this paper, regression analysis of Material Removal
Rate MRR is proposed for Sinker EDM operations. The experiments were carried out on Mild Steel
with copper electrode. Experiments were conducted by varying the controllable parameters -
machining voltage, peak current and pulse-on time. From experimental results, the MRR was
determined for each machining performance criteria. The experimentally found MRR with machining
parameters was then mathematically modeled by using the regression analysis method. The developed
model was validated with a set of experimental data and appeared to be satisfactory.

Keywords: Electrical Discharge Machining, Regression Analysis, Material Removal Rate.

1. Introduction
Electrical Discharge Machine (EDM) is now become the most important accepted technologies in
manufacturing industries since many complex 3D shapes can be machined using a simple shaped tool

65
 RTMST
electrode. Electrical discharge machine (EDM) is an important ‘non-traditional manufacturing
method’, developed in the late 1940s and has been accepted worldwide as a standard processing
manufacture of forming tools to produce plastics moldings, die castings, forging dies and etc. New
developments in the field of material science have led to new engineering metallic materials,
composite materials, and high tech ceramics, having good mechanical properties and thermal
characteristics as well as sufficient electrical conductivity so that they can readily be machined by
spark erosion. At the present time, Electrical discharge machine (EDM) is a widespread technique used
in industry for high precision machining of all types of conductive materials such as: metals, metallic
alloys, graphite, or even some ceramic materials, of whatsoever hardness [1]. Electrical discharge
machine (EDM) technology is increasingly being used in tool, die and mould making industries, for
machining of heat treated tool steels and advanced materials (super alloys, ceramics, and metal matrix
composites) requiring high precision, complex shapes and high surface finish. Traditional machining
technique is often based on the material removal using tool material harder than the work material and
is unable to machine them economically [2]. An electrical discharge machining (EDM) is based on the
eroding effect of an electric spark on both the electrodes used. Electrical discharge machining (EDM)
actually is a process of utilizing the removal phenomenon of electrical-discharge in dielectric.
Therefore, the electrode plays an important role, which affects the material removal rate and the tool
wear rate [3].

In this paper, a study will be performed on the influence of the factors of peak current, pulse on time
and voltage. The material used in this study is a Mild Steel. Mathematical models will be obtained
using Regression technique to select the optimum machining conditions for machining Mild Steel
using EDM [4].

The objectives of this paper are stated as follows:

 To evaluate the performance of EDM on Mild Steel with respect to various responses such as
surface quality and material removal rate (MRR).
 To establish mathematical model for all responses involved which are surface quality and
material removal rate (MRR).

2. Experimental set up
For this experiment the whole work can be down by Electric Discharge Machine, model AGIETRON
COMPACT 1 (die-sinking type) with positive polarity for electrode was used to conduct the
experiments. Commercial grade EDM oil (specific gravity= 0.763, freezing point= 94°C) was used as
dielectric fluid with a flushing pressure of 0.2 kgf/cm2 .Experiments were conducted with positive
polarity of electrode [5].

66
 RTMST

Figure 1. Electric Discharge Machine, model AGIETRON COMPACT 1 (die-sinking type)

A number of experiments were conducted to study the effects of various machining parameters on
EDM process. These studies were undertaken to investigate the effects of discharge current, pulse on
time and voltage on Material Removal Rate. The selected workpiece material for the research work is
Mild steel. Experiments were conducted on AGIETRON COMPACT 1 die sinking machine. An
electrolytic pure copper with a diameter of 30 mm was used as a tool electrode (positive polarity) and
workpiece materials used were steel square plates of dimensions 95*85mm2 and of thickness 6 mm.
Commercial grade EDM oil ( specific gravity = 0.763, freezing point = 940C ) was used as dielectric
fluid. Lateral flushing with a pressure of 2.94 X 105 Dyne/cm2 was used.

2.1 Conduct of experiment

Copper tube tool with 38mm diameter was used as electrode and the AGITON (die-sinking type) of
EDM machine are used. Commercial grade EDM oil (specific gravity= 0.763, freezing point= 94°C)
was used as dielectric fluid. Copper was used as tool material.

For a three factor are tackled with a total number of 8 experiments were performed on die sinking
EDM. This machine capacity is 1 kg and accuracy is 0.001 gram. Design factors selected There are a
large number of factors to consider within the EDM process, but in this work voltage (V) , discharge
current (I) and pulse- on time (Ton) have only been taken into account as design factors. The reason
why these three factors have been selected as design factors is that they are the most widespread and
used amongst EDM researchers.

2.2 Dielectric fluid

A dielectric is an electrical insulator that can be polarized by an applied electric field. When a
dielectric is placed in an electric field, electric charges do not flow through the material as they do in a
conductor, but only slightly shift from their average equilibrium positions causing dielectric
polarization. Because of dielectric polarization, positive charges are displaced toward the field and
negative charges shift in the opposite direction. This creates an internal electric field which reduces the
overall field within the dielectric itself. If a dielectric is composed of weakly bonded molecules, those
molecules not only become polarized, but also reorient so that their symmetry axis aligns to the field.

67
 RTMST
Although the term "insulator" implies low electrical conduction, "dielectric" is typically used to
describe materials with a high polarizability. The latter is expressed by a number called the dielectric
constant. The term insulator is generally used to indicate electrical obstruction while the term dielectric
is used to indicate the energy storing capacity of the material (by means of polarization). A common,
yet notable, example of a dielectric is the electrically insulating material between the metallic plates of
a capacitor. The polarization of the dielectric by the applied electric field increases the capacitor's
surface charge [6].

During the EDM process the workpiece and the electrode are submerged in the dielectric oil, which is
an electrical insulator that helps to control the arc discharge. The dielectric oil, that provides a means
of flushing, is pumped through the arc gap. This removes suspended particles of workpiece material
and electrode from the work cavity.

The dielectric fluid has the following functions:

(a) It helps in initiating discharge by serving as a conducting medium when ionized, and conveys the
spark. It concentrates the energy to a very narrow region.
(b) It helps in quenching the spark, cooling the work, tool electrode and enables arcing to be prevented.
(c) It carries away the eroded metal along with it.
(d) It acts as a coolant in quenching the sparks.

2.3 Tool Material

Electrode material should be such that it would not undergo much tool wear when it is impinged by
positive ions. Thus the localized temperature rise has to be less by tailoring or properly choosing its
properties or even when temperature increases, there would be less melting. Further, the tool should be
easily workable as intricate shaped geometric features are machined in EDM. Thus the basic
characteristics of electrode materials are:
• High electrical conductivity
• High thermal conductivity
• Higher density
• High melting point
• Easy manufacturability
• Cost – cheap
Just as the workpiece is vaporized, the electrode wears as well. It can lose its shape before a cavity is
completed. For this reason, the electrode must be made from a material that is much less affected by
this process. An electrode material must be selected that does not wear down (or one that wears much
slower). The electrode material must also be conductive [7].
 Graphite is the material most often chosen for the electrodes. It is easy to machine, conducts
electricity, and has good wear characteristics. However, there are other materials that work well
under certain conditions.

 Pure tungsten, tungsten carbide, copper, and tungsten are only a few of the additional materials
used for electrodes. Each of them offers different characteristics to produce a desired result.
Depending on the accuracy and surface finish required, multiple electrodes may be required to
complete the job. In sinker EDM terms, these are referred to as rough, semi-finish, and finishing
electrodes. As a finishing electrode begins to wear, it can be used as a semi-finish. As a semi-finish
begins to wear it can be used as a rough. The new electrodes being introduced into the process would
be the finishers.

68
 RTMST
2.4 Selection of electrode
A pipe copper tool is being selected for the machining because it is a highly conductive tool, low cost,
low wear ratio, good machinability and finishing. In this research, there are two types of electrode
have been used i.e. cylindrical electrode (38mm dia,20 mm dia,10 mm dia.) and rectangular electrode
(9*7). Initially a cylindrical copper tool is being drilled to produce a 0.5 mm hole with 38 mm dia.

2.5 Evaluation of MRR

Material removal rate (MRR) refers to the amount of material removed from the workpiece per unit
time. Material removal rate is calculated by measuring the depth of cut and then multiplying it with the
area of cross-section of the cut.

MRR= ………… (1)

Whereas V = Total volume cut by electrode.

t = Machining time.

Table 1. Parameter selection and range selection

Control Level
Observed value
parameters Min. level Max. level
Voltage(volt) 20 30
Current(amp) 50 60 Material Removal
Rate(mm3/min)
Pulse on 150 200

Table 2. MRR determination

Pulse On Material Removal

SL. NO. Current (amp.) Voltage (volt.)
Time(µs) Rate(mm3/min)
1 20 50 150 18.89
2 20 50 200 20.55
3 20 60 150 19.63
4 20 60 200 25.12
5 30 50 150 21.60
6 30 50 200 23.63
7 30 60 150 22.91
8 30 60 200 25.61

3. Regression analysis
Based on the experimental data gathered, statistical regression analysis enabled to study the correlation
of process parameters with the MRR.

69
 RTMST
In this study, for three variables under consideration, a polynomial regression issued for modeling. The
coefficients of regression model can be estimated from the experimental results. The effects of these
variables and the interaction between them were included in this analyses and the developed model is
expressed as interaction equation:

Y=a+bX1+cX2 +……. +nXm …………. (2)

Where a, b, c etc are co-efficient of their corresponding parameter.

The unknown coefficients are determined from the experimental data. Since, EDM process is non-
linear in nature, a linear polynomial will be not able to predict the response accurately, and therefore
the second-order model (quadratic model) is found to be adequate to model the process.

Table 3. Result of experiment for performing regression analysis

SL. Material
Current Voltage Pulse On
NO A B C Removal
(Amp.) (Volt.) Time(µs)
. Rate(mm3/min)
1 20 50 150 -1 -1 -1 18.89
2 20 50 200 -1 -1 1 20.55
3 20 60 150 -1 1 -1 19.63
4 20 60 200 -1 1 1 25.12
5 30 50 150 1 -1 -1 21.60
6 30 50 200 1 -1 1 23.63
7 30 60 150 1 1 -1 22.91
8 30 60 200 1 1 1 25.61

Here current, Voltage and pulse on time denoted as A, B and C.

The above Equation can be rewritten as

Y = Co + Ca*A + Cb*B + Cc*C + Cd*A*B + Ce*A*C + Cf*B*C ……….. (3)

Putting the value of A, B, C in above equation

A= (X1 – 25)/5 ………. (4)

B = (X2 – 55)/5 ……… (5)

C = (X3 – 175)/25 ……….. (6)

We get,

Y = 22.5601+0.833877 * X1 – 0.32278 * X2 –0.1824 * X3 –0.010044 * X1X2--0.0002433 *X1X3 +

0.0045072 * X2X3 ……… (7)

Where: X1, X2 and X3 are Discharge Current (IP), Voltage (V) and pulse on time (Ton) respectively.

Table 4. Results showing the experimental and predicted value and error
70
 RTMST

Pulse On
SL. NO. Current (amp.) Voltage (volt.) Exp. MRR Pred. MRR Error
Time(µs)
1 20 50 200 18.89 18.77 0.1228
2 20 50 250 20.55 20.68 0.1189
3 20 60 200 19.63 20.29 0.6664
4 20 60 250 25.12 24.45 0.6703
5 30 50 200 21.60 21.72 0.1183
6 30 50 250 23.63 23.50 0.1237
7 30 60 200 22.91 22.24 0.6710
8 30 60 250 25.61 26.27 0.6657

3.1 Graph of experimental MRR, predicted MRR vs. no. experiment

The results obtained from the experiments are compared with the predicted value calculated from the
model. It can be seen that the regression model is reasonably well fitted with the observed values.

S c a tte r p lo t o f E X P . M R R , P R E D . M R R v s n o . o f r u n s
0 2 4 6 8
EX P. M R R PR ED. M R R
26 27

25 26

24 25

24
23
23
22
22
21
21
20
20
19 19

18 18
0 2 4 6 8
no. of runs

Figure 2. Scatter plot of experimental MRR and predicted MRR vs. no of runs

3.1.1 Effect of discharge current

Discharge current is one of the important process parameters in EDM as it directly governs the spark
energy. Figure shows that MRR increases with increase in discharge current. These trends match with
the experimental results and predicted results. Higher values of discharge current is recommended for
rough machining, while lower values are recommended for finishing operations.

71
 RTMST

S c a tte r p lo t o f M R R v s c u r r e n t
2 6 .5

2 6 .0
MRR

2 5 .5

2 5 .0

2 4 .5

20 22 24 26 28 30
curre nt

Figure 3. Effect of discharge current on MRR

At higher values of discharge current, the tool electrode erodes faster affecting the cost and
productivity of the EDM process.

3.1.2 Effect of spark on time

Spark on time is another important process parameter in EDM, which decides the time for the
discharge energy to be applied on the work surface during the total spark time (on+off).

S c a tte r p lo t o f M R R v s p u ls e o n tim e

26

25
MRR

24

23

22
150 160 170 180 190 200
p u ls e o n t im e

Figure 4. Effect of pulse-on time on MRR

Figure shows the variation of MRR with spark on time for a typical machining condition for Mild steel
and copper electrode. The amount of MRR increases with the increase of spark on time.

3.1.3 Effect of discharge voltage

Discharge voltage is an important process parameter which governs the heat flux applied on the
cathode surface.

72
 RTMST

S c a tte r p lo t o f M R R v s v o lta g e
2 6 .5

2 6 .0

2 5 .5
MRR

2 5 .0

2 4 .5

2 4 .0

2 3 .5

50 52 54 56 58 60
v o lt a g e

Figure 5. Effect of gap voltage on MRR

Fig.5 shows the variation of MRR with discharge voltage for various values of discharge current and
spark on time. MRR are seen to increase with the discharge voltage. Higher values of discharge
voltage increase the flux density producing higher local temperatures both at anode and cathode
surfaces. Higher values of discharge voltages are recommended for roughing application.

From the above results, it is seen that the performance measures of EDM process are
influenced by three interacting process parameters, viz. discharge current, discharge duration and
discharge voltage. A complex nonlinear relationship seems to exist between these process parameters
and process performance measures such as MRR.

4. Conclusion and scope of further work

In this research, a second order polynomial regression model of three parameters are developed and it
is found that discharge current, voltage and pulse on time significantly affect the material removal rate
(MRR). Material Removal Rate is directly proportional to linear effect of discharge current and pulse
on time and voltage. The maximum value of material removal rate is achieved with Ip = 30 amp, ton =
200 μs and V = 60 volt.

Experimental investigations were performed to consider the machining characteristics in EDM process
of Mild steel and the following results were concluded:
1. The regression technique is an important tool for representing the relation between machining
characteristic and EDM process input parameters, and to obtain mathematical models.
2. The proper and optimized input parameters to achieve a specific output parameter (MRR) and a
higher efficiency can be determined by theoretical and experimental characteristic diagrams,
especially the two dimensional scatter diagrams.
3. From this research it is observed that material removal rate increases with the increase of the peak
current, voltage and pulse-on time. Maximum value of material removal rate in this study was
achieved when Ip = 30 A, Ton = 200 μs and V = 60 volt.
4. Surface roughness detoriates when there is an increase amount of current and voltage with the low
value of current the surface roughness.

This research can also help researches and industries for developing a robust, reliable
knowledge base and early prediction of Material Removal Rate without experimenting with EDM
process.
73
 RTMST

A further research can be performed to obtain good surface finish and accuracy even on
increasing the amount of current and voltage.

References
[1]. Mohan Kumar Pradhan and Chandan Kumar Biswas “Modelling of machining parameters for MRR in
EDM using response surface methodology” Proceedings of NCMSTA’08 Conference, National
Conference on Mechanism Science and Technology: from Theory to Application November 13-14, 2008
National Institute of Technology, Hamirpur
[2].Vishal Parashar , A.Rehman, J.L.Bhagoria, Y.M.Puri “Statistical and regression analysis of Material
Removal Rate for wire cut Electro Discharge Machining of SS 304L using design of experiments” Vishal
Parashar et. al. / International Journal of Engineering Science and Technology Vol. 2(5), 2010, 1021-1028
[3]. Kuang-Yuan Kung & Jenn-Tsong Horng & Ko-Ta Chiang “Material removal rate and electrode wear ratio
study on the powder mixed electrical discharge machining of cobalt-bonded tungsten carbide” Int J Adv
Manuf Technolgy (2009) 40:95–104 DOI 10.1007/s00170-007-1307-2
[4]. Mohd Amri Lajis, H.C.D. Mohd Radzi, A.K.M. Nurul Amin “The Implementation of Taguchi Method on
EDM Process of Tungsten Carbide” European Journal of Scientific Research ISSN 1450-216X Vol.26 No.4
(2009), pp.609-617
[5]. S. S. KHANDARE & MITESH A. POPAT “Experimental Investigations of EDM to optimize Material
Removal Rate & Surface Roughness through Taguchi’s Technique of Design of Experiments” Second
International Conference on Emerging Trends in Engineering and Technology, ICETET-09
[6]. M. M. Rahman, Md. Ashikur Rahman Khan, K. Kadirgama M. M. Noor and Rosli A. Bakar “Modeling of
Material Removal on Machining of Ti-6Al-4V through EDM using Copper Tungsten Electrode and Positive
Polarity” World Academy of Science, Engineering and Technology 71 2010
[7]. M.R.Shabgard, R.M.Shotorbani “Mathematical Modeling of Machining Parameters in Electrical Discharge
Machining of FW4 Welded Steel” World Academy of Science, Engineering and Technology 52 2009

74
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Experimental Investigation and Modeling of Controllable

Parameters in CNC Wire-EDM

1
KINGSHUK MANDAL, 2 SUBRATA MAHATA, 3 DIPANKAR BOSE
1
Mechanical Engineering Department, Camellia Institute of Engineering, Madhyamgram, Kolkata-129
Phone No: +91-9674210974, Email: mandal.kingshuk1@gmail.com
2
Production Engineering Department, Jadavpur University, Kolkata-700032
Phone No: +91-9932859165, Email: subrata.2k9@gmail.com
3
Mechanical Engineering Department, National Institute of Technical Teachers’ Training & Research,
Kolkata-700106, Phone No: +91-9831631887, Email: dipan88@yahoo.co.in

Abstract
Computer Numerical Control Wire Electrical Discharge Machining (CNC-WEDM) is a thermo-
electric non-conventional machining process. WEDM process is playing a pivotal role in the
production of newer materials, especially for the aerospace and medical industries or high temperature
application area. Using WEDM technology, complicated cuts as well as shapes can be made through
difficult-to-machine electrically conductive components. The high degree of accuracy and the fine
surface quality make WEDM valuable. The present investigation deals with the formulation of
mathematical model for the modeling and analysis of the effects of machining parameters on the
performance characteristics in the WEDM process of Inconel 718 alloy. The proposed model has been
developed using a second order model of Response Surface Methodology (RSM) to explain the
influence of four significant machining parameters [Wire feed rate (WFR), Gap voltage (G), Pulse on
time (Ton), Pulse off time (Toff)] on the selected responses i.e. Material Removal Rate (MRR), Kerf
Width (k) and Surface Roughness (SR). The experimental plan adopts full-factorial central composite
design (2k factorial) for its computation. The separable influence of individual machining parameters
and the interaction between these parameters are also investigated by using the Analysis of Variance
(ANOVA). After analyzing the experimental as well as predicted results by formulating design matrix
and through regression analysis it has been observed that the model adopted is of significant
importance and the predicted results are in full coherence with the experimental findings.

Keywords
Wire-EDM, Inconel 718, Surface roughness (Ra), Response Surface Methodology (RSM), Analysis of
Variance (ANOVA).

1. INTRODUCTION
Electrical discharge machining is a machining method primarily used for hard metals or those that
would be very difficult to machine with traditional techniques. EDM typically works with materials
that are electrically conductive, although methods for machining insulating ceramics with EDM have
also been proposed. EDM can cut intricate contours or cavities in pre-hardened steel without the need
for heat treatment to soften and re-harden them. This method can be used with any other metal or
75
 RTMST
metal alloy such as titanium, tungsten and Inconel. Also, applications of this process to shape
polycrystalline diamond tools have been reported [1]. Ideally, EDM can be seen as a series of
breakdown and restoration of the liquid dielectric in-between the electrodes. However, caution should
be exerted in considering such a statement because it is an idealized model of the process, introduced
to describe the fundamental ideas underlying the process. Yet, any practical application involves many
aspects that may also need to be considered. For instance, the removal of the debris from the inter-
electrode volume is likely to be always partial. Thus the electrical proprieties of the dielectric in the
inter-electrodes volume can be different from their nominal values and can even vary with time. The
inter-electrode distance, often also referred to as spark-gap, is the end result of the control algorithms
of the specific machine used [2]. The presence of these small craters on the tool results in the gradual
erosion of the electrode. This erosion of the tool-electrode is also referred to as wear. Strategies are
needed to counteract the detrimental effect of the wear on the geometry of the work piece. One
possibility is that of continuously replacing the tool-electrode during a machining operation. This is
what happens if a continuously replaced wire is used as electrode. In this case, the correspondent EDM
process is also called wire EDM. The tool-electrode can also be used in such a way that only a small
portion of it is actually engaged in the machining process and this portion is changed on a regular basis
[3].

2. EXPERIMENTAL SETUP AND WORKPIECE MATERIAL

Fig. 1: Photographic view of experimental set up (EZEECUT PLUS Wire Cut EDM)

The entire experiment was conducted on the EZEECUT Wire-Cut Electrical Discharge Machine. It
consists of
(i) A machine tool.
(ii) A power supply unit.
(iii) A dielectric unit.
Software named RRAPTV3.XX was used to make the contours used during cutting.
Inconel 718 which is nickel based super alloy. It has a wide range of application in different industries,
automobile and medical domain.

3. RESPONSE SURFACE METHODOLOGY (RSM)

RSM is an important subject in the statistical design of experiments, the response surface methodology
(RSM) is a collection of mathematical and statistical techniques useful for the modeling and analysis
76
 RTMST
of problems in which a response of interest is influenced by several variables and the objective is to
optimize this response [4].
At present, RSM is successfully used and applied in many divers’ fields having problems on response
such as chemical engineering, agricultural and biological and even computer development and process
improvement. RSM has a low order non linear behavior with regular experimental domain and
respectively small factor reason, due to its limitation in building a model to fit the data over a irregular
experimental region. Moreover, the main advantage of RSM is its ability to exhibit the factor
contributions from the coefficients in the regression model [5].

The first-order model for two independent variables in RSM interaction introduces curvature with the
response function. Often the curvature in the true response surface is strong enough that the first order
model (even with the interaction term include) is inadequate. A first-order model (two variables) as
follows:
η = β0 + β1x1 + β2x2 (1)

Where η = Output or response,

β0, β1, β2 = Linear regression coefficient and
x1, x2, = Input independent variable.

A second-order model will likely be required in these situations. For the case of two variables, the
second order model is-

η = β0 + β1x1 + β2x2 + β11x21 + β222x2 + β12x1x2 (2)

Where β11, β12 = coefficients (interaction),

β222 = coefficient (quadratic) and
x12 = input variables (quadratic).

4. RESULT ANALYSIS AND MATHEMATICAL MODELING

4.1 Experimental results
Table 1: Experimental data

Experimental Information Results

Std Run Machining Parameters

Order Order
Wire Pulse- Pulse- Gap Kerf Width Material Surface
feed on- off- voltage removal Roughness
(k)
rate Time Time rate (SR)
(MRR)
1 9 45 23 5 14 0.398 6.015 3.181

77
 RTMST
2 19 75 23 5 14 0.469 8.844 3.455

3 20 45 53 5 14 0.281 4.911 4.698

4 13 75 53 5 14 0.310 5.388 4.999

5 5 45 23 9 14 0.265 2.122 3.698

6 2 75 23 9 14 0.345 4.081 4.175

7 1 45 53 9 14 0.280 3.078 4.270

8 16 75 53 9 14 0.272 2.614 4.778

9 11 45 23 5 34 0.285 4.910 2.589

10 14 75 23 5 34 0.386 6.328 3.179

11 12 45 53 5 34 0.337 5.499 4.185

12 10 75 53 5 34 0.310 4.533 4.867

13 7 45 23 9 34 0.313 2.699 3.600

14 6 75 23 9 34 0.353 3.261 4.355

15 8 45 53 9 34 0.400 5.322 4.311

16 15 75 53 9 34 0.375 3.501 5.231

17 3 60 38 7 24 0.411 5.670 4.412

18 17 60 38 7 24 0.421 5.699 4.415

19 18 60 38 7 24 0.400 5.611 4.419

20 4 60 38 7 24 0.411 5.635 4.400

21 27 30 38 7 24 0.360 3.300 3.101

22 23 90 38 7 24 0.328 4.338 4.220

23 26 60 8 7 24 0.300 4.166 2.500

24 24 60 68 7 24 0.258 3.301 4.818

25 30 60 38 3 24 0.303 6.599 4.199

26 25 60 38 11 24 0.370 1.652 4.951

78
 RTMST
27 28 60 38 7 4 0.438 6.707 4.444

28 22 60 38 7 44 0.598 6.441 4.396

29 29 60 38 7 24 0.404 5.681 4.421

30 21 60 38 7 24 0.423 5.688 4.416

4.2 Analysis for Kerf width (k)

Table 2: ANOVA for kerf width

Source DF Seq SS Adj MS F P

Regression 14 0.136 0.009 7.15 0.000

Linear 4 0.015 0.012 8.88 0.001

Square 4 0.083 0.020 15.14 0.031

Interaction 6 0.037 0.006 4.55 0.008

Residual error 15 0.020 0.001
Lack-off-fit 10 0.019 0.001 4.27 0.001
Pure error 5 0.001 0.000

Total 29 0.156

S = 0.0368564, PREES = 0.1155880, R-Sq = 98.97%, R-Sq (pred) = 97.10%, R-Sq (adj) = 98.81%
N or ma l P r ob a bility P lot
( r e s p o n s e is K e r f W id th )
99

95

90

80
70
Percent

60
50
40
30
20

10

1
- 0 .0 5 0 - 0 .0 2 5 0 .0 0 0 0 .0 2 5 0 .0 5 0 0 .0 7 5
R e s id u a l

Fig 2: Normal probability plot for kerf width

79
 RTMST

4.3 Analysis for Material Removal Rate (MRR)

Table 3: ANOVA for MRR

Source DF Seq SS Adj MS F P

Regression 14 74.329 5.309 10030.20 0.000

Linear 4 39.350 4.663 8809.23 0.000

Square 4 16.729 4.182 7901.41 0.002

Interaction 6 18.249 3.041 5746.01 0.005

Residual error 15 0.007 0.000

Lack-off-fit 10 0.0032 0.000 0.19 0.008

Pure error 5 0.005 0.001

Total 29 74.337

S = 0.0230071, PRESS = 0.0207816, R-sq = 99.99%, R-sq (pred) = 99.97%, R-sq (adj) = 99.98%.

N o r m a l P r o b a b i li ty P lo t
( r e s p o n s e is M R R )
99

95

90

80
70
Percent

60
50
40
30
20

10

1
- 0 .0 5 0 - 0 .0 2 5 0 .0 0 0 0 .0 2 5 0 .0 5 0
R e s id u a l

Fig 3: Normal probability plot for MRR

80
 RTMST
4.4 Analysis for Surface roughness (Ra)

Table 4: ANOVA for surface roughness

Source DF Seq SS Adj MS F P

Regression 14 13.951 0.996 704.26 0.000

Linear 4 10.757 0.610 431.43 0.000

Square 4 1.925 0.481 340.26 0.021

Interaction 6 1.268 0.211 149.36 0.003

Residual error 15 0.021 0.001

Lack-off-fit 10 0.020 0.002 3.56 0.011

Pure error 5 0.000 0.000

Total 29 13.972

S = 0.0412564, PREES = 0.2015880, R-Sq = 92.87%, R-Sq (pred) = 96.10%, R-Sq (adj) = 98.64%

N ormal P robability P lot

(response is R a)
99

95

90

80
70
Percent

60
50
40
30
20

10

1
-0.075 -0.050 -0.025 0.000 0.025 0.050
Re sidual

Fig 4: Normal probability plot for surface roughness

ANOVA is carried out for all others response models given in table 2, 3 &4. The ANOVA and the F-
test have been performed to check the adequacy of the models as well as the significance of the
individual’s parameters. The ANOVA table of the quadratic model with other adequacy measures R2,
R2 (adj), & R2 (pred) are given below of the corresponding table. The associated p-value is less than
0.05 for model (i.e. α = 0.05, or 95% confidence level) indicates that the model terms are statistically
significant. The values of lack-of-fit (F-value) are 4.27, 0.19 & 3.56 implies that the lack of fit is so
significant relative to the pure error.

81
 RTMST
Figure 2, 3 & 4 display the normal probability of the residuals for responses. The responses are falling
on a straight line, which means that the errors are normally distributed and also implies that the model
proposed adequate and there is no reason to suspect any violation of the independence or constant
variance assumption.

5. CONCLUSSION
An attempt has been made in the investigation for modeling and optimization of machining parameters
by response surface methodology using a brass wire and Inconel-718 work-piece material. The main
investigation centre around to study the effects of control parameters on the kerf-width, material
removal rate and surface roughness. Moreover, this work contributes to the development of a response
surface methodology based model for prediction of the desired machining quality in terms of outputs
or responses.

It is observed from validation experiments, for the developed regression equation, that the estimated
results are in good agreement with the measure data. This indicates that the developed model can
predict the response adequately within the limit of machining parameters being used.

REFFENCES

[1] J. Chris, R. Morgan, Ryan Vallance and R. Eric, Marsh., “Micro machining glass with polycrystalline
diamond tools shaped by micro electro discharge machining.” Journal of Micromechanics and Micro
engineering, volume 14, pp 1687–1692, 2004.
[2] M. Naotake, F. Yasushi, N. Takayuki, S. Tanic and F. Katsushi. “Assisting Electrode Method for
Machining Insulating Ceramics.” CIRP Annals – Manufacturing Technology. Volume 45, Issue 1, pp
201–204, 1996.
[3] R. F. Feng-Tsai Weng, Shyua and Chen Siang Hsub. “Fabrication of micro- electrodes by multi-EDM
grinding process”. Journal of Materials Processing Technology. Volume 140, Issues 1-3, 22, pp 332–
334, 2003.
[4] P. Paracha. “Some experimental Design and optimization Technique in Response surface
metholodogy.” Masters thesis.MSU-IIt, Iligan City, 1999.
[5] K. Y. Beyounis and A. G. Olabi. “Optimization of different welding process using statistical and
numerical approache reference guide”, Advance in engineering software, Vol.39, pp.483-496, 2008.

82
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Multi-Objective Optimization of Electrochemical Deburring

Process Using Grey-based Taguchi Method

1M. Singha, 2S. Sarkar, 3S. Mitra and 4A.S. Kuar

Production Engineering Department, Jadavpur University, Kolkata-32, India
1
E-mail: manoj.singha.ju@gmail.com
2
E-mail: s.sarkar.ju@gmail.com
3
E-mail: souren_mit@rediffmail.com
4
E-mail: askuar@rediffmail.com

Abstract: This paper discusses the application of Taguchi method in combination with grey relational
analysis to optimize the parameters for machining of burrs in electrochemical deburring (ECD) when
the performance characteristics include change in burr height and base material removal. A multi-
response performance characteristic was used for optimization of parameters with application of grey
relational analysis. The machining parameters are selected as deburring time, gap between tool and
burr tip, electrolyte concentration and deburring voltage where as the response variables selected as
change in burr height and base material removal. The results from confirmation runs indicated that the
determined optimal combination of machining parameters improved the performance of the machining
process.

Keywords: ECD; Taguchi method; Burr height; Base material removal

1. Introduction

Burrs are produced after machining operations, such

as grinding, drilling, milling, engraving or turning, etc. Deburring is basically smoothing the rough
edges or any other part of the metal that needs a smooth surface finish. There is no standard procedure
to remove burrs having different shape and dimensions. Conventional deburring processes necessitate
time, labour and other associated costs. Electrochemical deburring has been found to offer a potential
solution to these problems [4,7-8]. In ECD process any type of conducting materials can be used as the
83
 RTMST
workpiece material apart from of its hardness. Due to electrochemical dissolution between tool and
burr tip, the material removed from burr tip. The performance of this process is influenced by many
factors, such as deburring voltage, gap between tool and burr tip, electrolyte concentration and
deburring time and others. They have significant influences on change in burr height and base material
removal. Optimal selection of process parameters is very much essential, as it is a costly process to
increase burr removal rate considerably and at same time, to achieve lower base material removal. This
study aimed to find an optimal process parameter capable of higher removal of burr and reduce the
base material removal through Taguchi method in combination with grey relational analysis.

2. Literature Review

A lot of researchers have studied the ECD methods. Choi and Kim explain the mechanism of ECD by
using electroplated Cubic Boron Nitride (CBN) wheels that examined the deburring efficiency and the
deburring performance through various electrolytic currents and other electrochemical conditions for
an internal cross hole [5]. Ghabrial and Ebeid pointed out that stationary machining is suitable for
deburring, embossing, finishing of dies with intricate profiles for forging, pressure-casting and
extrusion processes [6]. Shome et al. presented an investigation on the ECD of stainless steel using the
response surface methodology [3]. However, further research work is needed to improve the accuracy
and performance of the electrochemical deburring process to yield better product quality.
Taguchi’s viewpoint for the robust design is an important and dominant tool for the purpose of
designing and improving product quality. The original Taguchi method is designed to optimize a
single quality characteristic. Sharma et al. presented a Taguchi methodology with L9 orthogonal array
is employed for optimizing both cut qualities individually and simultaneously [1]. Optimization of
multiple quality characteristics is much more complicated than that of a single quality characteristic
[9]. To overcome this problem grey based Taguchi method has been successfully applied in different
advanced manufacturing processes to optimize the control parameters [2]. However, the applications
of grey-based Taguchi method for optimizing electrochemical deburring process with multiple quality
characteristics have not been reported yet.
In the present investigation, ECD of die steel has been considered. This material is extremely difficult
to machine by conventional method due to its excellent strength property. Electrochemical deburring is
found to be a potential machining process which may be utilized effectively for machining of this
material and an effort is made to optimize the ECD process with consideration of multiple deburring
characteristics, e.g. change in burr height and base material removal using Taguchi method in
combination with grey relational analysis. Planned experiments and subsequent analyses are carried
out to find the optimal parametric combination. Additional experiment has been carried out to verify
predicted result at optimum level and an effective improvement is observed in deburring quality.

3. Experimentation

3.1 Experimental Set-up

To carry out experimental investigation, an ECD system having the provisions of controlled electrical
power supply, electrolyte flow system and machining chamber was developed. The electrical circuitry
of the ECD setup includes a DC power supply with electrical elements for sort-circuit prevention,
spark detection and auto tripping operation of the ECD system. The electrolyte flow system consists of
a filter, pump, electrolyte storage tank, pressure gauge and flow measuring device, etc.

84
 RTMST

Pressure
gauge
Flow meter

Power supply and

control unit
Tool
Electrolyte
Machining chamber distribution pipes
with work holding
device Electrolyte reservoir
Pump

Fig. 1. Developed ECD setup

The machining chamber is made of transparent perspex sheet. Workpiece was fixed with machining
chamber by the fixing arrangement. The fixture is used to hold the flat workpiece tightly and with a
proper accuracy for better performance. A strap clamp, a square block and a stud with nut used as a
work holding device. This strap clamp and stud are based on the lever principle to amplify the required
clamping force. By tightening the stud, clamping force is transferred to the workpiece. Square block
work as a fulcrum about which the lever acts while the clamping force is applied at the stud by
tightening the bolt. Tool holding device is a tool post and a tool holder and they were adjusted by
tightening the tool. The tool and workpiece material was copper and die-steel respectively. A
photograph of the developed ECD set-up is shown in Fig. 1.

3.2 Experimental Planning

The experimentation scheme based on Taguchi method was designed in such a way as to explore the
influence of the various dominant machining parameters. The actual values of process parameters at
various levels in this set of investigations are as shown in Table 1. The experimental plan chosen for
studying the relationship between the controlling parameters and the various machining criteria, L16
orthogonal array and is shown in Table 2.
For carrying out the experiments, a rectangular tip solid copper tool and triangular edge workpieces
were used. NaNO3 salt solution of varying concentration was chosen as the electrolyte, because of its
good conductivity and non-passive characteristics.

Table 1. Independent factor and their levels

Character Control factors Levels

1 2 3 4
V Voltage (volts) 13 17 21 25
G Gap between electrode tip (mm) 0.1 0.2 0.3 0.4
C Concentration (gm/lit) 16 20 24 28
T Time (min) 3 4 5 6

Table 2. Combination of input factors and the measured responses

85
 RTMST

Exp. Control Factors Responses

No. V (volts) G (mm) C (gm/lit) T (min) Change in burr Base material
height (mm) removal (mm)
1 13 0.1 16 3 0.354 0.036
2 13 0.2 20 4 0.341 0.034

3 13 0.3 24 5 0.302 0.023

4 13 0.4 28 6 0.389 0.019
5 17 0.1 20 5 0.507 0.041
6 17 0.2 16 6 0.496 0.039
7 17 0.3 28 3 0.318 0.015
8 17 0.4 24 4 0.317 0.013
9 21 0.1 24 6 0.803 0.034
10 21 0.2 28 5 0.837 0.028
11 21 0.3 16 4 0.343 0.022
12 21 0.4 20 3 0.325 0.016
13 25 0.1 28 4 0.942 0.034
14 25 0.2 24 3 0.887 0.033
15 25 0.3 20 6 0.716 0.017
16 25 0.4 16 5 0.513 0.012

4. Experimental Observations
Fig. 2 and 3 show the main effects plots of four control factors on change in burr height and base
material removal. From Fig. 2 it is observed that change in burr height has an increasing trend with
increase of deburring voltage, electrolyte concentration and deburring time and the change in burr
height has a decreasing trend with the increase of gap between tool and burr tip. Current density
increases with increase of voltage and electrolyte concentration as a result burr removal rate increases.
With decreases of gap between tool and burr tip, current density increases in the gap between tool and
burr tip due to which removal rate of burr increases.
The base material removal rate has been found to have a decreasing trend with the increase of
deburring voltage, gap between tool and burr tip and electrolyte concentration, as depicted in Fig. 3.
Increase of deburring voltage and electrolyte concentration, causes quick removal of burr for which
gap increases and current density decreases, as result base material removal decreases. Similarly with
the increase of deburring time, the base material removal increases.

86
 RTMST

Main Effects Plot for Means

Fig. 2. Mean effects of ECD parameters on change in burr height

Main Effects Plot for Means

Fig. 3. Mean effects of ECD parameters on base material removal

5. Parametric Optimization of ECD Process

5.1 Evaluation of Optimal Process Condition

From production viewpoint higher amount of burr removal can be categorized as ‘larger-the-better’
problem. Minimization of base material removal can be categorized as a ‘smaller-the-better’ type
problem. These two contradictory response parameters i.e. change in burr height and base material
removal requires different level setting of the machining parameters for their optimization. So proper
machining parameter set up for simultaneous optimization of the two responses is important. In the
present paper, Grey relational analysis (GRA) has been used to renovate the multi-response
optimization model into a single response Grey relational grade.
In grey relational analysis, the raw experimental data is first normalized. The normalized
results, , for larger-the-better quality characteristics can be expressed as:

x = , i = 1,2,........,n; j = 1,2,……..,m (1)

For smaller-the-better quality characteristic, the normalized results, , can be expressed as:

x = , i = 1,2,........,n; j = 1,2,……..,m (2)

87
 RTMST

The grey relational coefficient is calculated to express the relationship between the ideal and actual
normalized experimental results. Ideal result is the best result which is expressed as . The relational
coefficient can be calculated as:
∆ ∆
Y( , )= , i = 1,2,........,n; j = 1,2,……..,m (3)
∆ ∆

where ∆ = − (4)

∆ = min[∆ , i = 1,2,........,n; j = 1,2,……..,m ] (5)

∆ = max[∆ , i = 1,2,........,n; j = 1,2,……..,m] (6)

ζ is the distinguishing coefficient, which is defined in the range of 0 ≤ ζ ≤ 1.

The grey relational grade can be determined as the average of the grey relational coefficients
associated with each response parameter. It can be expressed as:

Г( , )= ∑ ( , ) (7)

Equation (1) is used to normalize the experimental value when the target of the original value is having
the characteristic of larger-the-better. When the smaller-the-better is a characteristic, base material
removal is normalized using this equation (2) and the grey relational coefficient for change in burr
height and base material removal is determined as shown in Table 3. It also exhibits the grey relational
grade using equation (7). In the present report, experiment number 16 has the best multi-response
characteristics among the 16 experiments because it corresponds to the highest grey relational grade.
The mean grey relational grade for the voltage at levels 1, 2, 3 and 4 can be calculated by averaging
the grey relational grade for the experiments 1 - 4, 5 – 8, 9 – 12 and 13 – 16 respectively. Similarly the
mean grey relational grade is calculated for the rest parameters for their levels in Table 4.
The Fig. 4 shows the optimal parametric combination where the optimal ECD parameter setting is to
maintain voltage at level 4 (V=25volts), gap at level 4 (G=0.4mm), concentration at level 4
(C=28gm/lit) and time at level 2 (T=4min) for maximizing the change in burr height and for
minimizing the base material removal.

Table 3. Evaluation of grey relational grade

Exp. No. Normalized values Grey relational coefficients Grey Rank

relational
Change in burr Base material Change in burr Base material grade
height (mm) removal (mm) height (mm) removal (mm)
1 0.081 0.172 0.352 0.378 0.365 16
2 0.061 0.241 0.347 0.397 0.372 15
3 0 0.621 0.333 0.569 0.451 12
4 0.136 0.759 0.367 0.675 0.521 10
5 0.320 0 0.424 0.333 0.379 14
6 0.303 0.069 0.418 0.349 0.384 13
7 0.025 0.897 0.339 0.829 0.584 7
8 0.023 0.966 0.339 0.936 0.638 4

88
 RTMST
9 0.783 0.241 0.697 0.397 0.547 9
10 0.836 0.448 0.753 0.475 0.614 6
11 0.064 0.655 0.348 0.592 0.470 11
12 0.036 0.862 0.342 0.784 0.563 8
13 1 0.241 1 0.397 0.699 2
14 0.914 0.276 0.853 0.408 0.631 5
15 0.647 0.827 0.586 0.743 0.665 3
16 0.329 1 0.427 1 0.714 1

Table 4. Response table for overall grey relational grade

Deburring parameters Grey relational grade

Level 1 Level 2 Level 3 Level 4 Delta Rank
V 0.427 0.496 0.549 0.677 0.250 1
G 0.498 0.500 0.543 0.609 0.112 3
C 0.483 0.495 0.567 0.605 0.122 2
T 0.536 0.545 0.539 0.529 0.016 4
Total mean value of the grey relational grade = 0.537

Table 5. Results of confirmatory experiment

Initial parameter setting Optimal parameters

Prediction Experimental
Level V4 G4 C1 T3 V4 G4 C4 T2 V4 G4 C4 T2
Change in burr height (mm) 0.514 0.668
Base material removal (mm) 0.012 0.011
Grey relational grade 0.714 0.825 0.770
Improvement of the grey relational grade = 0.056

Voltage
Gap between tool and burr tip
Concentration
Time
Grey relational grade

Factor levels

89
 RTMST
Fig. 4. Plot of graph for grey relational grade

5.2 Confirmation Tests

After evaluating the optimal parameter settings, the final stage is to predict and verify the improvement
of quality characteristics using optimal parameter setting. The predicted optimum grey relational grade
can be calculated as:

γ =γ +∑ (γ − γ )
(8)

where:
γ = Total mean of the grey relational grade
γ = mean grey relational grade at optimum level
o = number of ECD parameters that significantly affect the quality characteristics.
Table 5 shows the comparison of the predicted and actual machining performance, using their
respective optimal parameters. The improvement in grey relational grade at optimum level proves the
effectiveness of the proposed approach in relation to product/process optimization, where more than
one objective has to be fulfilled simultaneously.

6. Conclusions

The grey relational analysis combined with the Taguchi method is applied to find the optimum
parameter setting for electrochemical deburring of die steel. Planned experiment and subsequent
analysis has been carried out to evaluate the optimal parametric combinations of different main
influencing process parameters, such as deburring voltage, concentration of the electrolyte, initial
inter-electrode gap and machining time on change in burr height and base material removal. The effort
has been made to minimize the base material removal and to maximize the change in burr height,
simultaneously. The proposed optimization technique converts the multiple quality characteristics of
electrochemical deburring process to a single performance characteristic, thus, simplifies the
optimization process. Finally a confirmation test is conducted which show effective improvement of
the quality characteristics in the electrochemical deburring process.

90
 RTMST
References
[1]. A. Sharma, V. Yadava, Optimisation of cut qualities during pulsed Nd:YAG laser cutting of SUPERNI 718
thin sheet for straight profile, International Journal of Manufacturing Technology and Management, Vol.
24, No.1/2/3/4, 2011, pp. 108 – 123.
[2]. B. Acherjee, A. S. Kuar, S. Mitra, and D. Misra, Application of grey-based Taguchi method for
simultaneous optimization of multiple quality characteristics in laser transmission welding process of
thermoplastics, International Journal of Advanced Manufacturing Technology, Vol. 56, 2011, pp.995–
1006.
[3]. D. Shome, S. Mitra and S. Sarkar, Response surface methodology-based approach to electrochemical
deburrring of SS304 stainless steel workpiece, International Journal of Product Development, Vol. 6, No.
1, 2008, pp.2-15.
[4]. I-H. Choi and J-D. Kim, A study of the characteristics of the electrochemical deburring of a governor-
shaft cross hole, Journal of Materials Processing Technology, Vol. 75, 1998a, pp.198–203.
[5]. I-H. Choi and J-D. Kim, Electrochemical deburring system using electroplated CBN wheels, International
Journal of Machine Tools and Manufacture, Vol. 38, 1998b, pp.29–40.
[6]. S.R. Ghabrial and S.J. Ebeid, Beneficial effect of air-electrolyte mixtures in stationery electrochemical
machining, Precision Engineering, Vol. 3, 1981, pp.221-223.
[7]. S. Sarkar, S. Mitra and B. Bhattacharya, Mathematical modeling for controlled electrochemical deburring
(ECD), Journal of Materials Processing Technology, Vol. 147, 2004, pp.241–246.
[8]. V.K. Jain, Advanced Machining Processes, Allied Publishers, New Delhi, 2002.
[9]. Y.S. Tarng and W.H. Yang, Optimization of the weld bead geometry in gas tungsten arc welding by
Taguchi method, International Journal of Advanced Manufacturing Technology, Vol. 14, No. 8, 1998,
pp.549-554.

91
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

AN INVESTIGATION ON TEMPERATURE DISTRIBUTION

OF AN ERW PIPE DURING TIG WELDING
Tathagata Bhattacharya a, Asish Bandyopadhyayb,

Pradip Kumar Palc

a
M/s M.N.Dastur & Co. (P) Ltd, P-17, Mission Row Extension, Kolkata-700013, West Bengal, India

tathagata.b@dasturco.in
b
Mechanical Engineering Department, Jadavpur University, Kolkata-700032, West Bengal, India

asishbanerjee@yahoo.com
c
Mechanical Engineering Department, Jadavpur University, Kolkata-700032, West Bengal, India

pradippal54@yahoo.com

Abstract:
The present investigation deals with the determination of temperature profile on
the external surface along axis of an ERW pipe, during TIG Welding. Three samples, each of OD 516
mm pipe of a given length of material ASTM A 106 Gr. B seamless type, are selected for
experimentation. The temperatures on the external surface at different locations on each of these
seamless pipes during welding have been measured with the help of electronic pyrometers. The
temperature variation along the axis of pipe from the weld arc has been represented in graphs. Then,
through separate analytical methods the temperature distribution across the external surface of a pipe
for a given point heat source is theoretically calculated, and the results are compared with the
measured temperatures. They are found to be well in agreement with the measured values of
temperatures.

Key words:
Weld arc, point heat source, pyrometer, temperature profile

92
 RTMST
Introduction:
TIG welding is a common industrial joining process. In this process, a welding
arc is maintained between the non consumable tungsten electrode and the work piece in a protective
inert gas atmosphere. This inert gas atmosphere is adopted in order to prevent the metal from being
oxidized. The TIG welding process gives a very high quality of finished weld.

In all process plants and steel plants, huge quantity of corrosive fuel gases is generated from various
chemical reactions. These corrosive fuel gases need to be carried from their generation points to their
consumption points, to be used as fuel and increase the overall plant fuel efficiency. For transporting
this large amount of gas, big diameter pipelines are fabricated and erected in the plant site, joining the
gas generation point to the gas consumption point. Metal sheets are welded to get a leak-proof, perfect
pipe joint. Corrosive fuel gases like Coke Oven gas, Converter gas and Blast Furnace gas are to be
carried through pipes that do not have any oxidized part inside to prevent explosion. Since TIG
welding protects the parent metal from getting oxidized during welding, it is, therefore, generally
adopted for welding of the pipes meant for the above mentioned services. Seamless pipes,
manufactured by extrusion process, are chosen for carrying of high velocity high pressure oxygen gas
in a steel plant. These pipes have high circumferential strength .They too are required to be welded in
such a way that does not leave any oxidized metal part in its internal surface to prevent accumulation
of kinetic energy of the flowing gas , leading to explosion. So for this special case of Oxygen pipe
welding also, TIG welding method is adopted.

V.v.d. Mee et. al. [1] carried out experiments for understanding of quality of TIG weld in connection
with hydrogen content .They found out that susceptibility to hydrogen cracking of the weld, using the
GMA or GTA process was limited. On the basis of experiments, they concluded that a moisture level
of even 1000 ppm in the shielding gas in combination with correct welding parameters resulted in
sufficiently low hydrogen content in the weld, thus making it brittle. M.F.Arenas and V.F. Acoff [2]
determined the weldability of Gamma Titanium Aluminide welds produced by Gas Tungsten Arc
Welding. Through experiment they determined the weld nature of spot and butt jointing of Gamma
Titanium Aluminides, at welding current levels of 50 A to 150 A, without pre-heat. They received
crack free weld joints for high welding currents and concluded that a careful selection of welding
parameters were required for successful joining of Ti-Al alloys using Gas Tungsten Arc Welding
process. J. Luo et. al. [3] performed experiments to determine a new approach to fluid flow model in
Gas Tungsten Arc weld pool using longitudinally electromagnetic control. By using a current density
measuring probe, they first calculated the distribution arc current density during TIG welding and
modeled a distribution of welding arc current flux. They also presented a new mathematical model for
body force in TIG weld pool and performed experiments with LD10CS Aluminium alloy to develop
the fluid flow in the weld pool for the control process. A. Traidia et. al. [4] presented a numerically
simulated Pulsed TIG Welding model with partial and full penetration and thereafter deduced that
heat transfer and fluid flow in the weld pool were driven by a combined effect of electromagnetic
force, buoyancy force, surface tension force and latent heat.

A considerable amount of welding power is spent for carrying out TIG Welding. Therefore, it is
important to obtain the welding energy analysis and vis-à-vis actual weld power required during
93
 RTMST
welding process. To make the welding process more energy efficient, effort should be given to gather
knowledge regarding the actual weld power utilization and weld energy loss through various means,
during welding. It is important to determine the temperature profile on the periphery of a metallic pipe
during welding, because that in turn helps to calculate the mean temperature of the pipe and vis-a-vis
helps in energy analysis. The determination of rate of cooling of the weld zone, during and after the
completion of welding, is also important because of determination of grain orientation and post weld
mechanical characteristics of the welded pipe. Very little reporting is available in so far as the
temperature distribution on the external pipe surface during TIG welding process is concerned. But
some literature is there in connection with temperature profile and weld bead orientations during
Electro-slag welding and Friction welding processes.

T.Debroy and J.Szekely [5] analytically developed a mathematical model to represent the three
dimensional temperature field in the slag, metal pool and base plate region of an Electro-slag welding
process. Then they carried out laboratory experiments to physically determine the temperature profile
in the slag, weld pool and base plate region of an Electro-slag welding process and found the analytical
and experimental results to be in fairly good agreement. They found out that in Electro-slag welding
system the Heat affected zone is relatively uniform and its size is not markedly affected by the process
parameters. Furthermore they concluded that the percentage dilution can be successfully influenced by
process parameters like liquidus temperature and plate gap. N. Karunakaran and V. Balasubramanian
[6] experimentally determined the Temperature distribution and weld bead profiles of constant current
and pulsed current gas tungsten arc welded aluminium alloy joints and compared the results. They
found out that the use of pulsed current technique improved the tensile properties, hardness profiles
and micro-structural grain sizes of the welds, compared with continuous current welding, due to grain
refinement occurring in the fusion zone. S.S.A.Akbari Mousavi and A. R. Kelishami [7] performed
experiments in friction welding to conclude that variations in welding temperature, stress, strain etc.
helps determine optimum weldment design. A finite Element method involving Johnson-Cook
equation was used by them for predicting the state of the coupled thermo-mechanical system. During
experiments, the variation in temperature, stress, strain rate, were continuously investigated and
analyzed. The results obtained through numerical analysis were found in good agreement with those
found during experimentation. But rare publications are found in respect to temperature profile on the
external pipe surface during welding processes. Hence in the present work a study has been carried out
for temperature profile on the external pipe surface during welding processes, in particular TIG
welding process.

Experimental procedure:
Three samples of pipe piece having outside diameters OD 516 mm and thickness 8 mm each, are
taken. This pipe sample is an oxygen gas pipeline which is extensively used in a Steel Plant piping
scenario, with specification of ASTM A 106 Gr. B. Each sample has a length of 125 mm, for
convenience of handling and welding purpose.

The specification of the Welding Machine used for the purpose is given below:

i) Make : Alstom SR Welding Systems

94
 RTMST
ii) Input voltage : 800 V
iii) Maximum current at output circuit : 1000 A

The welding electrodes have 20 mm diameter and 300 mm length. Argon and Nitrogen mixture is used
as inert gas at a pressure 7 kg/cm2 (gauge)

The set-up for TIG Welding is given in Fig. 1.

Fig.1. The experimental set-up for TIG Welding

The co-ordinate system is manually put up on the external surface of the pipe with the help of markers.
The Y-Axis and the X-Axis are set up as indicated below in Fig. 2. The Y Axis is the curved axis that
runs across the periphery of the pipe along its mouth, and the X axis is the one that runs along the
horizontal direction. Six points are selected on the semi-circle for y co-ordinate (since it is symmetric
about the diameter) and five such sections for different values of X have been selected to cover the
entire length of the pipe. The locations of the points are indicated in Fig. 2.

95
 RTMST

Fig. 2. Location of the points for measurement of temperature on the pipe surface

The instantaneous temperature is measured at the points with the help of a “Metravi” make pyrometer.
The technical specification of the pyrometer is as follows:

i) Model No. : Metravi MT-09.

ii) Type : Non contact digital thermometer with laser sensor for temperature
iii) Range : 500 C to 38000 C.
iv) Display : Back light LCD Display with auto-power off facility
v) Sighting Ratio : Distance of sighting ratio is 12:1.
vi) Emmissivity : 0.95
vii) Accuracy :  0.5%

The instantaneous temperature values that are measured for each pipe piece by the above-mentioned
instrument are recorded for tabulation and comparison. The same method is repeated for the rest of the
2 samples.

96
 RTMST
Mathematical Modeling:

The total power required for welding is P =V I  (1)

where,

V = Welding voltage (in V)

I = Welding current (in A)

 =Efficiency of the welding machine

In Fig. 3 below, a co-ordinate system developed on a thick metal sheet subjected to TIG welding is
represented.

Fig. 3. Co-ordinate System for measuring temperature due to a point heat source moving on the
surface of a thick plate

From theoretical study of heat transfer, [10], it is known, when a point heat source of power = q W, is
moving across the length of a plate at velocity (v m/s), in the direction of positive Y-axis as indicated
in Fig. 3. , then the instantaneous temperature at a point, P, (x, y), on the plate is given by:

( ̇) ( )/ )
= (2)

where,

r = x 2  ( R )2

k=conductivity constant

97
 RTMST
α=diffusion constant

For formulation of a moving point heat source, it is considered that the point heat source is static at the
origin and the plate is moving in the opposite direction with velocity v m/s.

Now, ̇ = VI η

where

V=Welding voltage

I=Welding Current

η =Welding efficiency

K = conductivity constant = 35 W/m/° C [8, 9]

 =diffusion constant = 1.2 X 10-5 m/s [8,9]

Now for applying this formula for curved surface of a pipe, the Y co-ordinate is replaced by polar co-
ordinate system and it is expressed as Y= Rθ

Using the values of the constant quantities, and making the necessary corrections for geometrical
orientations, the Radius of the pipe ( R) and Angle ( θ ) expressed in radians , the location of the points
on the surface of pipe take the shape as shown in Fig. 2.

The Fig.2. indicates the location of the points whose temperature is to be measured, lying on the
external surface of the seamless pipe.

So, by application of the standard formula of temperature distribution (after making necessary
corrections), on to a metallic pipe, the theoretical temperature distribution is calculated, and the
corresponding experimental temperature of the points lying on the pipe surface are also found out
simultaneously, by Pyrometer and the comparison is presented thereafter.

98
 RTMST
Accordingly, the OD 516 mm pipe is taken and it is subjected to the following welding condition as
shown in Table 1.

Radius of pipe 250 mm

Voltage 80 V

Current 80 A

̇ 5.44 kW

v 1.78 x 10-3 m/s

k 35 W/m/°c

 1.2 x 10-5 m/s

Table 1: Welding parameters of OD 516 mm, seamless pipe of material ASTM A 106, Gr. B

Now, for the above given welding conditions, for three different values of θ , the corresponding
experimental and theoretical temperatures as recorded for various values of x, are given below in the
following Tables 2 ,3 and 4.

Table 2 below shows the theoretical and experimental temperatures received for a value of θ = 0° on
the pipe surface.

Value of θ = 0°

Value of x in mm Theoretical temperature Measured temperature

in °C in °C

25 989.49 700
50 494.74 300
75 329.83 250
100 247.37 165
125 197.90 145

Table 2: Comparison of theoretical and experimental temperatures for θ= 0° and various values
of x ranging from 25 mm to 125 mm, on OD 516 mm pipe

Theoretically at x=0, θ=0°, we have r=0 and so temperature tends to ∞. Experimentally it is found that
temperature is very high at that point i.e. of the order of 1600° C.It is also found from the table that, as
values of x increases along the x axis, the corresponding temperature values asymptotically die out .

99
 RTMST
The comparative graphical representation of the theoretical and experimental temperatures as indicated
in Table 2 are given in Fig. 4.

Temp. (° C
)
1000
900
800
700
600
500
400
300
200 Dist. along x
100
0
0 50 100 150
Theoretical Temperature
Experimental Temperature

Fig. 4. :Graphical representation of theoretical and experimental temperatures indicated in

Table 2.

Table 3 below shows the theoretical and experimental temperatures received for a value of θ = 10° on
the pipe surface.

Value of θ = 10°

Value of x in mm Theoretical temperature Measured temperature

in °C in °C

0 5.61 24
25 33.93 30
50 66.1 31
75 82.15 37
100 86.59 40
125 85.48 39

Table 3: Comparison of theoretical and experimental temperatures for θ= 10° and various
values of x ranging from 0 to 125 mm, on OD 516 mm pipe.

100
 RTMST
It is found that, for θ=10°, temperature is somewhat less at x=0, but it asymptotically increases
gradually upto a certain value of x, i.e. x=100 and then again temperature decreases , as value of x
increases further beyond the point x=100.

The comparative graphical representation of the theoretical and experimental temperatures as indicated
in Table 3 are given in Fig. 5.

Temp. (° C
) 90

60

30

0 Dist. along x
0 50 100 150

Theoretical Temperature
Experimental Temperature

Fig. 5. : Graphical representation of theoretical and experimental temperatures indicated in

Table 3.

Table 4 below shows the theoretical and experimental temperatures received for a value of θ = 15° on
the pipe surface.

Value of θ = 15°

Value of x in mm Theoretical temperature Measured temperature

in °C in °C

0 0.37 24
25 3.01 24
50 9.81 24
75 18.55 24
100 26.30 26
125 31.96 30
Table 4: Comparison of theoretical and experimental temperatures for θ= 15° and various
values of x ranging from 0 to 125 mm, on OD 516 mm pipe.

101
 RTMST
It is found that, unlike the previous example, for θ=15°, theoretical temperature is almost 0° C at
x=0, but it asymptotically increases gradually , as x increases beyond the point x=0, whereas the
experimental temperature at any point never goes below the ambient temperature.

The following Fig.6 shows the comparative graphical representation of temperatures indicated in Table
4.

. p meT
100.00

75.00

50.00

25.00

0.00 Dist. along

0 50 100 150

Theoretical Temperature
Experimental Temperature

Fig. 6 :Graphical representation of theoretical and experimental temperatures indicated in Table

4

Conclusions:
From the above results the following conclusions can be drawn:

(1) For all the 3 samples of OD 516 mm pipes, it is found that theoretical temperature rise on the pipe
surface is always greater than the experimentally received values.

(2) The percentage difference between the theoretical and experimental temperature recorded varies
between 10 to 15% maximum in all the 3 samples.

(3) For all the three pipe samples it is noted that, at the point x=0, θ=0, the temperature is very high
and it asymptotically drops as x gradually increases.

(4) It is noted that for pipe OD 516 mm during determination of instantaneous temperature on the
pipe surface, beyond an arc of approximately 20° at the center of pipe, the temperature dies down
sharply and no appreciable temperature indication can be determined by pyrometer. Also the
theoretical value of temperature is less than the ambient temperature in some cases, but in practice the
experimental temperature never drops below the ambient.
102
 RTMST
Acknowledgement:
The authors sincerely express their gratitude and thankfulness to Mr. S. Bhaumik, as well as to Mr.
B.N.Rao, for their sincere support, and for rendering technical help for preparing this paper.

References:
[1].V.v.d. Mee, H. Meelker and R.v.d.Schelde., How to control Hydrogen level in duplex
stainless steel welding by making use of GTAW and TIG Welding process, American Welding
Journal, Vol. VIII, January 1999, pp. 65-70.

[2].M.F. Arenas & V.F. Acoff., Analysis of Gamma Titanium Aluminide welds by making use
TIG welding Process, American Welding Journal, Vol. II, May 2003, pp. 345-350.

[3].J. Luo, Q. Luo, Y.H.Lin and J. Xue., New approach in Fluid Flow Model in GTAW
and TIG Weld pool by using longitudinal electromagnetic control, American Welding Society
Journal, Vol. VIII. , August 2003, pp. 80-84.

[4].A. Traidia, F. Roger and E. Guyot , Numerical Simulation of Pulsed TIG Welding-
Partial and Full Penetration, American Welding Society Journal, Vol. VI. , August 2011,
pp. 86-91.

[5]. T. Debroy, J.Szekely and T.W.Eagar, Temperature profile, the size of the heat
affected zone and dilution in electro-slag welding, Journal of Material Science and
Engineering, MIT, Cambridge, USA, October 1982 Volume,: pp.181-193.

[6]. N. Karunakaran & V. Balasubramanian, Effect of pulsed current on temperature

distribution, weld bead profiles and characteristics of gas tungsten arc welded aluminium
alloy joints , Journal of trans non ferrous metallurgical society of China, April 2010
Volume, pp 278-286.

[7]. S.A.A.Akbari Mousavi & A. Rahbar Kelishami, Experimental and Numerical

Analysis of the Friction Welding Process for the 4340 Steel and Mild Steel Combinations,
American Welding Journal, Volume 87, July 2008, pp. 178-188.

[8]. Piping Handbook, 1976, pp. 35-38,1st Edition, Oxford University Press, Oxford, UK.

[9]. Piping Handbook, 1983, pp. 125-126, 1st edition, Tata-McGraw Hills, London, UK.

[10]. Metallurgy of Welding, 1982, pp-176, 1st Edition, Oxford University Press, Oxford, UK.

103
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18th and 19th April, 2013, NITTTR, Kolkata

An Experimental Investigation on Dimensional Accuracy and

Burr Formation in Slot Milling Operation

A. GHOSH1, A. RAWANI2, G. MAITY3, S. MAJEE4 and Santanu DAS5

Department of Mechanical Engineering, Kalyani Government Engineering College,
Kalyani- 741235, West Bengal
Phone No.: +919433956050, fax No. 5+913325821309
5

Email: 1a.ghosh865@gmail.com, 2rawani.jay@gmail.com, 3ganesh.maity@rediffmail.com,

4
sagar.kgec@rediffmail.com, 5sdas.me@gmail.com

Abstract: End milling is widely used in various manufacturing industries, including the automobile
and aerospace sectors where quality is an important factor in the production of plane surface, slots,
pockets, precision moulds and dies, etc. The quality of the milled surface is adjudged by the surface
finish, dimensional accuracy and presence of burr at the edges. In the last few decades, CNC milling
machines have been employed in manufacturing industries for better productivity, reproducibility and
improved quality. However, some inaccuracies are often detected in CNC operations. These
inaccuracies may come from gradual wearing out of the insert, vibration during machining, lack of
rigidity of machine tools, irregularities in the workpiece material, or other reasons. In the present work,
a rectangular aluminium block has been used for various slot milling operations (T-slot, square U-slot,
round U-slot) by a cutter of 16 mm diameter with a TiN-coated carbide insert on a CNC milling
machine. Linear and curvilinear accuracy of the slots has been judged by using various measuring
instruments. The size of burr formed has also been observed under a tool makers’ microscope. From
the observations, 1% to 4% linear deviation and less than 1% angular deviation is found. Presence of
some top, entrance, and side burr, etc. is found out and reported in the proposed paper.

Keywords: Aluminium alloy, Dimensional accuracy, Burr, CNC, End milling

1. Introduction
Milling process is frequently used to machine complex shapes like slots, pockets, precision moulds and
dies, etc. for aerospace industry, automobile industry and other areas where high quality surface is the
major area of concern. Discrete part manufacturing using CNC milling machine is gaining popularity
in modern manufacturing industries because it can provide high quality products at a lower
manufacturing cost. But some inaccuracies like rough surface, dimensional inaccuracy, burr, etc. are
often detected in CNC milling operation. These inaccuracies may come from improper selection of
process parameters, vibration during machining, gradual wearing out of the insert, lack of rigidity of
machine tools, irregularities in the workpiece material, etc. Usually dimensional accuracies are
checked through coordinate measuring machine (CMM) as a post process inspection. But recently it
104
 RTMST
has been proposed to use a machine-mounted touch probe which can provide part size information
directly to the CNC controller during machining thus forming a closed-loop process control to prevent
deviations in machining [1].

Fallbohmer et al. [2] reported that adjusting spindle RPM and feed rate to maintain nearly constant
chip load in the machining can optimize CNC programs. Ning et al. [3] pointed out that if the chip type
and its removal were carefully controlled in the high speed end milling, surface integrity, product
dimension, and tool life could be significantly improved. Diniz et al. [4] investigated the process
conditions for high speed turning on steel using coated carbide inserts without using cutting fluid. It
was reported that proper increase in feed and tool nose radius, and decrease in cutting speed could
improve overall turning performance without using coolant. Yih-fong et al. [5] tried to optimize high
speed CNC milling process by Taguchi based S/N ratio concept and performing ANOVA analysis.
Cutting speed was found to be most important parameter affecting the performance. Benardos et al.
[6] developed an artificial neural network (ANN) model based on Taguchi’s design of experiments to
predict surface roughness depending on process parameters. Law et al. [7] prepared a tool deflection
model to predict the deflection errors in straight segments and corners. Values of the predicted result
closely matched with the actual measurement by CMM.

Tsai et al [8] developed an in-process surface recognition system based on back-propagation neural
networks in end milling operations. Spindle speed, feed rate, depth of cut and the vibration average per
revolution were taken as four input neurons. The experimental results confirmed that the proposed
model could provide a high accuracy rate (96-99%). Raja et al. [9] made a mathematical model using
Particle Swarm Optimization (PSO) technique for predicting the effect of machining parameters such
as cutting speed, feed rate and depth of cut on the surface roughness of face milled surface. It was
reported that feed rate had greater influence on surface roughness when compared to cutting speed and
depth of cut. Since PSO could be able to obtain a global solution within a reasonable execution time, it
could be used on on-line systems for the selection of optimal cutting parameters. Lu et al. [10] used
gray relational analysis coupled with principal component analysis to optimize the cutting parameters
to obtain desired surface finish and metal removal rate in high speed end milling operation.
Thangarasu et al. [11] optimized the high speed CNC milling operation of aluminium using Taguchi
based RSM technique. Box-Behnken L27 orthogonal array was used as orthogonal design.

In the present work, dimensional accuracy of T and U slots machined by CNC milling operation, has
been judged and formation of burr in different parts of the slots has been observed.

2. Experimental Procedure

In the present work, an aluminium block of dimension 157 mm x 107 mm x 60 mm has been used as
base metal. Akshara VF 30 vertical 3-axis CNC milling machine (Fig. 1) has been used for slot milling
operation. This machine provides best performance at room temperature (10-45°C) and offers several
cutting speeds. A 16 mm diameter cutter (Sandvik A1B14-4032070) with one TiN-coated carbide
insert (Savdvik R390-11T308M-PM/4230) has been used. The workpiece has been set in a tapered
fashion to examine the variability of the depth of slot machined.

105
 RTMST

Fig. 1. The CNC milling machine used

First, one T slot, one round U slot and one square U slot (Fig. 2) have been machined on the
aluminium block by generating CNC part programs. It has been already said that for an efficient CNC
operation, proper selection of process parameters i.e., cutting speed, feed rate and depth of cut is
necessary [12]. The cutting parameters used in slot milling operations have been mentioned in the
Table 1.

Fig. 2. View of T, square U and round U slots

Table 1. Selected process parameters

Type of slot Selected process parameters

Depth of cut Feed (mm/rev) Speed (RPM) Cutting velocity (m/min)
T-slot Maximum 2 mm
Round U-slot and the reduced 0.05 2000 100.5
Square U-slot gradually

After the slot milling operation, obtained dimensions like depth of slot and width of slot have been
measured using digital depth gauge, and vernier caliper respectively. A tool makers’ microscope
(Model No.: 176-812E, XY range: 100 mm x 50 mm, magnification: 30x) has been used to measure
the corner angle and burr. A burr is the projected portion of the ductile workpiece, and is attached
beyond an edge of it. This needs be removed in the finished parts. The height of burrs has been
observed under a tool makers’ microscope, and qualitatively classified following Table 2.

Table 2. Classification of burr height

Sl. No. Height of burr Classification of burr

106
 RTMST
i Small but visible with bare eye ****
ii Tiny but visible with bare eye ***
iii Considerable but not visible with bare eye **
iv Negligible burr *

3. Results and Discussions

In the present work, job accuracy is tried to assess that can be achieved in slot milling operation while
making different types of job profiles. After machining the slots, percentage deviation of actual results
from desired depth of cut, slot width and corner angles have been measured by various measuring
instruments. Percentage deviation is calculated as (measured value – desired value) / (desired value).
The amount of burr formed during machining is also noted.

3.1 Measurement of T-Slot

A schematic diagram of the T–slot is shown in Fig. 3. In Table 3, measurements of depth of slot are
shown. It is observed that % deviation of measured error is within +3.50% which is within the
accepted limit of ±5%. As the metal is kept in taper position in its length, the desired depth of slot has
been adjusted throughout its length. In Table 4, measurements of slot width are shown in which
maximum deviation of the slot width is found out to be -2.38%.

Fig. 3. Schematic diagram of the T-slot

Table 3. Measurement of depth of slot in the T-slot

No. of Position Distance from the Desired value Measured value Percentage
observations end (A-A′) and of depth of slot of depth of slot deviation
(D′-C′) (mm) (mm) (mm) (%)
1 A-A′ 0 2 2.07 +3.50
2 B-B′ 23 2 2.06 +3.00
3 C-C′ 45 2 2.05 +2.50
4 D-D′ 61 2 2.04 +2.00
5 E-E′ 84 2 2.03 +1.50
6 F-F′ 107 2 2.06 +3.00
7 D′-C′ 0 2.00 2.07 +3.50

107
 RTMST
8 G-G′ 30 2.14 2.19 +2.33
9 H-H′ 60 2.27 2.33 +2.76
10 I-I′ 90 2.41 2.47 +2.48
11 J-J′ 120 2.54 2.61 +2.75

Table 4. Measurement of width of slot in the T-slot

No. of Position Distance from the Desired value Measured value Percentage
observations end (A-A′) and of depth of slot of depth of slot deviation
(D′-C′) (mm) (mm) (mm) (%)
1 A-A′ 0 16 15.62 - 2.38
2 B-B′ 23 16 15.64 - 2.25
3 C-C′ 45 16 15.68 - 2.00
4 D-D′ 61 16 15.64 - 2.25
5 E-E′ 84 16 15.66 - 2.13
6 F-F′ 107 16 15.64 - 2.25
7 D′-C′ 0 16 15.68 - 2.00
8 G-G′ 30 16 15.67 - 2.06
9 H-H′ 60 16 15.65 - 2.19
10 I-I′ 90 16 15.63 - 2.31
11 J-J′ 120 16 15.64 -2.25
In Table 5, measurements of right angles are shown. Quite low % deviation is seen in producing the
angular portions. In Table 6, burr heights measured at different positions are shown. During the cutter
entry and exit, entry and exit burrs are seen respectively. In rest of the portions, top burrs are observed.
As the edges cannot provide much support, burrs are formed there due to plastic deformation occurring
along the negative shear plane [13,14].

Table 5. Measurement of right angles in the T-slot

No. of Position of Desired Measured value of the angles Percentage

observations angles value of the Main scale Vernier scale Total deviation (%)
angles readings readings reading
1 B′C′G′ 90° 90° 24 90.24° + 0.27
2 E′D′G 90° 90° 18 90.18° + 0.20

Table 6. Measurement of burr height along the T-slot

No. of Range of Position Height of burr Type of burr

observations Left hand side Right hand side
(from entrance) (from entrance)
1 A-A′ *** **** Entrance burr
2 A-A′ to B-B′ *** ** Top burr
3 B-B′ to C-C′ ** * Top burr
4 C-C′ to D-D′ ** *** Top burr
5 D-D′ to E-E′ * ** Top burr
6 E-E′ to F-F′ ** * Top burr
7 F-F′ *** *** Exit burr
108
 RTMST
8 D′-C′ *** **** Entrance burr
9 D′-C′ to G-G′ ** *** Top burr
10 G-G′ to H-H′ * ** Top burr
11 H-H′ to I-I′ ** ** Top burr
12 I-I′ to J-J′ *** ** Top burr
13 J-J′ *** *** Exit burr

3.2 Measurement of Square U-Slots

A schematic diagram of the square U-slot is shown in Fig. 4. In Table 7, measurements of depth of slot of the
square U-slot are shown. The percentage deviation of the slot depth is within the acceptable limit of ±5%
(+1.64% to -3.37%).

Fig. 4. Schematic diagram of the square U-slot

In Table 8, measurements of slot width of the square U-slot are shown. Here also, maximum % deviation of
slot width is found out to be -2.88% which is acceptable. In Table 9, measurements of corner right angles are
given showing only 0.20 to 0.27% deviation in the measured value.

In Table 10, burr heights and their position and type at different locations of the slot are shown. Types of burrs
observed are entrance and exit burrs at the cutter entry and exit portions respectively, and elsewhere, top
burrs are there as expected.

Table 7. Measurement of depth of slot in the square U-slot

No. of Position Desired value of Measured value of Percentage

observations depth of slot (mm) depth of slot (mm) deviation (%)
1 A-A′ 2.22 2.17 - 2.25
2 B-B′ 2.22 2.15 - 3.15
3 C-C′ 2.22 2.18 - 1.80
4 D-C′ 2.17 2.14 - 1.38
5 E-E′ 2.00 2.03 + 1.05
6 F-F′ 1.83 1.86 + 1.64
109
 RTMST
7 G-F′ 1.78 1.74 - 2.24
8 H-H′ 1.78 1.74 - 2.24
9 I-I′ 1.78 1.72 - 3.37

Table 8. Measurement of width in the square U-slot

No. of Position Desired value of Measured value of Percentage

observations slot width (mm) slot width (mm) deviation (%)
1 A-A′ 16 15.59 - 2.56
2 B-B′ 16 15.57 - 2.69
3 C-C′ 16 15.55 - 2.81
4 D-C′ 16 15.57 - 2.69
5 E-E′ 16 15.55 - 2.81
6 F-F′ 16 15.55 - 2.81
7 G-F′ 16 15.57 - 2.69
8 H-H′ 16 15.55 - 2.81
9 I-I′ 16 15.54 - 2.88

Table 9. Measurement of right angles in the square U-slot

No. of Position Desired Measured value of corner angle Percentage

observations of value of Main scale Vernier scale Total deviation
angles corner angle readings readings reading (%)
1 B′C′E′ 90º 90º 24 90.24º + 0.27
2 E′F′H′ 90º 90º 18 90.18º + 0.20

Table 10. Measurement of burr along the square U-slot

No. of Range of Height of burr Type of burr

observations position Left hand side Right hand side
(from entrance) (from entrance)
1 A-A′ **** *** Entrance burr
2 A-A′ to B-B′ *** *** Top burr
3 B-B′ to C-C′ ** ** Top burr
4 C-C′ to D-C′ *** **** Top burr
5 D-C′ to E-E′ * ** Top burr
6 E-E′ to F-F′ *** ** Top burr
7 F-F′ to G-F′ ** ** Top burr
8 G-F′ to H-H′ ** * Top burr
9 H-H′ to I-I′ ** ** Top burr
10 I-I′ **** **** Exit burr

110
 RTMST

3.3 Measurements of Round U-Slots

Fig. 5 is a schematic diagram of the round U–slot. In Table 11, measured values and percentage deviations of
depth of slot of the square U-slot are shown. Deviations in the depth of slot are quite less (within -1.00% to
+2.50%), and these are within accepted limits. Measured values of slot width are tabulated in Table 12, in
which maximum deviation of -2.81% is observed, when ±5% deviation is considered to be acceptable. In Table
13, height of burr, their position and type at different locations of the slot are shown. In this case also, similar
types of burr are observed expectedly as that seen in square U-slot.

Fig. 5. Round U- slot

Table 11. Measurement of depth of slot of round U- slot

No. of Position Desired Measured Percentage

observations value (mm) value (mm) deviation (%)
1 A-A′ 2 1.98 - 1.00
2 B-B′ 2 2.03 + 1.50
3 C-C′ 2 2.05 + 2.50
4 D-D′ 2 2.01 + 0.50
5 E-E′ 2 1.98 - 1.00
Table 12. Measurement of slot width of round U- slot

No. of Position Theoretical value Measured values of Percentage

observations of width (mm) slot width (mm) deviation (%)
1 A-A′ 16 15.58 - 2.63
2 B-B′ 16 15.57 - 2.68
3 C-C′ 16 15.58 - 2.63
4 D-D′ 16 15.56 - 2.75
5 E-E′ 16 15.55 - 2.81

111
 RTMST
Table 13. Measurement of burr along round U- slot

No. of Position Height of burr Type of burr

observations Left hand side Right hand side
(outer side) (inner side)
1 A-A′ *** *** Entrance burr
2 A-A′ to B-B′ ** ** Top burr
3 B-B′ to C-C′ * * Top burr
4 C-C′ to D-D′ * * Top burr
5 D-D′ to E-E′ ** ** Top burr
6 E-E′ ** *** Exit burr

4. Conclusions

From the above experimental results on different slot milling operations done on a CNC milling
machine, following conclusions can be made.

i) For all the three slots, deviation obtained in depth and width of slot is within the accepted range
of ± 5%.
ii) In all the cases, deviation obtained for corner angles are quite low (< 0.30%).
iii) Burrs at the entrance and exit are large enough. In other positions, top burrs are found.
iv) Proper selection of cutting velocity, feed and depth of cut through analysis of results of
experiments by varying cutting conditions is necessary, and this can be the future scope of
work. Use of coolants to get better surface finish may also be explored.

Acknowledgement

Authors sincerely acknowledge the contribution of Sri Ashis Kr. Bhattacharya of the Manufacturing
Technology Laboratory of Kalyani Govt. Engineering College, Kalyani throughout various stages of
experimentation.

References
[1]. Y. Kwon, T.L. Ttseng, Y. Ertekin, Characterization of closed-loop measurement accuracy in precision
CNC milling, Robotics and Computer-Integrated Manufacturing, Vol. 22, 2006, pp. 288-296.
[2]. P. Fallbohmer, C.A. Rodriguez, T. Ozel, T. Altan, High-speed machining of cast iron and alloy steels
for die and mold manufacturing, Journal of Materials Processing Technology, Vol. 98, 2000, pp. 104–
115.
[3]. Y. Ning, M. Rahman, Y.S. Wong, Investigation of chip formation in high speed end milling, Journal of
Materials Processing Technology, Vol.113, 2001, pp. 360–367.
[4]. A.E. Diniz, R.Micaroni, Cutting condition for finish turning process aiming: the use of dry cutting,
International Journal of Machine Tools & Manufacture, Vol. 42, 2002, pp. 899–904.
[5]. T. Yih-fong, J. Ming-der, Dimensional quality optimisation of high-speed CNC milling process with
dynamic quality characteristic, Robotics and Computer-Integrated Manufacturing, Vol. 21, 2005, pp.
506-517.
[6]. P.G. Benardos, G.C. Vosnikos, Prediction of surface roughness in CNC face milling using neural
networks and Taguchi design of experiments, Robotics and Computer-Integrated Manufacturing, Vol.
18, 2002, pp. 343-354.
112
 RTMST
[7]. K.M.Y. Law, A. Geddam, Prediction of contour accuracy in the end milling of pockets, Journal of
Materials Processing Technology, Vol. 113, 2001, pp. 399-405.
[8]. Y-H. Tsai, J.C. Chen, S-J. Lou, An in-process surface recognition system based on neural networks in
end milling cutting operations, International Journal of Machine Tools & Manufacture, Vol. 39, 1999,
pp. 583-605.
[9]. S.B. Raja, N. Baskar, Application of particle swarm optimization technique for achieving desired milled
surface roughness in minimum machining time, Expert Systems with Applications, Vol. 39, 2012, pp.
5982-5989.
[10]. H.S. Lu, C.K. Chang, N.C. Hwang, C.T. Chung, Grey relational analysis coupled with principal
component analysis for optimization design of the cutting parameters in high-speed end milling,
Journal of Material processing Technology, Vol. 209, 2009, pp.3808-3817.
[11]. V.S. Thangarasu, R. Sivasubramanian, Study of high speed CNC milling of aluminium: optimization of
parameters using Taguchi based RSM, European Journal of Scientific Research, Vol. 74, No.3, 2012,
pp. 350-363.
[12]. J.S. Narang, Computer Aided Manufacturing (CNC & Robotics), Dhanpat Rai & Co. (P) Ltd. Educational &
Technical Publishers, Revised Edition, 2006, pp. 45-95.
[13]. P.P. Saha, S. Das, Burr minimization in face milling: An edge beveling approach, Proceedings of the
Institutions of Mechanical Engineers, Part B, Journal of Engineering Manufacture, Vol. 225, No. 9,
2011, pp. 1528-1534.
[14]. P.P. Saha, A. Das, S. Das, On reduction of formation of burr in face milling of 45C8 steels, Materials and
Manufacturing Processes, Accepted for publication, 2013.

113
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Determining the effect of grinding parameters on surface

roughness in traverse cut cylindrical grinding process using
Taguchi method

Ramesh RUDRAPATIa Pradip Kumar PALb Asish BANDYOPADHYAYc

Ph.D. Scholara, Professorb, c
Mechanical Engineering Department, Jadavpur University, Kolkata, West Bengal – 700032, India
E-mails: (a). rameshrudrapati@gmail.com (b). pradippal54@yahoo.com (c).
asishbanerjee@yahoo.com

________________________________________________________________________
Abstract
This paper presents an effective approach to optimize grinding parameters for traverse cut cylindrical
grinding process while grinding of mild steel. Cylindrical grinding is extensively used in the metal
cutting industries to make cylindrical parts with good surface finish. Accurate and intricate grinding
are the strengths, but, grinding parameter selection and surface quality still remains as major
challenges. The main objective of this study is to minimize surface roughness parameters Ra and Rz
simultaneously. Infeed, longitudinal feed and work speed are the three control factors taken at various
levels. The experiments are designed based on L27 orthogonal array of Taguchi method. Grey
relational analysis has been employed for convert multi-objective problem into single objective
problem. The signal to noise ratio (S/N) and analysis of variance (ANOVA) have been used to identify
significant grinding parameters on surface roughness. Optimum grinding condition has been obtained
from main effect plots. Confirmatory experiment has also been conducted to validate the optimized
grinding condition.
Keywords: Cylindrical grinding, Surface roughness, Grey relational analysis, Taguchi method,
Analysis of variance.

I. Introduction

Cylindrical grinding is the mostly used manufacturing process in any metal cutting industries to
produce cylindrical surfaces with very high level of accuracy and surface finish. The study of metal
cutting focuses, among others, machine parameter settings influencing process efficiency and output
114
 RTMST
quality characteristics. A significant improvement in process efficiency may be obtained by process
parameter optimization that identifies and determines the regions of critical process control factors
leading to desired output responses with acceptable variations ensuring a lower cost of manufacturing
[1]. Optimization in cylindrical grinding is bit complex, because of lot of interactive parameter such as
wheel parameters: abrasives, grain size, grade, structure, binder, shape and dimension etc.; work-piece
parameter: mechanical and chemical properties; process parameter: infeed (depth of cut), longitudinal
feed, work speed, grinding wheel speed, dressing condition etc.; and machine parameters: static and
dynamic characteristics, spindle system, table system [2]; are acting while grinding. But, through
systematic optimization methodology based on design of experiments (DOE), it can possible to obtain
better surface finish by controlling grinding process.
Surface finish is treated as technological quality of any machined surface, and it plays key role in its
performance. The arithmetic average height parameter (Ra) and ten-point height roughness (Rz) are
important roughness parameters used to specify the quality of the machined job. In which Ra is most
universally accepted and used roughness parameter to describe the quality of surface finish of the
machined part. Review of available literature shows that Ra has been focused for most of the studies,
but, Ra parameter alone cannot provide full description of surface finish. If the number of roughness
parameters used is increased, full description of the surface profile can be obtained [3]. The present
study aims at consideration of two roughness parameters Ra and Rz combinedly for the surface
roughness generated in cylindrical grinding operation. With consideration of two objectives
simultaneously, this problem becomes multi-objective problem. Grey relational analysis technique
which is used in the present case to converting multi-objective problem in to a single objective
problem. It has also been attempted to optimize surface roughness prediction using signal-to-noise
ratio and main effect plots.

2. Literature Review

Since grinding is the last operation in most of the production processes in the metal cutting industry,
surface finish of the ground components have greater influence on the quality of the product. As
already mentioned above, surface finish in cylindrical grinding has been found to be influenced in
varying amounts by a number of factors. Literature is fairly extensive in regard to optimization of
process parameters in different machining operations for single objective optimization by Taguchi
method and multi objective optimization by using grey relation analysis combined with Taguchi
method. Some of the previous studies are mentioned here.
The Taguchi method is a useful technique for the design of high quality systems. It provides simple,
efficient and systematic approach to optimize designs for improved performance, quality and reduced
cost [4]. Zhang et al [5] used application of Taguchi method to optimize surface quality in CNC face
milling operation. Taguchi method cum grey relational analysis was used in turning operation to study
the significant control parameters influencing roughness average, roughness maximum and roundness
individually and to optimize them simultaneously [6]. IIhan and Akkus [7] used Taguchi method to
minimize surface roughness (Ra and Rz). The statistical methods of signal to noise ratio and analysis of
variance were applied to investigate effects of process parameters on surface roughness in CNC
turning. The optimum cutting parameters for minimum roughness was determined by using main effect
plots. Lin and Lin [8] conducted orthogonal array experiment, and grey-Taguchi method was used for
prediction of multiple characteristics: material removal rate, surface roughness and electrode wear
ratio. Caydas and Hascalik [9] optimized two cutting parameters in laser with consideration of
multiple-performance characteristics: surface roughness, top kerf width and width of heat affected

115
 RTMST
zone. However, application of this technique (grey-Taguchi method) in optimization of cylindrical
grinding process is found to be limited so far, in the literature.

3. Experimental Planning Methods

3.1. Design of Experiments (DOE)

DOE refers to the planning, designing, and analyzing the experiment so that valid and objective
information can be drawn effectively and with greater efficiency. DOE is based on the objective of
desensitizing a product’s performance characteristics to variation in critical product and process design
variables [10]. Among other methods, Taguchi method is an important experimental design method of
DOE. Orthogonal array (OA) in the Taguchi method is standard tables which used to design the
experiments. In the present study, an L27 OA (Table 1) is chosen so that the individual effects of the
chosen grinding variables on responses Ra and Rz can be studied.

Table 1. L27 Orthogonal array (OA).

Input parameter
S. No Infeed Longitudinal Work speed
(A) feed (B) (C)
1 1 1 1
2 1 1 1
3 1 1 1
4 1 2 2
5 1 2 2
6 1 2 2
7 1 3 3
8 1 3 3
9 1 3 3
10 2 1 2
11 2 1 2
12 2 1 2
13 2 2 3
14 2 2 3
15 2 2 3
16 2 3 1
17 2 3 1
18 2 3 1
19 3 1 3
20 3 1 3
21 3 1 3
22 3 2 1
23 3 2 1
24 3 2 1
25 3 3 2
26 3 3 2
27 3 3 2

116
 RTMST

3.2 Grey Relational Analysis

Optimization of multiple performance characteristics is not straight forward and much more
complicated than of single performance characteristics. To solve the multi-responses (Ra and Rz) in the
present work, grey relational analysis has been used to convert multiple optimization problems into
single objective optimization problem. Grey relational analysis is then coupled with Taguchi method to
identify the significant grinding variables on responses and optimize them (grinding variables). Grey
theory provides the following steps to optimize the grinding parameters.
1. Normalization of surface roughness parameters Ra and Rz.
2. Calculation grey relational coefficient.
3. Calculation of grey relational grade by averaging the grey relational coefficient.
4. Analyze the experimental results using the grey relational grade, statistical analysis of variance and
signal noise ratio.
5. Selection of optimal levels of process parameters.
6. Verification of the optimal levels by conducting confirmatory test.

3.2.1. Data Pre-Processing

In grey relational analysis, the experimental results are first normalized in the range between zero and
unity. This process of normalization is called the “grey relational generation”. The next step is to
calculate the grey relational coefficients from the normalized experimental data to express the
relationship between the desired and actual experimental data. Then the overall grey relational grade is
calculated by averaging the grey relational coefficient corresponding to each selected response. The
overall evolution of the multiple responses is based on the grey relational grade. Grey relational
analysis converts a multi objective problem into a single objective problem with the objective function
of overall grey relational grade. The corresponding level of parametric combination with highest grey
relational grade is considered as the optimum parametric combination [8].
In grey relational analysis, the normalized data corresponding to lower-the-better criterion can be
expressed as follows:

max yi(k )  yi (k )
xi(k )  (1)
max yi(k )  min yi (k )

where xi (k) is the value after grey relational generation while min yi(k) and max yi(k) are
respectively the smallest and largest values of yi(k) for the kth response. The grey relational co-
efficient can be calculated as

 min   max
 i( k ) 
 0i (k )  r  max
(2)
where Δ0i = ||x0(k) - xi(k)|| = difference of the absolute value between x0(k) and xi(k); ∆min =
smallest value of  0i ; ∆max = largest value of  0i ; and here ‘r’ is distinguish co- efficient which is
used to adjust the difference of the relational coefficient, usually r is within the set {0, 1}, generally ‘r’
value is taken as 0.5. i(k) is Grey relation co-efficient. After obtaining the grey relational coefficient,
grey relational grade is determined by taking the average of the grey relational coefficient. The grey
relational grade can be expressed as follows:

117
 RTMST
n
i    i(k ) / n
k 1
(3)

where ‘n’ is the number of performance characteristics.

Signal-to-noise ratio (S/N) of Taguchi method is used in the present study to identify significant input
parameters on output responses from grey relational grade. Based on the criteria of the experiment,
S/N ratio can be categorized into: lower-the-better (LB); Higher-the-better (HB); and Nominal-the-
better (NB) [11]. In the present study of maximization of grey relational grade, HB criterion needs to
be used.

Lower-the-better: S / N  10log1/ n   y 2 
(4)
Higher-the-better: S / N  10log1/ n  1/ y 2 
(5)
Nominal-the-better: S / N  10 log y / S y2 (1)
(6)

where, y is the average of observed data, s y2 is the variation of ‘y’, n is the number of observations, and
‘y’ is the observed data.

3. Experimental Procedure

L27 orthogonal array is used for conducting the experiments on mild steel work piece in traverse cut
cylindrical grinding machine. Three varied levels of the process variables infeed, longitudinal feed and
work speed are selected in the present investigation. The selected input parameters and their levels are
shown in Table 2. The experimental set up is shown in Fig. 1. After completing the experiments,
surface roughness has been measured using stylus-type Profilometer: Talysurf (Taylor Hobson,
Sutronic 3+). The surface roughness measurement set up is shown in Fig. 2.

Table 2. Input parameters and their levels

Levels
Input parameters
1 2 3
Infeed (A) mm/cycle 0.04 0.05 0.06
Longitudinal feed (B) mm/s 70 80 90
Work speed (C) 80 112 160

118
 RTMST

Fig. 1. Experimental set up

Fig. 2. Surface roughness measurement set up

4. Results and Discussion

The experimental results are shown in in Table 3. These data have been used to identify the effects of
process parameters on surface roughness parameters Ra and Rz. In order to convert multi-objective
optimization problem of Ra and Rz into single objective optimization problem, the Taguchi design of
experiment is integrated with the grey relational analysis. In grey relational analysis, measured
experimental data i.e. surface roughness parameters Ra and Rz have been normalized by using Eq. 1
and shown in Table 4. This process is known as known as grey relational generation as already
mentioned above. Grey relational coefficients for both responses have been calculated based on
normalized experimental data by using Eq. 2 (Table 4). Then the grey relational grade corresponding
to selected responses has been determined by averaging the grey relational coefficients with use of Eq.
3 (Table 4). Furthermore, analysis of signal to noise ratio of Taguchi method has been applied on grey
119
 RTMST
relational grade to identify the significant process parameters, and to predict the optimum condition for
minimizing both surface roughness parameters Ra and Rz combinedly.
Table 3. Output responses

Experiments Output responses

Input parameters surface roughness
S. No
Infeed (A) Longitudinal feed (B) Work speed
Ra μm Rz μm
mm/cycle mm/s (C) rpm

1 1 1 1 0.663 4.76
2 1 1 1 0.558 4.24
3 1 1 1 0.576 4.15
4 1 2 2 0.637 4.54
5 1 2 2 0.525 4.38
6 1 2 2 0.533 4.37
7 1 3 3 0.641 4.48
8 1 3 3 0.637 4.43
9 1 3 3 0.620 4.63
10 2 1 2 0.604 4.64
11 2 1 2 0.673 4.68
12 2 1 2 0.563 3.70
13 2 2 3 0.806 4.97
14 2 2 3 0.776 5.73
15 2 2 3 0.812 5.56
16 2 3 1 0.761 5.59
17 2 3 1 0.797 4.35
18 2 3 1 0.802 4.40
19 3 1 3 0.602 4.05
20 3 1 3 0.614 4.20
21 3 1 3 0.644 4.45
22 3 2 1 0.738 5.26
23 3 2 1 0.708 5.03
24 3 2 1 0.753 5.02
25 3 3 2 0.621 4.36
26 3 3 2 0.766 5.17
27 3 3 2 0.680 5.24

Table 4. Grey relational analysis for surface roughness parameters Ra and Rz

Grey relational Grey

Normalized data Values of ∆0i
S. No coefficient relational
Ra Rz Ra Rz Ra Rz Grade
1 0.519163763 0.477832512 0.480836237 0.522167488 0.5097690 0.48915662 0.49946286
2 0.885017422 0.733990148 0.114982578 0.266009852 94
0.8130311 7
0.65273311 0.73288214
3 0.822299652 0.778325123 0.177700348 0.221674877 61
0.7377892 90.69283276 0.71531098
03 5 4
120
 RTMST
4 0.609756098 0.586206897 0.390243902 0.413793103 0.5616438 0.54716981 0.55440682
5 1 0.665024631 0 0.334975369 136 1
0.59882005 3
0.79941002
6 0.972125436 0.669950739 0.027874564 0.330049261 0.9471947 90.60237388 90.77478430
7 0.595818815 0.615763547 0.404181185 0.384236453 19
0.5529865 70.56545961 30.55922306
8 0.609756098 0.640394089 0.390243902 0.359605911 13
0.5616438 0.58166189 1
0.57165286
9 0.668989547 0.541871921 0.331010453 0.458128079 36
0.6016771 10.5218509 30.56176402
10 0.724738676 0.536945813 0.275261324 0.463054187 49
0.6449438 0.51918158 4
0.58206270
11 0.484320557 0.517241379 0.515679443 0.482758621 20.4922813 60.50877193 30.50052661
12 0.867595819 1 0.132404181 0 04
0.7906336 1 70.89531680
13 0.020905923 0.374384236 0.979094077 0.625615764 09
0.3380447 0.44420131 4
0.39112303
14 0.12543554 0 0.87456446 1 59
0.3637515 30.33333333 60.34854245
15 0 0.083743842 1 0.916256158 84
0.3333333 3
0.35304347 9
0.34318840
16 0.177700348 0.068965517 0.822299652 0.931034483 33
0.3781291 80.34939759 60.36376335
17 0.052264808 0.679802956 0.947735192 0.320197044 17
0.3453670 0.60960961 40.47748831
18 0.034843206 0.655172414 0.965156794 0.344827586 28
0.3412604 0.59183673 9
0.46654856
19 0.731707317 0.827586207 0.268292683 0.172413793 04
0.6507936 50.74358974 90.69719169
20 0.68989547 0.753694581 0.31010453 0.246305419 51
0.6172043 4
0.66996699 7
0.64358564
21 0.585365854 0.630541872 0.414634146 0.369458128 01
0.5466666 70.57507082 90.56086874
22 0.257839721 0.231527094 0.742160279 0.768472906 67
0.4025245 20.39417475 40.39834965
23 0.362369338 0.344827586 0.637630662 0.655172414 44
0.4395099 7
0.43283582 1
0.43617288
24 0.205574913 0.349753695 0.794425087 0.650246305 54
0.3862718 10.43468950 70.41048068
25 0.665505226 0.674876847 0.334494774 0.325123153 71
0.5991649 7
0.60597014 9
0.60256753
26 0.160278746 0.275862069 0.839721254 0.724137931 27
0.3732119 90.40845070 80.39083133
27 0.459930314 0.24137931 0.540069686 0.75862069 64
0.4807370 40.39726027 40.43899864
18 4 6

4.1. Analysis of Variance (ANOVA)

Table 6 and 7 shows the analysis of variance (ANOVA) and response table for signal-to-noise ratio
(S/N) ratio of grey relational grade. In ANOVA, P- value is determined which factor is significance. If
P-value of a particular factor becomes less than 0.05; then it is concluded that influence of the
respective factor on output response (95% confidence level).is significant. Table 6 reveals that all the
process parameters do not have much influence on grey relational grade i.e. surface roughness
parameters Ra and Rz as its ‘P’ values are greater than 0.05.
The delta statistics in Table 7 measures the size of the effect by taking the difference between the
highest average and lowest average value for each response, process parameters are ranked from the
highest to lowest effect. It is conclude from 7 that infeed is the most significant variable on both
surface roughness parameter Ra and Rz, next is longitudinal feed followed by work speed.

Table 6. Analysis of signal to noise ratio (AS/N)

Source DF Seq SS Adj SS Adj F P

MS
A 2 12.048 12.048 6.024 2.79 0.264
B 2 10.875 10.875 5.438 2.52 0.284
C 2 3.9000 3.9000 1.950 0.90 0.525

121
 RTMST
Residual error 2 4.317 4.317 2.158
Total 8 31.140

Table 7. Response table for s/n ratio

Level A B C
1 -4.161 -4.197 -6.411
2 -6.890 -6.633 -4.850
3 -6.188 -6.408 -5.978
Delta 2.729 2.436 1.562
Rank 1 2 3

A B
-4
Mean of S/N ratios for grey relational grade

-5

-6

-7
1 2 3 1 2 3
C
-4

-5

-6

-7
1 2 3
Signal-to-noise: Larger is better

Fig. 3. Main effect plots

4.2. Prediction of Optimal Process Condition
Main effect plots of input parameters on mean of grey relational grade is drawn from response table
for S/N ratios (Table 7) and shown in Fig. 3. From this plot, Optimum grinding parametric
combination (level 1 of infeed (0.04 mm/cycle), level 1 of longitudinal feed (70 mm/s) and level 2 of
work speed (112 rpm)) is obtained for minimizing surface roughness parameters Ra and Rz
simultaneously. Obtained optimum grinding condition is shown in Table 8.

5. Confirmatory Experiment
After obtaining the optimal parametric setting, the improvement of output responses (Ra and Rz) has
been verified. The confirmatory experiment (Table 8) confirms that surface roughness parameters Ra
and Rz have been minimized / improved compared to initial experimental results (Table 2).
Table 8. Optimum conditions obtained by Taguchi method

122
 RTMST
Optimum grinding condition obtained by Obtained output responses (Ra and Rz) at
Taguchi method optimum grinding condition
Input parameters Level Ra Rz
Infeed (A) mm/cycle 1
Longitudinal feed (B) mm/s 1 0.520 μm 3.65 μm
Work speed (C) rpm 2

6. Conclusions
In the present study, integrated grey relational analysis cum Taguchi methodology have been
employed for optimizing grinding variables to minimize surface roughness parameters Ra and Rz
combinedly in traverse cut cylindrical grinding. The following important points are drawn from this
work:
1. From analysis of variance (ANOVA), it is noted that all input parameters do not have direct effect
on surface roughness (Ra and Rz).
2. Response table for signal to noise ratio reveals that infeed is the highest significant factor on
surface roughness next is longitudinal feed and followed by work speed.
3. Optimum grinding condition: infeed = 0.04 mm/cycle, longitudinal feed = 70 mm/s and work
speed = 112 rpm is obtained from main effect plots of mean signal-to-noise ratio.
4. The confirmatory test confirms that proposed technique (grey relational analysis - Taguchi
method) is very effective in optimization of surface roughness parameters Ra and Rz
simultaneously in traverse cut cylindrical grinding.

7. Acknowledgement

Research support provided by Council of Scientific and Industrial Research (CSIR), India: File No:
9/96 (0723)2K12-EMR-I dated 27/02/1012 to Ramesh Rudrapati (one of the authors) is gratefully
acknowledged.

8. References
[1] Mukherjee and p. K. Ray. A review of optimization techniques in metal cutting processes, Computers &
Industrial engineering, Vol. 50, 2006, pp. 15-34
[2] J. S. Kwak. Application of Taguchi and response surface methodologies for geometric error in surface
grinding process, International Journal of Machine Tools & Manufacturing, Vol. 45, 2005, pp. 327-
334.
[3] E.S. Gadelmawla, M.M. Koura, T.M.A. Maksoud, I.M. Elewa and H.H.Soliman. Roughness
parameters, Journal of Material Technology, Vol. 123, 2002, pp.133-145.

[4] W.H. Yang and Y.S. Tang. Design of cutting parameters for turning operations based on Taguchi
method, Journal of Material processing Technology, Vol. 84, 1998, pp. 122-129.
[5] J. Z. Zhang, J. C. Chen and E. D. Kirby. Surface roughness optimization in an end-milling operation
using the Taguhi design method, Journal of Material processing Technology, Vol. 184, 2007, pp. 233-
239.
[6] b. C. Tzeng, Y. H. Lin, Y. K. Yanga and M.C. Jeng. Optimization of turning operations with multiple
performance characteristics using the Taguchi method and Grey relational analysis, Journal of
Materials Processing Technology, Vol. 2 0 9, 2 0 0 9, pp. 2753–2759
[7] IIhan Asilturk and Harun Akkus. Determining the effect of cutting parameters on surface roughness in
hard turning using the Taguchi method, Measurement, Vol. 44, 2011, pp. 1697-1704.
123
 RTMST
[8] J.L. Lin and C.L. Lin. The use of the orthogonal array with grey relational analysis to optimize the
electrical discharge machining process with multiple performance characteristics, International Journal
of machine Tools &Manufacture, Vol. 42, 2002, pp. 237-244
[9] U. Caydas and A. Hascalik. Use of the grey relational analysis to determine optimum laser cutting
parameters with multi-performance characteristics, Optics & Laser Technology, Vol. 40, 2008, pp. 987-
994.
[10] S.K Das and P.sahoo. Tribological characteristics of electroless Ni-B coating and optimization of
coating parameters using Taguchi grey relational analysis, Materials and Design, Vol. 32, 2011, pp.
2228-2238.
[11] S. Datta, A. Bandyopadhyay and P. K. Pal. Slag recycling in submerged arc welding and its influence
on weld quality leading to parametric optimization, International Journal of Advanced Manufacturing
Technology, Vol.39, 2008, pp. 229-238.

124
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Machining of Cylindrical Surface by Electric Discharge

Machine

SANASAM Sunderlal Singh*, P. K. PATOWARI

Department of Mechanical Engineering
National Institute of Technology Silchar
P.O. Silchar 788010, Assam, India
[e-mail: sunderlal2031@gmail.com, ppatowari@yahoo.com]
(*Corresponding author)

Abstract: Die-sinking Electrical Discharge Machining (EDM) is a non-conventional machining

process which is widely used to machine prismatic components having flat surfaces irrespective of its
hardness. In this work, an attachment has been designed and fabricated which is fitted to the die-
sinking EDM for an attempt to machine cylindrical surface. The effects of machining parameters such
as pulsed current, pulse on-time and workpiece rotation on Material Removal Rate (MRR) and Surface
Roughness (Ra) were analyzed. An MRR up to the extent of 9.7mg/min with surface roughness of 4.5
µm has been obtained at pulsed current of 16A, pulse on-time of 1010µs and rotation of the workpiece
at 8rpm.

Keywords: Electrical Discharge Machining, Material Removal Rate

1. Introduction
Electrical Discharge Machining is a non-conventional machining process in which the erosion of work
material takes place by a rapid succession of short time electrical discharges occurring between the
tool electrode and the workpiece immersed in dielectric. When a voltage pulse is applied to the
electrodes, an electric spark discharge occurs within the inter-electrode gap, i.e., spark gap leading to
localised heating of work surface and further melting and vaporization of work surface and hence
machining takes place. Various parameters in EDM are gap voltage, peak current, polarity, on-time,
off-time, flushing pressure and pulse frequency.

125
 RTMST
The hardness and strength of the difficult-to-machine work-materials are no longer the dominating
factors that affect the tool wear and hinder the machining process. This makes the EDM process
particularly suitable for machining hard, difficult-to-machine materials. Machining of cylindrical parts
has been already reported by Dr. Masuzawa’s research group at University of Tokyo using Wire EDM
[1].Use of rotary electrode [2], application of magnetic field [3], planetary motion of tool electrode
[4], force flushing, vibration of electrode in the two cases of low frequency (about 50 Hz) and high
frequency (more than 20 KHz) are some of the techniques available for preventing abnormal
discharges. The centrifugal force in rotary EDM improves the gap flushing and machining efficiency
using Cu tool electrode of radius 4mm and length of 25mm and AISI D2 tool steel as workpiece of
8mm diameter rotating at high speed upto a range of 5000rpm [5]. A new mechanism of cutting pipe
combined with electrical discharge machining had been introduced by Chen et al. [6] in which both
MRR and TWR increases.

Till date, there have been few investigations into rotary motion of tool electrode in EDM for
machining of cylindrical workpiece. In this work, an attempt has been made to machine cylindrical
workpiece by EDM. The concept of machining cylindrical workpiece is as illustrated in Fig.1.

Fig.1. Schematic diagram of cutting profile generated

In conventional EDM, the tool used for the spark erosion is an electrode whose shape is a negative
replica of the contour to be produced on the work while in this work; a circular profile is generated as
the workpiece is rotated relative to the tool.

2. Experimental setup

An electrical-discharge machine model F25 series of Sparkonix India private limited with servo-
control was used to conduct the experiments. Fig.2 shows the experimental set up of the EDM
processes, in which a rotation mechanism was developed and installed in the machining tank. A DC
motor rotates the workpiece at varying speed. EDM was applied to the central part of the specimen.
Copper is used as the electrode material and mild steel as the work material.

126
 RTMST

Fig. 2Experimental set-up

Experiments were conducted with different combinations of the varying parameters at different levels.
The experimental conditions are as listed in table1.

Table1. Experimental conditions

Dielectric used EDM oil

Work material Mild steel rod of 14mm dia
Electrode material Copper rod of 6mm dia
W/P rotation speed (rpm) 4,8,12
Pulse current (amps) 8,16,20
On time (µsec) 463, 1010
Duty factor (%) 58, 72
Flushing pressure (kg/cm2) 5
Polarity Normal
Machining time (mins) 20

127
 RTMST
3. Results and discussion

The workpiece specimen and the tool after EDM operation are shown in Fig. 3.

3(a) 3(b)
Fig. 3(a) Workpiece and 3(b) Tool after machining

3.1 Material removal rate

Difference in weight of the workpiece and the tool has been measured using Ishida DX precision
electronic balance and the Material Removal Rate (MRR) was calculated. Necessary plots showing the
relationship between the current and material removal rate at two different pulsed On-times has been
made as shown in Fig. 4. It has been observed that MRR increases with increase in both Current and
Pulsed On-time owing to the higher energy input rate.

MRR Vs Current
MRR (mg/min)

463 µsec
1010µsec

At 8rpm

Current (amps)

128
 RTMST
Fig.4Effect of Current on MRR with varying pulse On-time

3.2 Surface roughness

Surface roughness has been measured using Handysurf E-35A Portable Surface Measuring Unit.
Measurements are taken at three different positions and its average values have been calculated and a
graph is plotted for different readings at varying parameters. Fig. 5 shows the effect of current on
Surface roughness at varying pulsed On-time. It has been observed that higher current and higher
pulsed On-time results in rough surface finish.

Ra Vs Current
Ra (µm)

463 µsec
1010µsec

At 8rpm

Current (amps)

Fig.5 Effect of current on Surface roughness (Ra) at varying pulsed On-time

3.3 Effect of workpiece rotation

MRR Vs RPM
MRRT (mg/min)

463 µsec
1010 µsec

At 16A

RPM

Fig.6 Effect of workpiece rotation speed on MRR at varying pulsed On-time

129
 RTMST
Figure 6 shows the effect of workpiece rotation speed on MRR at varying pulsed On-time. It has been
observed that with increase in rotational speed of the workpiece, MRR initially increases and then
decreases at higher rpm which can be attributed to effective flushing. Finally unstable discharge at
higher rpm results in decrease of MRR at higher rpm.

3.4 Surface morphology

To study the machined top layer of the workpiece, Leica DM-500X Optical microscope has been used.
Micrographs of the machined surface are as shown in Fig.5 and Fig.6. Analysis of the micrographs
shows that increase in Current at same pulsed On-time results in coarse surface finish. Further,
increase in Pulse On-time at constant current results in more coarse surface finish. Migration of tool
electrode (copper) can be seen in all the cases.

5(a) 5(b)
Fig. 5Influence of current, (a) 8A and (b) 16A, over the machined surface at an On-time time of 106 µsec

6(a) 6(b)
Fig. 6Influence of Pulsed On-time, (a) 18 µsec and (b) 463 µsec, at current of 16A

130
 RTMST
4. Conclusion
Design and fabrication of the attachment for machining of cylindrical workpiece has been successfully
achieved and following conclusions can be drawn after the experimentation.
1. MRR increases with increase in current and pulsed On-time owing to higher energy input rate.
2. Higher current and higher On-time results in rougher EDM machined surface.
3. Rotation of the workpiece also contributes to MRR.

References

[1]. Masuzawa T, Fujino M, Kobayashi K, Suzuki and Fujii H. Wire electrode discharge grinding system for
machining very fine rods, International Conference on Computer-Aided Production Engineering,
Edinburgh, 1986, pp. 247-254

[2]. M. Ghoreishi, J. Atkinson. A comparative experimental study of machining characteristics in vibratory,

rotary and vibro-rotary electro-discharge machining, Fournals of Material Processing Technology, vol. 120,
2002, pp. 374-384

[3]. H.E. DeBryun. Effect of a magnetic field on the gap clearing in EDM, Ann. CIRP vol. 27, 1978, pp 93-95

[4]. T. Masuzawa, C.J. Heuvelman. A self-flushing method with spark erosion machining, Ann. CIRP, vol. 32,
1983, pp 109-111

[5]. Y.H. Guu, H. Hocheng. Effects of workpiece rotation on machinability during electrical-discharge
machining, Materials and Manufacturing Processes, 2001, pp. 91-101

[6]. Chen, S.L., Lin, M.H., Hsieh, S.F., Chiou. S.Y. The characteristics of cutting pipe mechanism with multi-
electrodes in EDM, Journal of materials processing technology, vol.203, 2008, pp. 461– 464

131
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18th and 19th April, 2013, NITTTR, Kolkata

Modelling and Optimization of Thermal Energy of CNC High

Speed End Milling of EN 362
Indranil CHAKRABORTY1 and N.K. MANDAL2
1. Mechanical Engineering Department
National Institute of Technical Teacher’s Training and Research,
Sector III Saltlake, Kolkata-700106
Email id: icneal33@gmail.com

2. Mechanical Engineering Department

National Institute of Technical Teacher’s Training and Research,
Sector III Saltlake, Kolkata-700106
Email id: mandal_nirmal@yahoo.com

Abstract:
Study of thermal behavior in end milling has been the subject of many current investigations using,
either, analytical, numerical or experimental approaches. In this research, experimental approach is
used. The effects of different cutting conditions like cutting speed, feed rate and depth of cut on cutting
temperature are studied. It is clear, higher cutting temperature decreases surface quality of finished
product, also it affects total tool life. In this experimental study, dry milling of EN 362 is considered,
since dry milling is much more economical and environment friendly. It is very important to obtain an
optimum combination of input cutting parameters. RSM and ANOVA is used for modeling and
optimization.

Keywords: Thermal behavior, cutting parameters, material, modeling & optimization.

1. Introduction
In order to optimize cutting conditions in a metal cutting process, the common conducted
investigations are concerning the evolution of the temperature distribution, since it highly effect the
quality of the end product i.e. it’s surface roughness. It is well known that the main sources of heat
generated, as a result of plastic deformation and friction occurring along the shear plane and tool-chip
interface, are respectively the primary and the secondary shear zones. Also, it is found from many
earlier research, most of the heat flows with the chip (almost 90%) while a small portion is transferred
to the tool and the work piece material. In last decades, numerical methods have become the main tool
for predicting metal cutting thermal behavior due to the advent of digital computers. In this project
experimental approach is used. The effects of different cutting conditions as cutting speed, feed rate
and depth of cut on cutting temperature are studied. In this experiment, dry milling of SAE 8620/EN
362 is considered, RSM and ANOVA is used for modeling and optimization.

132
 RTMST
M Gostimirovic, P Kovac and M Sekulic[1] said machining process causes development of large
quantities of thermal energy within a relatively narrow area of the cutting zone. The generated thermal
energy and the problems of its evacuation from the cutting zone account for high temperatures in
machining. These increased temperatures exert a pronounced negative effect on the tool and work
piece.C.K. Toh[2] investigated the feasibility of employing high depths of cut (10–20mm axial depth
of cut) by using an infrared red technique to measure the chip surface temperature. R. Komanduri, Z.B.
Hou[3] studied and presented Several techniques for the measurement of heat and the temperatures
generated in various manufacturing processes and tribological applications. G.M. Pittalà, M.
Monno[4]used a new methodology to the temperature prediction of milling. The temperature of the
workpiece, during the milling operation, has been measured using infrared camera. XuanceZhou,
QingshunBai, Kai Yang, ZhiLuo[5] produced the FEM module of micro-milling and established a
relationship between cutting force, damage of the tool and temperature.

2. Analytical study of temperature in end milling

The schematic diagram (Fig.1) given below shows a end milling process having following parameters
with their usual conventional symbols.

Fig.1: Mechanism of milling process

Where,

α is radial rake angle

λs is axial rake angle
f is feed mm/rev
d is depth of cut
t is chip thickness
ϕ is shear plane angle

133
 RTMST

Fig. 2: Parallel sided shear zone mode

The resultant force R is shown in Fig.1. Its inclination to the primary shear plane is, from geometry, (f
+ l – a). From the resolving R onto the shear plane, dividing it by the area of the plane and equating the
result to k leads to,
R = ------ (1)
ø (ø )

The cutting and thrust force components, Fc and Ft, also defined in Figure.1, are
( )
Fc = (ø )
------ (2)
ø

( )
Ft = ------ (3)
ø (ø )

Merchant suggested that chip thickness may take up a value to minimize the energy of cutting. The
well-known equation results,
ϕ = π/4 – (λ - α)/2 ------ (4)

Specific work during machining:

The work done per unit machined volume, the specific work, in metal cutting is Fc/(fd). The
dimensionless specific work, may be defined as Fc/(kfd). If we manipulate equation (2), we get,

( )
= = + tan (ø + λ − α) ------ (5)
ø (ø ) ø

Of the total specific work, some is expended on primary shear deformation and some on rake face
friction work. The specific primary shear work, Up, is the product of shear force kfd /sinf and velocity
discontinuity on the plane. After ‘non-dimensionalizing’ with respect to kfd, we get,

= ------ (6)
ø (ø )

Now, if we draw a velocity diagram from Fig.1,

(π/2-α)

Fig. 3: Velocity diagram

We can get,

= = ------ (7)
ø (ø )

134
 RTMST
Thermal modelling:
If all the primary shear work of equation (6) were converted to heat and all were
convected into the chip, it would cause a mean temperature rise ΔT1 in the chip,

∆ 1 = = ------ (8)
ø (ø )

ρc – Heat capacity of the chip material

For finding the total heat during the machining operation it needed to be clear about that,
Total specific work = Primary shear work + Rake face friction work

Fig. 4: Temperature zones in milling process

3. Instruments and material used:

Instruments
The instrument that is to be used to measure the temperature of end milling operation is an Infrared
Thermometer (METRAVI MT-18).

Fig. 5: Infrared Thermometer Fig. 6:Basic setup of Infrared Thermometer

Work material
The material to be used for the experimentation is SAE 8620/En362. SAE 8620 is a low nickel -
chromium - molybdenum medium hardenability, case hardening (carburizing) steel, generally supplied
in the as rolled condition with a maximum Brinell hardness of 280 (Rc30). Typical uses are: Arbors,
Bearings, Bushings, Cam Shafts, Differential Pinions, Guide Pins, King Pins, Pistons Pins, Splined
Shafts, Ratchets, and Sleeves etc.

Chemical Composition:

Carbon 0.23 Chromium 0.75

Silicon 0.40 Molybdenum 0.25
135
 RTMST
Manganese 0.95 Phosphorous 0.04
Nickel 0.75 Sulphur 0.04

Table 1. Chemical composition of EN 362

Tool
Commercially available PVD coated cemented carbide end mill inserts by Mitsubishi material corp.
are used in this investigation.
Cutter diameter: 20 mm
Rake angle: 11o
No. of teeth: 2 Nos.
Setting angle: 90o
4. Experiment and data collection

Experiments are designed with the help of using Response Surface Methodology (RSM). The software
used for DOE (Design of experiment) is Minitab16.

Parameters of the setting

Control factor Symbol

Spindle speed Factor A

Feed Rate Factor B

Depth of cut Factor C

Selected input Parameter

Control Level1 Level2 Level3 Units

factors/levels -1 0 1

Factor A 1500 2000 2500 rpm

Factor B 0.01 0.03 0.05 mm/rev
Factor C 0.20 0.35 0.50 mm

Design of experiments (DOE)

For selected input parameters experiments are designed using Response Surface Methodology (RSM)
full factorial design. For this purpose software Minitab 16 is used

Central Composite Design

Factors: 3 Replicates: 1
Base runs: 20 Total runs: 20
Base blocks: 1 Total blocks: 1

Two-level factorial: Full factorial

Cube points: 8
136
 RTMST
Center points in cube: 6
Axial points: 6
Center points in axial: 0

Alpha: 1

Sl. No. Run order A B C

1 1 0 0 0
2 2 -1 0 0
3 3 0 0 0
4 4 0 0 0
5 5 -1 -1 -1
6 6 0 0 0
7 7 1 1 1
8 8 -1 -1 1
9 9 -1 1 -1
10 10 0 -1 0
11 11 0 1 0
12 12 -1 1 1
13 13 1 0 0
14 14 1 -1 1
15 15 0 0 -1
16 16 0 0 0
17 17 1 -1 -1
18 18 0 0 1
19 19 0 0 0
20 20 1 1 -1

Table 2. Design of experiment and run order

Experimentation
After DOE, 20 experiments are carried out in CNC vertical End milling. For each experiment cutting
temperature and surface roughness is measured.

Run Order A B C Cutting

Speed Feed DOC Temperature
0
rpm Mm/rev mm c
1 0 0 0 126.4
2 -1 0 0 88.8
3 0 0 0 119.8
4 0 0 0 121.3
5 -1 -1 -1 73.7
6 0 0 0 118
7 1 1 1 214.2
8 -1 -1 1 79.2
9 -1 1 -1 82.3
10 0 -1 0 99.2

137
 RTMST
11 0 1 0 153.9
12 -1 1 1 135.9
13 1 0 0 114.3
14 1 -1 1 109.2
15 0 0 -1 85
16 0 0 0 124.7
17 1 -1 -1 67.8
18 0 0 1 135.3
19 0 0 0 128.2
20 1 1 -1 97.8

Table 3. Experimental data

ANOVA table has been used to summarize the test for significance of regression model, test for
significance for individual model coefficient and test for lack of fit. Summary output revealed that
quadratic model is statistically significant for response at the two different conditions. Significant
model terms are identified at 95% confidence level Goodness of fit. Degrees of freedom (DF) mean
the number of value that can independently of one another.

Term Coefficient SE Coefficient T P

constant 120.827 1.866 64.764 0
A 14.34 1.716 8.356 0
B 25.5 1.716 14.356 0
C 26.72 1.716 14.859 0
A*A -15.918 3.273 15.57 0.001
B*B 9.082 3.273 2.775 0.02
C*C -7.318 3.273 -2.236 0.049
A*B 8.713 1.919 4.541 0.001
A*C 12.338 1.919 6.43 0
B*C 15.338 1.919 8.02 0

Table 4. Estimated Regression Coefficients for cutting temperature

The mathematical relationship for correlating the cutting temperature using uncoded data and the
considered process variables has been obtained by using MINITAB Statistical Software 16 is as
follows:

T = 120.827+14.34*SPEED+25.5*FEED+26.72*DOC-15.918SPEED*SPEED
+9.082FEED*FEED-7.318DOC*DOC+8.713SPEED*FEED
+12.338SPEED*DOC+15.338FEED*DOC

138
 RTMST
Graphs and Plots:

Fig. 7: Main effects plot for cutting temperature

Fig. 8: Interaction plot

139
 RTMST

Surface Plot of CUTTING TEMP. vs C, A Surface Plot of CUTTING TEMP. vs B, A

Hold Values Hold Values
B 0 C 0

175
140
150
CUTTING TEMP. 120 CUTTING TEMP.
125
100
1 1
80 100

0 C 0 B
-1 -1
0 -1 0 -1
A 1 A 1

Surface Plot of CUTTING TEMP. vs C, B

Hold Values
A 0

175

CUTTING TEMP. 150

125
1
100
0 C
-1
0 -1
B 1

Fig.9: Surface plots

Optimization of Cutting Temperature:

Optimization is done using RSM and it is found that the combination of the highest value of speed and
lowest value of feed & depth of cut of the cutting parameters give least cutting temperature.

140
 RTMST
Optimal A B C
High 1.0 1.0 1.0
D Cur [1.0] [-1.0] [-1.0]
0.00000 Low -1.0 -1.0 -1.0

Composite
Desirability
0.00000

CUTTING
Minimum
y = 63.1302
d = 0.00000

Fig.10: Optimization plot for cutting temperature

Conformation Experiment:
In the present study response surface equation is derived from quadratic regression fit, so to verify
taking the independent variable values within the range for which the formula was derived performed
their validity conformation test. one conformation experiment was performed for cutting temperature.

SPEED FEED RATE DEPTH OF CUT CUTTING TEMP.

Exp. Predicted Error

2500 0.01 0.2 65.8 63.13 4.22%

Table 4.Conformation experimentation

5. Conclusion:
The present study develops cutting temperature models for three different parameters namely
RPM, feed rate and Depth of cut for turning process on EN 362 high carbon steel using
Response Surface Method. The second order response models have been validated with
analysis of variances. It is found that all the three machining parameters and their interactions
have significant effect on cutting temperature, considered in the present study. The value of
cutting temperature increases with decrease in RPM and increase in feed and depth of cut.
For minimum cutting temperature within the experimental range are RPM 2500rpm, Feed rate
0.01mm/rev and Depth of Cut 0.2mm.
141
 RTMST

6. References:

1. An inverse heat transfer problem for optimization of the thermal process in machining
M Gostimirovic∗, P Kovac and M Sekulic
Department of Production Engineering, Faculty of Technical Science,
University of Novi Sad, Trg D. Obradovica 6, 21000 Novi Sad, Serbia

2. Comparison of chip surface temperature between up and down milling orientations in high speed rough
milling of hardened steel, C.K. Toh∗
University of Birmingham, School of Engineering (Mechanical), Edgbaston Park Road, Birmingham
B152TT, UK

3. A review of the experimental techniques for the measurement of heat and temperatures generated in
some manufacturing processes and tribology, R. Komanduri *, Z.B. Hou
Mechanical and Aerospace Engineering, Oklahoma State University, Stillwater, OK 74078, USA

4. A new approach to the prediction of temperature of the workpiece of face milling operations of Ti-6Al-
4V,G.M. Pittalà, M. Monno
Department of Mechanical Engineering, Politecnico di Milano, Via La Masa 1,20156, Milano, Italy

5. Relationship between Cutting Temperature and Cutting Parameters of Micro-milling

Xuance Zhou, QingshunBai, Kai Yang, ZhiLuo.

6. FE thermal analysis of 3d ball end milling model Based on analytical study of oblique cuttingAbdallah
Nasri, Mihed Ben Said, Wassila Bouzid Sai
Laboratoire de Génie de Production Mécanique et Matériaux LGPM2,
ENIS, BPW 3038 Sfax, Tunisie

142
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

INITIATION TIME FOR ELECTRICAL DISCHARGE

THROUGH LIQUID MEDIUM

Debasish DAS
M.Tech. scholar, Department of Mechanical Engineering
National Institute of Technology Durgapur
Durgapur- 713209, India
dd4311@gmail.com; +91 9735431338

Debasish NANDI
Astt, Prof., Department of Mechanical Engineering
NSHM Knowledge Campus Durgapur
Durgapur- 713212, India
debasish_nandi1971@yahoo.co.in; +91 9434934195

Indrajit BASAK
Professor, Department of Mechanical Engineering
National Institute of Technology Durgapur
Durgapur- 713209, India
basak_indrajit@yahoo.com; +91 8900200404

ABSTRACT
Electric Discharge Machining (EDM), Electrochemical Discharge Machining (ECDM) and
Electrochemical Arc Machining (ECAM) are the non-conventional machining processes, which uses
the energy from electrical discharge for machining. They can be used to machine electrically
conductive materials as well as non-conductive materials. The discharge characteristics influenced by
a number of parameters such as power supply type and voltage, dielectric or electrolyte property, tool
and work piece materials, inter-electrode gap etc.
The energy and frequency of discharge to carry out machining is main consideration for engineers and
well discussed. The working voltage ranges for these processes are well established. But the pre
discharge phenomenon and mechanism of discharge in these voltage ranges is not investigated
sufficiently.
From the approach of physics, the mechanism of discharge at higher voltage range, kV to MV, is well
discussed and the available information includes dielectric pre-breakdown, streamer formation and
143
 RTMST
propagation, bubble generation etc. Also, the effects of dielectric properties and gases generated during
the process are established.
In the present work, study has been made to establish the time required to initiate discharge with
pulsed DC power supply in working voltage range of EDM, ECDM and ECAM. A specific time, at a
given condition, is required to initiate discharge. The dielectrics used are kerosene and distilled water
in stagnant and flowing condition. The voltage range selected is between 100V to 240V, and inter-
electrode gap of 0.005mm to 0.5mm. Copper is used as tool (cathode) while steel is used as work-piece
(anode).
It is concluded that the initiation time depend on the input voltage characteristic, dielectric property
and condition as well as to some extent to the electrode materials. If the critical time is not allowed, no
discharge can take place. The findings are in concurrence with the works of the high voltage
discharge.

Keywords: pre-breakdown, dielectric, inter-electrode gap, pulse duration, discharge voltage.

1. Introduction
In Electric Discharge Machining (EDM), the energy of electrical spark is utilized for the removal of
metal. Most commonly, a spark gap in the order of 10µm, and kerosene or deionised water as
dielectric medium are used. However, in certain situations special EDM oil may be used as dielectric.
The operating voltage lies between 80V and 300V. Controlled erosion of material occurs by the spark
energy, which turns into thermal energy through a series of discrete electrical discharges at the spark
gap. Due to electron bombardment on the non-conducting dielectric, it ionizes and leads to gaseous
plasma channel formation. Through this plasma channel, bombardment of ions and electrons occur on
the electrodes and intense localized heating takes place, causing melting even direct vaporization.
Within the plasma channel, the vapor bubble expands due to heating but, when the applied potential is
cut off, the heat source vanishes and the pressure inside the bubble reduces rapidly. This causes
implosion of the bubble and the dielectric rushes to the work zone and washes away the removed
materials and debris. This cycle is repeated for machining operations.

For Electrochemical Arc Machining (ECAM), the set up is similar to EDM, with inter-electrode gap is
of the order of 0.1-1mm. Here, electrolyte is used similar to Electrochemical machining (ECM).
Aqueous solution of NaCl, NaOH, KOH, HCl and NaF of different concentration are some of the most
commonly used electrolytes. Here, the commonly used operating voltage ranges from 25 to 80V.
Initially, at low voltages, electrochemical etching as in ECM is seen. However, as the voltage is
increased, bubbles accumulate near the cathode, thereby leads to gaseous plasma channel formation
and discharge occurs by EDM nature.

The set up for Electrochemical Discharge Machining (ECDM) is somehow different from the above
two. Though the electrolyte solutions similar to ECM or ECAM are used, two electrodes of largely
different surface area are chosen and placed large distance apart. The cathode, being small in
dimension has more charge density. Here, the work piece is not a part of the electrical circuit and. Any
material placed at a small gap near the cathode receives some part of the energy of discharge between
the cathode and the electrolyte and get machined. Here, the commonly used operating voltage ranges
from 15 to 30V. The input discharge energy during the processes of EDM, ECAM and ECDM
influences the material removal rate as well as surface roughness. As a result the machining aspects

144
 RTMST
remained the main center of research. In the present work the mechanism of the initiation of the
discharge has been explored rather than the energy of the discharge.

2. Literature Survey
It has been noted in the course of literature survey that the discharge mechanism is little explored at the
working voltage range of EDM, ECDM or ECAM and only a few work is available related to such low
voltage discharge and small inter electrode gap. However substantial of work is available on high
voltage discharge phenomenon from an approach from Physics, where the inter electrode gap is large.
Survey of these literature has been made to examine the discharge initiation, as the mechanism should
remain same for all the inter electrode gap condition. Hereafter, the surveyed literatures are grouped
as high voltage discharge and low voltage discharge.

2.1. High voltage discharge

The research on high voltage discharge phenomenon is mainly in the field of electrical engineering and
of plasma physics. W.G. Chadband [1] described the sequence of positive and negative streamer
breakdown through initiation, propagation, return stroke and arc formation stages. For positive
streamers, during initiation phase, when a high voltage is applied, a small spheroidal structure appears
at anode tip. During propagation phase, as the applied voltage is increased, filamentary structure is
found to grow, propagate and several branching of streamers occur. In the return stroke, the streamer
reaches the opposite electrode. Upon touching the cathode, it rapidly begins to thicken from the plane
cathode end, becoming luminous in the process. In the final step, arcs are formed. For negative
streamers, during initiation phase, small discharges cause local heating. This reduces density and
promotes bubble formation. It leads to electron avalanche. During propagation, the bubble expands
while in contact with cathode tip. Simultaneous branching occurs. The return stroke is absent. Before
reaching the opposite electrode, after crossing a certain distance, the streamers starts thickening and
luminous.

According to F. M. J. McCluskey et al. [2], an extremely weak conduction current (of order 10-12A) is
initially measured for lower voltages. However, when the applied voltage, while continuously
increasing finally crosses a certain threshold value (VT), the current jumps by several orders of
magnitude (10-7A) and becomes discontinuous in time but made up of identical evenly spaced
impulses. This immediate jump to space charge limited current at V=VT is indicative of electron
avalanche processes. The emitted light detected by the photomultiplier has a similar curve.

145
 RTMST

Fig 1. Current jump leading to electron avalanche [2]

I.M. Gavrilov et al. [3] worked on the dynamics of pre breakdown phenomena in a uniform field in
water for a breakdown time lag of 10-7 to 10-3 sec, and confirmed the existence of two types of
streamers. The probability of appearance of subsonic and supersonic streamers are described,
according to which, supersonic streamers appear over higher voltages compared to subsonic voltages.
Development of fast streamers results from kinetic phase transition in the liquid with subsequent
ionization of discharge channel, accompanied by shock waves.
F. M. J. McCluskey et al. [4] analyzed breakdown and pre-breakdown phenomena in liquids under
positive impulse voltages. Aluminum plate of 200mm diameter is used as work piece and Stainless
steel with tip radii 3μm is used as tool. Benzyl Toluene and mineral oil as dielectrics are used. They
found that varying the inter-electrode gap between 50-200mm didn’t provide significant changes in
streamer behavior. They also found a linear voltage-spark gap ratio.

146
 RTMST
Fig. 2. Voltage vs. Spark Gap graph [4]

E.O. Forster et al. [5] experimented the effect of different liquid dielectrics on breakdown voltage for
an inter-electrode gap of 5mm. Highly polished steel ball of 12.7mm diameter and a gramophone
needle with conical tip having a radius of curvature of ~35μm (approx) is used as electrodes with
trapezoidal voltage pulse. The voltage inter-electrode gap ratios with corresponding dielectrics were all
linear. Some of their results are shown in figure 3.

Perfluro polythers Galden XAD (800mW) Commercial transformer oil (Agip ITE360)

Fig 3. Voltage vs. Spark Gap graph [5]

O. Lesaint et al. [6] experimentally analyzed positive streamer electrical properties and propagation in
large oil gaps with Al plate of 200mm in diameter as work piece and a steel point of tip radius 100μm
as tool. It was observed that as the applied voltage is increased, the average values of longitudinal field
El within the streamers decreases. They also found that for some dielectrics an increase in voltage
induces branching of streamers. They also found a linear voltage-spark gap ratio as shown in figure 4.

147
 RTMST
Fig 4. Voltage vs. Spark Gap graph [6]

The positive and negative streamers originate from anode and cathode respectively. Based on the
velocity of propagation, M. T. Do et al. [7], classified the streamers as subsonic and supersonic
streamers, having different features. The subsonic streamers have bush like appearances, slow
propagation velocities of the order of m/s, higher probability of appearance at lower voltages of the
order of kilo volts, shows polarity effect, propagates in gaseous phases mostly due to vaporization. The
supersonic streamer have filament type appearances, higher propagation velocities of the order of
km/s, higher probability of appearance at higher voltages of the order of mega volts, do not show
polarity effect, propagates in liquid phases mostly due to ionization and dissociation.

Subsonic (Bush type) Supersonic (Filament type)

Fig 5. Streamers [7]

2.2. Low voltage discharge

Though the machining interest is the main objective in research works on EDM, ECDM and ECAM,
some works also explored the pre discharge as well as post discharge phenomenon.

The effect of additive in dielectric is studied by Han-Ming Chow et al. [8] for the micro-slit machining
of titanium alloy using electro discharge machining. Thin copper rotational diskette electrode is used
as tool (anode). The tool is rotated in between 0-150rpm. Ti-6Al-4V of thickness 1mm is used as work
piece (cathode). Kerosene is used as dielectric with a flow rate of 60l/h. SiC powders of diameter 1μm
with 0-75g/l concentration and aluminum powers of 5g/l concentration are used as additives in
dielectric. The power supply voltage is 250V and the gap voltage is 25V. The duty ratio is kept 80%.
The pulse duration (ton) used are 2, 5, 10, and 20μs. They concluded that presence of additives
increases MRR, depth of cut and surface roughness.

S.N. Joshi et al. [9] developed a non-linear, transient, thermo-physical modeling of die-sinking EDM
process using finite element method (FEM) to predict the shape of the crater cavity and MRR. Using
this model, the discharge parameters are analyzed parametrically. The inter-electrode gap is of the
order of 5-10μm. The applied voltage is of the order of 200V which breaks the dielectric for which the
discharge voltage is 25-30V. Special EDM oil with 20-40kpa is used as dielectric. Duty cycle of 50%
is maintained throughout the experiment. Steel is used as work piece (anode), while copper is used as
148
 RTMST
tool (cathode). The theoretical model’s prediction is very close to experimental results. Further, they
recommended different energy distribution factors for different energy zones.
A comparative study of ECM, ECDM and EDM experimentally by drilling a steel plate with a hollow
graphite pipe is as tool has been made by M. Kubota [10]. The voltage range was 20-30V and the gap
between the tool and work piece is 0.015-0.02mm. The inter-electrode gap is 2.5mm. He concluded
that metal removal in ECDM is a summation of ECM and EDM. ECDM could be practically used
either by using high speed ECM with maximum current density and slight discharge in smooth
machining, or to supply maximum electric power in small working area and to get maximum feed
speed in rough machining.
The spark generation in electrochemical discharge machining (ECDM) saw examined by Basak et al.
[11] to develop a theoretical model for predictions of discharge parameters and found that hydrogen
gases are generated at the cathode, when current passes through the electrolyte. These bubbles grow in
size and number leading to constriction, which cause ohmic heating of electrolyte, leading to vapor
bubble generation. At critical condition, these bubbles lead to isolation of tool and electrolyte, which is
called blanketing, which cause sparking.
Basak et al. [12] also worked on the mechanism of material removal in electrochemical discharge
machining (ECDM) to develop a theoretical model and performed experimental verification. This
model predicts the characteristics of the MRR for varying input parameters with the possibility of
enhancing the capacity of the process. An external inductance was used along with the inbuilt
inductance to observe the differences. In their experiment, a variable full wave rectified DC power
supply of frequency 100Hz. The voltage was varied between 0-80V. The gravity feed mechanism
maintains the vertical feed with constant contact between tool and work piece. Aqueous solutions of
25% NaCl, 25%-40% KOH and 25%-35% NaOH were used as electrolytes. Glass slides of 0.15cm
thickness were used as work piece. An inter electrode gap of 5cm, with direct contact between cathode
and work piece was chosen.
Kulkarni et al. [13] conducted an experimental study of time varying current in the circuit to identify
the mechanism of discharge in electrochemical discharge machining (ECDM). The ECDM cell is
fabricated using a Borosilicate glass flask of 10cm diameter. The inter-electrode gap is kept 3cm. The
separation of the cathode tip and the work piece is in a range of 500μm. The work piece is a square
copper block 2.5cm x 2.5cm with 0.6mm thickness. The experiments are carried out at a voltage of
150-155V, 1%-5% wt/vol aqueous solution of HCl.
Sarkar et al. [14] performed a parametric analysis on electrochemical discharge machining (ECDM)
thereby developed a second order, non-linear response surface methodology based mathematical
model to relate various machining parameters such as applied voltage, electrolyte concentration, inter-
electrode gap etc. with machining process criteria such as MRR, radial overcut (ROC), thickness of
heat affected zone (HAZ) etc. The mathematical model is verified using ANOVA. For the experiment,
ECDM micro drilling was performed on 20 x 20mm and 5mm thick silicon nitride ceramics using a
stainless steel tool with 400μm diameter with Pulsed DC power source. Inter-electrode gap varies from
27-39mm. 10%-30% wt/vol aqueous solution of NaOH is used as electrolyte. The experiments are
carried out at 50V, 54V, 60V, 66V and 70V.
Nandi et al. [15] analyzed the behavior of bubbles generated in electro-chemical discharge machining
(ECDM) to investigate the condition of the cell, up to critical voltage, with the objective to find the
effect of the cell resistance. The experiments were conducted with acidic, alkaline and neutral
electrolytes with commonly used electrolytes such as HCl, NaCl, and NaOH respectively. Carbon rod

149
 RTMST
was used as larger electrode (anode), while copper rod with diameters 0.5mm, 0.8mm, 1mm and
1.6mm are used as tool (cathode). The immersion depth used are 1, 2 and 3mm. Full wave rectified DC
supply is used as power supply. The voltage is varied between 0-20V while the corresponding current
varies between 0-3.2A. They plotted V-I curves to find three distinct zones. Zone 1 is linear with
voltage directly proportional to current. Its linearity is disturbed by local turbulence. For zone 2, the
increase in current is more for slight increase in voltage & the effect of turbulence is negligible. For
zone 3, rate of increase of current reduces with increase in voltage. It is also observed that path
resistance depends inversely on contact area of tool electrode and the electrolyte.

T. H. Drake et al. [16] analyzed the aspects of drilling holes by ECAM and establish the effects of
process variables such as voltage, feed rate etc. on MRR and TWR. They used an inter-electrode gap
of 1-4mm, a voltage range of 25-55V, and 25% wt/vol NaNO3 as electrolyte. 50mm long hollow
copper tube with inside and outside diameters being 1.325mm and 3.175mm respectively is used as
tool (cathode). Blocks of mild steel, 25mm long and 12.5mm by 12.5mm in cross section was used as
work piece (anode). The electrolyte pressure was kept at 30bar during drilling and the feed rate at
15mm/min. They concluded that the machining rate of ECAM is higher to that of EDM. However, the
dimensional accuracy of the prior is better.
Theoretical and experimental investigation of the relative effect of spark erosion and electrochemical
dissolution of ECAM was made by J. A. McGeough et al. [17], to establish the conditions favorable to
ECM or sparking. Theoretical analysis is compared with experimental results for the effects of pulsed
voltage and vibrating tool electrode waveforms. In the experiments, the inter-electrode gap is varied
between 0.1-0.6mm and the voltage is varied between 14-32V. 200g/l NaNO3 is used as electrolyte. A
100Hz pulsed rectified sinusoidal unsmoothed waveform is used as power source with the phase angle
varying from 0-90 degree. They observed that as the phase angle is increased, the equilibrium gap
reduces.
I. M. Crichton et al. [18] studied the discharge mechanisms in ECAM using 200μS pulses between
2mm diameter silver steel electrodes. The inter-electrode gap was between 10-90μm with input voltage
is between 0-85V. 3M NaCl and 3M NaNO are used as electrolyte. It provides an idea of discharge in
wide range of gaps in electrolyte.
X. W. Ni et al. [19] worked on an optical study of electrical discharge in electrochemical arc
machining. They analyzed cathodic evolution of hydrogen bubbles and observed the effect of wave
front optically during operation and found that either the charged bubble or the current wave front is
responsible for the electrical discharge. This behavior is different from that of static condition.

3. Present Work
From the surveyed literatures, the range of inter electrode gap and the discharge voltage are identified
and used in the present work. It is observed from the study of the high voltage discharge that a
particular series of events are necessary to initiate an electrical discharge between the electrodes. The
objective of the present work is set to investigate electrode gap- discharge characteristic as well as the
pre discharge time requirement at the machining voltage range. The assumption made is that the
pattern and phenomenon of pre discharge at high voltage range (kV-MV) and that of the low voltage
range (<kV) are same. Copper (ASTM electrical grade C11000) is used as tool (cathode) material and
150
 RTMST
steel (AISI1213) is used as work piece (anode) material, for all the experiments, as it is the observed
most common material combination for the related experiments. The other chosen working parameters
are as follows

3.1. EXPERIMENT CONDITION

Spark gap= 0.005mm to 0.5mm; Dielectric= Kerosene and Distilled water; Flow condition= Stagnant
and mild flow; Power supply= Pulsed DC (frequency 1- 1000Hz and Duty factor of the pulse= 1-
100%); Input voltage= 100- 240VDC; Ballast resistance= 250Ω in series.

3.2. EXPERIMENTAL SET UP

Fig 6. Schematic of the experimental set up

Figure 6 shows the experimental set up. The tool is held by an insulated pin vice and the job plate on
an insulated surface. The electrolyte flow is made by static head of the tank when necessary. A digital
height gauge is used to set the inter electrode distance and the zero gap was detected by the electrical
continuity by a separate low voltage circuit. For convenience of plotting the characteristics the inter
electrode distances are set at 0.005mm, 0.010mm and so on and the pulse voltage is increased up to the
point of the discharge initiation. Figure 7 shows the photographic image of the set up and the specimen
151
 RTMST
tool and work piece. The input voltage shape and the current spikes detected by the oscilloscope at the
onset of electrical discharge are shown in figure 8.

Fig 7. Experimental set up and the electrode specimens.

Fig 8. Input voltage shape and current spike at discharge

4. Observations
A fixed duty ratio of 0.5 is used to find the characteristic relation between the discharge voltages and
inter electrode gap for input voltage frequency 800Hz and 1000Hz. The other working conditions are
already described in chapter 3. Figure 9 shows the plot between discharge voltages and inter electrode
gaps for different pulse frequency as well as different dielectric with different flow conditions.

152
 RTMST

240 240

200 200

160 160

Voltage
Voltage

120 120

80 80

40 40
Flowing Flowing
Stagnant Stagnant
0 0
0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
Inter electrode gap (mm) Inter electrode gap (mm)
Frequency: 800Hz / Dielectric: Distilled Water Frequency: 1000Hz / Dielectric: Distilled Water

240 240

200 200

160 160
Voltage
Voltage

120 120

80 80

40 40
Flowing Flowing
Stagnant Stagnant
0 0
0.00 0.01 0.02 0.03 0.04 0.05 0.00 0.01 0.02 0.03 0.04 0.05
Inter electrode gap (mm) Inter electrode gap (mm)
Frequency: 800Hz / Dielectric: Kerosene Frequency: 1000Hz / Dielectric: Kerosene

Fig 9. Discharge voltage – inter electrode gap characteristic

Figure 10 shows relation between the discharge voltage and duty ratio at given working conditions.
The time duration of the pulse can be estimated from the pulse frequency and the duty factor. A strong
relationship has been observed between the pulse time and discharge voltage. This relation has

240 240
Flowing Flowing
Stagnant Stagnant
200 200

160 160
Voltage
Voltage

120 120

80 80

40 40

0 0
020 10 30 40 50 0 10
20 30 40 50
Duty Ratio (%) Duty Ratio (%)
Frequency: 1000Hz / Dielectric: Distilled Water / Inter electrode gap 0.005mm Frequency: 1000Hz / Dielectric: Kerosene / Inter electrode gap: 0.005mm

Fig 10. Discharge voltage – duty ratio characteristic

153
 RTMST
been shown in figure 11 for different inter electrode gap.

Fig 11. Discharge voltage – pulse time characteristic

5. Analysis and Discussion

From the discharge voltage vs. inter-electrode gap characteristics (Fig 9), it is clear that discharge
voltage is linearly proportional to the inter-electrode gap at any given condition. It is higher for the
flowing condition than the stagnant condition. These are in agreement with the observed mechanism
from the literature survey, as the required conditions for the discharge such as the bubble
accumulation, channel growth etc, are disturbed due to the flow. Increase in frequency keeping duty
ratio same or keeping frequency same decrease in duty ratio, means smaller pulse duration. Figure 9
also indicates with smaller pulse duration discharge voltage required for the discharge is higher. This
phenomenon is clearly shown in figure 10. As the time allowed is smaller, it requires higher rate to
complete the entire pre discharge requirement, which can only be facilitated by higher applied voltage.
Figure 11 explain the effect of inter electrode gap on the discharge voltage. As the pulse duration
increase the voltage approaches to a minimum equilibrium, indicating more than the minimum
required time for the discharge has been allowed. The required voltage increases with smaller pulse

154
 RTMST
time, in agreement of the indication of figure 10. On the other hand, below a lower critical value of
the pulse duration, no discharge can take place even at a highest value of the input voltage within the
working range.
Both the dielectric, distilled water and kerosene has shown the same characteristic. This seems
unnatural as they are of different dielectric strength. This should be investigated further. Also further
experimentations are necessary to investigate the effect of the electrode material on the discharge
voltage.

6. Conclusion
It is observed from the literature survey that even if the inter electrode gap is large, the discharge can
occur when the applied potential is high. The magnitude of the applied voltage required for such
discharge depends on the type of dielectric medium as well as the inter electrode gap. The variation of
the voltage with inter electrode gap is linear for all the cases, when the medium remain same. The
conducted experiments at the working voltage range of electrically assisted nonconventional
machining also show the similar nature. Hence, it can be concluded that the mechanism of discharge
remains same imperative of the applied potential. It is observed from the pulse duration of the applied
voltage that it requires certain time for discharge initiation. This initiation time depends on the input
voltage magnitude, dielectric property and flow condition of the dielectric. The effect of the properties
of the electrode material may have the influence on the discharge characteristic, which was not put
under the scope of the present work.

7. Acknowledgement

The authors are thankful to the authority of the NIT Durgapur for supporting the experimental
investigation and research of the ongoing work.

8. References
1. W.G. Chadband- From bubbles to breakdown, or vice-versa, Conduction and Breakdown in Dielectric
Liquids- ICDL, IEEE 11th International Conference on Engineered Materials, Dielectrics & Plasmas; Fields,
Waves & Electromagnetics, (DOI: 10.1109/ICDL.1993.593936), pp184 – 193, July 1993.

2. F. M. J. McCluskey, F. Huynh, A. Denat, N. Bonifaci and R. Kattan- Bubble formation in 'simple' and complex
liquids- Conference Record of the 1994 IEEE International Symposium on Electrical Insulation. Pittsburgh.
PA USA, (DOI: 10.1109/ELINSL.1994.401483), pp 406-409, June 1994.

3. I. M. Gavrilov, V. R. Kukhta, V. V. Lopatin, P. G. Petrov- Dynamics of Prebreakdown Phenomena in a

Uniform Field in Water- IEEE Transactions on Dielectrics and Electrical Insulation, (DOI:
10.1109/94.300293), vol. 1, issue 3, pp. 496-502, June 1994.

4. F. M. J. McCluskey, A. Denat, O. Lesaint- Breakdown and Pre-breakdown Phenomena in Liquids under

Positive Impulse Voltages- IEEE Transactions on Dielectrics and Electrical Insulation, (DOI:
10.1109/94.300277), vol 1, issue 3, pp 377-382, June 1994.

5. E. O. Forster, H. Yamashita, C. Mazzetti, M. Pompili, L. Caroli, S. Patrissi- The effect of the electrode gap on
breakdown in liquid dielectrics- IEEE Transactions on Dielectrics and Electrical Insulation- vol 1, issue 3, pp
440-446, (DOI:10.1109/94.300287), June 1994.

155
 RTMST

6. O. Lesaint, G. Massala- Positive streamer propagation in large oil gaps experimental characterization of
propagation modes, Dielectrics and electrical insulation- IEEE Transactions, vol 5 , issue 3 , (DOI:
10.1109/94.689425), pp 360-370, 1998.

7. M. T. Do, A. Nysveen- An Experimental Study on the Effect of DC Bias on Streamer Initiation and
Propagation in a Dielectric Liquid under Impulse Voltage- IEEE Transactions on Dielectrics and Electrical
Insulation, vol 16, issue 6, pp 1623-1631, December 2009.

8. Han-Ming Chow, Biing-Hwa Yan, Fuang-Yuan Huang, Jung-Cherng Hung- Study of added power in kerosene
for the micro-slit machining of titanium alloy using electro discharge machining- Journal of Materials
Processing Technology, vol. 101, issue 1-3, pp 95-103, 2000.

9. S. N. Joshi, S. S. Pande- Thermo-physical modeling of die-sinking EDM process- Journal of Manufacturing

Processes, Elsevier, vol. 12, issue 1, pp 45-56, January 2010.
10. Mamoru Kutoba- Metal removal in ECDM, ISEM-5, Switzerland, pp 217-220, June 1977.

11. Indrajit Basak, Amitabha Ghosh- Mechanism of spark generation during electrochemical discharge
machining: a theoretical model and experimental verification- Journal of Materials processing Technology,
Elsevier, no 62, pp 46-53, 1995.

12. Indrajit Basak, Amitabha Ghosh- Mechanism of material removal in electrochemical discharge machining: a
theoretical model and experimental verification, Journal of Materials processing Technology, Elsevier, no
71, pp 350-359, 1997.

13. A. Kulkarni, R. Sharan and G. K. Lal- An experimental study of discharge mechanism in electrochemical
discharge machining, International Journal of Machine Tools and Manufacture, Pergamon, no 42, pp1121-
1127, 2002.

14. B. R. Sarkar, B. Doloi and B. Bhattacharyya- Parametric analysis on electrochemical discharge machining of
silicon nitride ceramics, Springer Link, International Journal of Advanced Manufacturing Technology, vol
28, issue 9-10, pp 873-881, (DOI: 10.1007/s00170-004-2448-1) , 2006.

15. Debasish Nandi, Asit Baran Puri and Indrajit Basak- Behavior of bubbles generated in electro-chemical
discharge machining, International Journal of Engineering Science and Technology, ( ISSN: 0975-5462), vol
3, no 12, pp 8274-8280, December 2011.

16. T. H. Drake, J. A. McGeough- Aspects of drilling by Electrochemical Arc Machining- Proceedings of 22nd
IMTDR, Macmillan Publishers, pp 361-369, 1981.

17. J. A. McGeough, A. B. M. Khayry, and W. Munro- Theoretical and experimental investigation of the relative
effect of spark erosion and electrochemical dissolution in electrochemical arc machining- Annals of CIRP,
vol 32, no 1, pp 113-118, 1983.

18. I. M. Crichton, J. A. McGeough- Studies of the discharge mechanisms in ECAM- Journal of applied
electrochemistry, vol 15, pp113-119, 1985.

156
 RTMST
19. X. W. Ni, J. A. McGeough- An optical study of electrical discharge in electrochemical arc machining,
research and technological developments in nontraditional machining- Winter annual meeting of The
American society of mechanical engineers, Chicago, PED-vol 34, pp 63-73, 1988.

157
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Surface Modification of Graphite by EDM using Powder

Metallurgical Green Compact Tools

NGANGKHAM Devarani*, P. K. PATOWARI

Department of Mechanical Engineering
National Institute of Technology Silchar
P.O. Silchar 788010, Assam, India
[e-mail: dev.ngangkham@gmail.com, ppatowari@yahoo.com]
(*Corresponding author)

Abstract: At present, Electrical Discharge Machining (EDM) is a widespread technique used in

industry for precision machining of all types of conductive materials irrespective of the intensity of
hardness. Though EDM is essentially a material removal process, efforts have been made to use it as a
surface modification process. In this present work, an attempt has been made to deposit SiC on
graphite block using Si, Cu and graphite powder metallurgical green compact electrodes to obtain a
high hardness surface of the graphite block. Machining parameters like pulsed current, pulse on-time,
composition, compaction load are taken as input parameters to study the Material Transfer Rate
(MTR). An MTR up to an extent of 9mg/min with surface roughness of 1.5µm has been obtained at an
on-time of 106µsec and current of 2A.

Keywords: Electrical Discharge Machining, Powder Metallurgy Electrode, Surface Modification

1. Introduction:
At present, Electrical Discharge Machining (EDM) is a widespread technique used in industry for
high-precision machining of all types of conductive materials such as metals, metallic alloys, graphite
or even some ceramic materials irrespective of its hardness. EDM is a non-conventional manufacturing
158
 RTMST
process and proves to be an effective alternative as compared to traditional manufacturing processes. It
is an electro-thermal process in which the material is removed by a series of carefully controlled
discrete sparks (electrical discharges) generated between the tool and workpiece, submerged in
dielectric.

Migration of tool material onto the workpiece surface had been observed during machining by EDM
and hence this phenomenon has been exploited by many researchers in order to cause significant
transfer of selected tool material onto the machined surface for obtaining a specific functional
characteristics like improving resistance to wear, erosion and indentation, reducing friction, improve
lubrication and improving resistance to corrosion and oxidation, improve fatigue resistance and also
for biomedical applications where the machined surface by EDM shows good biocompatibility and
biological activity.

In EDM process using liquid dielectric, metal transfer from the tool electrode to the work surface
can be enhanced using powder compact tools with reverse polarity [1]. Wear of tool electrode
can be made higher by using green compact or sintered product as an electrode, which is desirable for
surface modification [2]. Experimental investigation on the surface modification of 5083 Al alloy by
Electrical Discharge Alloying (EDA) process with a Si–Fe alloy as an electrode shows that micro-
hardness of EDA alloyed layer was evidently higher than that of the base metal (5083 Al alloy) and
EDA alloyed layer with composite microstructures has good corrosion resistance in NaCl
aqueous solution [3]. Surface modification process by EDM with a Ti powder green compact
electrode carried out by Wang et al. [4] shows that TiC ceramic layer can be created over a metallic
surface. The hardness of the ceramic layer is more than three times as that of the base material and the
hardness changes gradually from the surface to the base material. Coating of TiC over Ni surface has
also been achieved by using multilayer electrodes composed of Ti and Graphite layers of same
dimensions stacked together [5].

In this present work, an attempt has been made to deposit SiC on graphite block using Si, Cu and
Graphite Powder Metallurgical (P/M) compact electrodes to obtain a high hardness surface of the
graphite block. Fig.1 shows the schematic diagram of the principle of surface modification.

Fig.1. Schematic diagram of the principle of surface modification

159
 RTMST
It is expected that during electrical discharge machining of Graphite with tool electrode of
Si/Cu/Graphite P/M tool, migration of Si from the tool electrode and carbon from both the tool and
decomposed hydrocarbon oil occurs at very high temperature leading to the formation of SiC layer
onto the graphite W/P surface.

2. Experimentation

2.1 Experimental set-up

Experiments have been conducted on Electrical-discharge machine model F25 series of Sparkonix (I)
Pvt. Limited, Pune, India, which is shown in Fig. 2. The machine can work up to current of 25A, on-
time of 2000µsec with maximum weight of W/P up to 650Kg and that of the tool up to 35Kg.

Fig. 2 EDM Machine

2.2 Preparation of Powder Metallurgical (P/M) Green Compact Electrodes

The compacts are prepared using KBr die set and manual pellet press manufactured by Kimaya
Engineers. In this process, silicon, graphite and cooper of mesh size 300,-60 and 325 respectively, are
mixed thoroughly for an hour using 4” mortar and pestle, with required composition by weight. A
160
 RTMST
small amount of zinc stearate mixed with alcohol is applied on the walls of the die which acts as
lubricant. The mixture is then poured into a die of 13mm diameter. The mixture of the powders is
compacted with required compaction loads (tons) in a pellet press.
The prepared compact electrodes are pasted to a copper rod of 5cm length and 13mm diameter with
the help of electrically conductive epoxy (Silver Loaded Epoxy Hardener and Silver Loaded Epoxy
adhesive) to make complete powder metallurgical compact electrodes. Figures 3(a) and 3(b) show the
pellet tool electrodes respectively.

3(a) 3(b)
Fig. 3(a) Pellet and 3(b) Tool electrode
2.3 Preparation of workpiece

The workpiece selected for the experimentation is graphite block of dimension 20208mm.
Necessary finishing operation has been done to ensure flatness and finishing of the workpiece surface.

2.4 Machining condition

Machining was performed over the graphite block with varying input parameters such as composition
of the tool, pulsed On-time, current, etc at different combinations. The parameters are as listed in
Table1.

Table1. Input parameters and values

Polarity Reverse
Flushing pressure 2kg/cm2
Tool Composition (% by wt.) 50Si/25G/25Cu, 40Si/30G/30Cu
Workpiece Graphite
On time (µs) 25, 57, 106
Compaction load (tons) 15
Current (A) 2,4
Machining time (minutes) 10
Flushing medium EDM oil

161
 RTMST

3. Results and discussion

The workpiece and tool after machining are as shown in Fig. 4. Increase in weight of the workpiece
and change in color indicates that material has been transferred from the tool to the workpiece. The
strength of the tool is not sufficient enough for working at higher current due to the weak strength and
presence of crack at the compacted tool.

4(a) 4(b)

Fig. 4(a) Workpiece and 4(b) Tool after machining

3.1 Material Transfer Rate (MTR)

Increase in weight of the work material has been measured using Ishida DX Precision balance and the
MTR has been evaluated by taking the increase in weight per unit of the machining time. Increase in
weight of the work material after machining indicates that material has been deposited over the work
surface. So, Material Transfer Rate (MTR) is the main output parameter considered.

Composition of the tool plays a major role in MTR, as indicated in Fig. 5. With increase in percentage
of Si composition of the tool, MTR increases. It has also been observed that increase in pulsed on-
time increases the MTR due to higher energy input rate, keeping current constant at 2A.

162
 RTMST

MTR Vs On-time
10

8
MTR (mg/min)

4 50Si/25G/25Cu
40Si/30G/30Cu
2

0 At 2A
25 57 106
On time (µsec)

Fig5. Effect of pulsed On-time on MTR at different tool composition at current of 2A

The effect of Pulsed on-time on MTR at two different current has been studied by keeping the
composition of the tool constant at 50Si/25G/25Cu as this composition have shown better effect on
MTR with reference to previous graph. Figure 6 shows that increase in on-time and current result in
higher MTR owing to higher energy input rate. An MTR up to an extent of 9mg/min has been obtained
at an on-time of 106µsec and current of 2A.

MTR Vs On-time
10
9
8
7
MTR (mg/min)

6
5
2A
4
3 4A
2
1
0
25 57 106
On-time (µsec) At 50Si/25G/25Cu

Fig. 6. Effect of pulsed On-time on MTR at different current using tool of composition 50Si/25G/25Cu

3.2 Surface roughness

Surface roughness of the deposited surface has been measured using Handysurf E-35A Portable
Surface Measuring Unit. Roughness value of 1.5 µm has been attained at the optimum condition (On-
time of 106 µsec, current of 2A and composition of 50Si/25G/25Cu). Fig. 7 shows the roughness
profile curve of the deposited surface at the optimum condition.
163
 RTMST

Fig. 7. Roughness profile curve of the deposited surface showing Ra of 1.5 µm at 106µsec on-time and 2A
current

4. Conclusion
From the detailed study of the experimental results, it can be concluded as under.

1. SiC layer can be deposited over graphite block by EDM using powder metallurgical green
compacted tools.

2. Increase in current, on-time and percentage of Si composition enhance the deposition.

3. Higher percentage of Si is not always a preferable condition as the strength of the compact is
weak which leads to breaking of the compacted tool during machining; which is a major
challenge in working with Si compacted P/M tool.

4. Tendency of arcing is higher at the point of crack present in the tool which is not a favorable
condition.

References

[1]. Gangadhar, M. S. Shunmugam, P. K. Philip. Surface Modification in Electro Discharge Processing with a
Powder Compact Tool Electrode, Wear, Vol. 143, 1989

[2]. N. Mohri, N Saito, Y. Tsunekawa. Metal Surface Modification by Electrical Discharge Machining with
Composite Electrode, Annals of the CIRP, Vol. 42, 1993

[3]. Kuralay Stambekova, Hung-Mao Lin, Jun-Yen Uan. Surface modification of 5083 Al alloy by electrical
discharge alloying processing with a 75 mass% Si–Fe alloy electrode, Applied Surface Science, 2012.

164
 RTMST
[4].Z. L. Wang, Y. Fang, P. N. Wu, W.S. Zhao, K. Cheng. Surface Modification Proccess by Electrical
Discharge Machining with a Ti powder green compact electrode, Journal of Material Processing
Technology, Vol. 129, 2002, pp. 139-142

[5].Yu-Lung Hwang, Chia-Lung Kuo, Shun-Fa Hwang. The coating of TiC layer on the surface of Nickel by
electric discharge coating (EDC) with multi-layer electrode, Journal of Material Processing Technology,
Vol 210, 2010, pp. 642-652

165
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

An Experimental Investigation of Optimum Flank Wear of

Carbide Inserts for Dry Turning of EN19 Tool Steel
Saurav NANDIa, Nirmal Kumar MANDALb
a
Assistant Professor, Mechanical Engineering Department,
Calcutta Institute of Engineering and Management
24/1A, Chandi Ghosh Road, Kolkata – 700040. India
Mob: 9230544014. E-mail: sauravsdh@gmail.com
b
Associate Professor, Mechanical Engineering Department,
NITTTR-Kolkata, Block-FC, Sector-III, Kolkata – 700106. India
Mob: 9432162792, E-mail: mandal_nirmal@yahoo.com

Abstract:

Flank wear of cutting tool is often selected as the tool life criterion because it determines the diametric
accuracy of machining, its stability and reliability. In this work, the effect of cutting parameters like
RPM, Feed Rate and Depth of Cut is studied on dry turning of EN 19 tool steel and the parameters are
optimized for minimum tool wear. RSM technique is employed to achieve the minimum tool wear.
Combined effects of the cutting parameters on tool wear (VB) are investigated by the help of
orthogonal array and ANOVA under dry turning. The optimum tool wear for a combination of cutting
parameters are determined and the tool wear is predicted.

Keywords: Tool wear, CNC-Turning, Response Surface Methodology.

1. Introduction

EN 19 tool steel is used in large scale in automobile industry. Turning operation is done to find the flank wear
measured with a tool makers microscope. Tool wear is defined as the amount of volume loss of tool material on
the contact surface due to the interactions between the tool and work piece, this contact surface is basically the
flank surface. Specifically, tool wear is described by wear rate (volume loss per unit area per unit time) and is
strongly determined by temperature, stresses, and relative sliding velocity generated at the contact interface.
Metal cutting tools are subjected to extremely arduous conditions, high surface loads, and high surface
temperatures arise because the chip slides at high speed along the tool rake face while exerting very high normal
pressures (and friction force) on this face. Now to minimize the heat development at the tool work piece
interference region cutting fluids are applied, but here we have incorporated dry turning condition to encourage
green machining and it is also seen the dry machining also deduces the production cost by 16% of the
machining cost.

166
 RTMST
Response Surface Method (RSM) is a mathematical and statistical technique which is used to find the optimum
condition for modeling and analysis of problems in which a response of interest is influenced by several
variables. Response surface helps in analyzing the influence of the independent variables on a specific
dependent variable (response) by quantifying the relationships amongst one or more measured responses and the
vital input factors. The mathematical models thus developed relating the machining responses and their factor
gives the optimum machining process. In RSM, the relationship between the response and the independent
variables is unknown. Usually when the response function is not known or non-linear, a second order model is
utilized in the form.
n n
y  b0   bi xi   bii xi2    bij xi x j   (1)
i 1 i 1 i j

2. Experimental Details
The following studies are carried out
 Identification of important process parameters and selection of their levels.
 Development of the design matrix. Conducting the experiments as per the design matrix.
 Recording of the response.
 Development of mathematical model using Response Surface Methodology.
 Calculation of the co-efficient of the polynomials.
 Analysis of experimental result for tool wear.
 Optimization of process parameter using Response Surface Methodology.

3. Response Surface Methodology

In equation 1, ε represents the noise or error observed in the response y. Response is (y - ε) and b’s are
the regression coefficients to be estimated. The least square technique is being used to fit a model
equation containing the input variables by minimizing the residual error measured by the sum of
square deviations between the actual and estimated responses. The calculated coefficients or the model
equations however need to be tested for statistical significance and thus the following tests are
performed.

To check the adequacy of the model for the responses in the experimentation, Analysis of Variance
(ANOVA) is used. ANOVA calculates the F-ratio, which is the ratio between the regression mean
square and the mean square error. The F-ratio, also called the variance ratio, is the ratio of variance
due to the effect of a factor (the model) and variance due to the error term. This ratio is used to
measure the significance of the model under investigation with respect to the variance of all the terms
included in the error term at the desired significance level, α. If the calculated value of F-ratio is higher
than the tabulated value of F-ratio for roughness, then the model is adequate at desired α level to
represent the relationship between machining response and the machining parameters.

In the ANOVA Table, there is a P-value or probability of significance for each independent variable in
the model the value of which shows whether the variable is significant or not. If the P-value is less or
equal to the selected α -level, then the effect of the variable is significant. If the P-value is greater than
the selected α -value, then it considered that the variable is not significant. Sometimes the individual
variables may not be significant. If the effect of interaction terms is significant, then the effect of each
factor is different at different levels of the other factors. ANOVA for different response variables are
167
 RTMST
carried out in the present study using commercial software Minitab (Minitab user manual, 2001) with
confidence level set at 95%, i.e., the α -level is set at 0.05.

4. Experimentation
The central composite design (CCD) is a design widely used for estimating second order polynomial
for the response variables in response surfaces methodology without using a complete full factorial
design of experiments. It is perhaps the most popular class of second order designs. Since introduced
by Box and Wilson (1951), the CCD has been studied and used by many researchers. To establish the
coefficients of a polynomial with quadratic terms, the experimental design must have at least three
levels of each factor. In CCD, there are three different point viz. factorial points, central points and
axial points. Factorial points are vertices of the n-dimensional cube which are coming from the full or
fractional factorial design where the factor levels are coded to -1, +1. Central point is the point at the
center of the design space. Axial points are located on the axes of the coordinate system symmetrically
with respect to the central point at a distance α from the design center.

4.1 Process Variables and Their Limits

The working ranges of the parameters for subsequent design of experiment, based on Response
Surface Methodology with rotatable design have been selected. In the present experimental study
spindle speed, feed rate and depth of cut have been considered as process variables. The process
variables with their units (and notations) are listed in Table 3.

Table 1 Process Variables

LEVEL
-1.682 -1 0 1 1.682
VARIABLES
DOC (mm) 0.159 0.5 1 1.5 1.841
FEED RATE(mm/rev) 0.032 0.1 0.2 0.3 0.368
RPM 230 400 650 900 1070

Table 2 Tool Wear Data (FW) with Coded an Un-coded Values

RUN ORDER CODED VALUES RESPONSE
RPM FEED DOC FW(mm)
1 0 0 1.681793 0.116
2 -1 -1 -1 0.081
3 -1 -1 1 0.105
4 -1 1 -1 0.09
5 1 -1 -1 0.115
6 1.681793 0 0 0.143
7 0 0 0 0.121
8 1 1 -1 0.118
9 1 1 1 0.135
10 -1.68179 0 0 0.088
11 0 1.681793 0 0.127
168
 RTMST
12 0 0 0 0.124
13 0 0 0 0.122
14 1 -1 1 0.131
15 0 -1.68179 0 0.118
16 0 0 0 0.123
17 0 0 0 0.121
18 0 0 0 0.12
19 -1 1 1 0.1
20 0 0 -1.68179 0.079

5. Result and Discussion

The second order response surface equation has been formed using Minitab software for the response
tool wear.
= 0.121893 + 0.01578 × + 0.001914 × + 0.009462 × ℎ − 0.00263 ×
× − 0.008994 × ℎ × ℎ
(2)

Table 3 ANOVA table for Tool Wear (FW)

Term Coef SE Coef T P
CONSTANT 0.1217662 0.000945 128.899 0.000
A 0.0157795 0.000738 21.371 0.000
B 0.0019138 0.000738 2.592 0.021
C 0.0094624 0.000738 12.815 0.000
A*A -0.002615 0.000715 -3.656 0.003
C*C -0.008978 0.000715 -12.554 0.000
S = 0.00272861 PRESS = 0.000277213
R-Sq = 98.26% R-Sq(pred) = 95.38% R-Sq(adj) = 97.64%

20 experiments were conducted in duplicate and the average values of Tool Wear with design matrix
were tabulated in Table 2. For analysis the data, the checking of goodness of fit of the model were
carried out. The model adequacy checking includes the test for significance of the regression model,
test for significance on model coefficients, and test for lack of fit. For this purpose, analysis of
variance (ANOVA) is performed. The fit summary recommended that the quadratic model is
statistically significant for analysis of Tool Wear.

Table 4 Analysis of Variance for Tool Wear (FW)

DF Seq SS Adj SS Adj MS F P
Regression 5 0.005894 0.005894 0.001179 158.34 0
Linear 3 0.004673 0.004673 0.001558 209.23 0
Square 2 0.001221 0.001221 0.006917 41.93 0
Interaction 3 0.015304 0.015304 0.000611 82 0
Residual Error 14 0.000104 0.000104 0.000007
Lack-of-Fit 9 0.000093 0.000093 0.00001 4.79 0.05
169
 RTMST

Pure Error 6 0.000011 0.000011 0.000002

Total 19 0.005999

5.1 Checking the Adequacy by Response Surface Methodology Variation of Tool Wear with
respect to input parameters

Figure.1 Main Effects Plot for Tool Wear (FW in mm)

Figure 2 Normal Probability Plot of Residuals for Tool Wear

Figure 3 Plot of Residuals vs. Fits for Tool Wear

The parametric analysis has been carried out to study the influences of the input process parameter
such as RPM, Feed Rate and Depth of Cut on the process responses, i.e. Tool Wear in dry turning
process. Contour plots and three-dimensional response surface plots were formed based on the RSM
quadratic models to evaluate the change of response surface. These plots can also give further
assessment of the correlation between the process parameters and responses. Figure 4 illustrates the
contour plot and response surfaces of Tool Wear with respect to input parameters RPM and Feed Rate.
The value of Tool Wear is shown to decrease with decrease of RPM and Feed Rate. In the Figure 5

170
 RTMST
contour plot and response surface of Tool Wear with Feed Rate and Depth of Cut is depicted. Tool
wear decreases with decrease in Feed Rate and Depth of Cut.

Figure 4 Variation in Tool Wear according to change in RPM and Feed Rate

Figure 5 Variation in Tool Wear according to change in Feed Rate and Depth of Cut

Figure 6 Variation in Tool Wear according to change in RPM and Depth of Cut

171
 RTMST
Figure 6 also depicts that the Tool Wear decreases with the decrease in respective RPM and Depth of
Cut. Thus lower the input parameters, lower the tool wear will be.

6. Conclusion
Experimental investigation on CNC turning machining on EN19 tool steel was performed with a view
to correlate the process parameters with the response as Tool Wear. The process has been successfully
modeled using RSM and model adequacy checking was also carried out using Minitab. The second-
order response models have been validated with analysis of variance. Finally, an attempt has been
made to estimate the optimum machining conditions to produce the best possible responses within the
experimental constraints. This research can also help researchers and industries for developing a
robust, reliable knowledge base and early prediction of Tool Wear without experimenting with turning
process for EN 19 tool steel.

References
[1] D. Dinakaran, S. Sampathkumar, J. Susai Mary,”Real Time Prediction of Flank Wear by Neuro Fuzzy
Technique in Turning”, JJMIE Volume 4, Number 6, December 2010 ISSN 1995-6665 Pages 725 – 732.
[2] R. Marumo, M. T. Letsatsi and O. S. Motsamai, “Fuzzing modelling of a single point lathe cutting tool”,
Journal of Mechanical Engineering Research Vol. 3(7), July 2011, pp. 264-288,.
[3] H. Chelladurai & V. K. Jain & N. S. Vyas, “Development of a cutting tool condition monitoring system for
high speed turning operation by vibration and strain analysis”, Int J Adv Manuf Technol (2008) 37:471–485
DOI 10.1007/s00170-007-0986-z.
[4] Vishal S. Sharma · S. K. Sharma · Ajay K. Sharma, “Cutting tool wear estimation for turning”, J Intell
Manuf (2008) 19:99–108 DOI 10.1007/s10845-007-0048-2.
[5] Yung-Chang Yen, Jörg Söhner, Blaine Lilly, Taylan Altan, ”Estimation of tool wear in orthogonal cutting
using the finite element analysis”, Journal of Materials Processing Technology 146 (2004) 82–91.
[6] W. Bouzid Sa, “An investigation of tool wear in high-speed turning of AISI 4340 steel”, Int J Adv Manuf
Technol (2005) 26: 330–334 DOI 10.1007/s00170-003-1991-5.
[7] Tug˘rul O¨ zel, Yig˘it Karpat, “Predictive modeling of surface roughness and tool wear in hard turning using
regression and neural networks”, International Journal of Machine Tools & Manufacture 45 (2005) 467–479.
[8] XIAOLI LI, “Real-time tool wear condition monitoring in turning”, INT. J. PROD. RES., 2001, VOL. 39,
NO. 5, 981–992.
[8] Viktor P. Astakhov, “The assessment of cutting tool wear”, International Journal of Machine Tools &
Manufacture 44 (2004) 637–647.

172
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18th and 19th April, 2013, NITTTR, Kolkata

Adaptation of Compliant Multi Bar Mechanism for

Generation of Compliant Segment Motion

Dr. Debasish Dutta 1, Subhankar Das2

1 Assistant Professor, Department of Mechanical Engineering, NIT Silchar
e-mail id: debasish.ddutta@gmail.com Mobile no. +919435740839

2 Department of Mechanical Engineering, NIT Silchar

e-mail id: subha.me31@gmail.com Mobile no. +919577477174

Abstract: This paper tells about a systematic procedure for synthesis of single-loop compliant
mechanism. This kind of mechanism characterized for no moving rigid-links compliant-segment
motion generation. This kind of complaint mechanism is applicable for adaptive structures. Compliant
four-bar mechanisms are constitutes of atleast one rigid moving link, and this kind of mechanism is
comprised of a rigid coupler link which in another way to say bearing motion generated rigid link
mechanism. An inverse elastica problem is cited as example and solved by loop-closure equation,
applicable to flexible segments. Finally equation adopted an optimization scheme and cited with
numerical example.
Key Words: Complaint Mechanism, Multi Bar Mechanism, Complaint Four bar Mechanism, Rigid
coupler Link, Rigid Moving Link.

173
 RTMST
1 Introduction:
Kinematic chain mechanism has traditionally been designed to comprise of links that
behave as “rigid” members. Such mechanism derives their mobility entirely from rigid-
body translations and/or rotations of links due to degree-of-freedom (DOF) permitted at
the kinematic pairs or connections between various links. Elastic deformations of
materials have also been utilized to generate useful motions in numerous mechanisms for
certain special advantages, but until mid-1960s such flexure generated mobility was
largely confined to small angular rotations between stiff members by means of flexure
hinge a short and thin metallic strip or a small “necked” down region of a thick blank of
material-that provides rotational DOF similar to that at a conventional pin joint 1. As
opposed to such flexure at joints, generating mobility through elastic deformations of
links by replacing one or more links in a conventional kinematic chain with slender
flexible members was first suggested by Burns and crossley [2], and this resulted in a
special class of mechanism called flexible link mechanism. Early works on flexible link
mechanism consisted of four-bar chains consisting one or two flexible members. Since
the late 1980s the scope of mechanism utilizing flexure has broadened tremendously
embracing mechanism with a variety of complex topologies. Today, all mechanism that
are designed to drive mobility from elastic deformations in some elements a flexural
hinge and/or a relatively long flexible segments of a mechanism have to come to be
broadly referred to as complaint mechanism. For more information on the background of
complaint mechanisms interested readers may refer Midha [3], Ananthasuresh [4], and
Sagree [5].
This paper concerns a class of complaint mechanism featuring a closed-loop kinematic
chain, i.e., a four-bar, and synthesized for one of the three conventional tasks viz. path,
function, or motion generation. This class of mechanism was first investigated by Burns
and Crossley [2]. They approximate the motion of the tip of a flexible cantilever strip as
equivalent to the rotation of a rigid-link, that is five-sixth of the length of the flexible
segment, and based on this approximation, they presented a graphical technique for
synthesis of a function generating four-bar linkage with a flexible coupler link. Sevak and
McLarnan [6] presented an optimization method based on iterative finite-element
analysis technique for synthesis of flexible mechanism with one or two flexible links in a
four-bar chain for function and path generation tasks. Howell and Midha [7] developed a
technique pseudo-rigid body model wherein a complaint mechanism is modeled as an
equivalent rigid-link mechanism, and used this technique to synthesize four-bar
complaint mechanisms containing a rigid input link, a rigid coupler and a flexible link for
motion generation task. In all of these works, two characteristics are prominent:
complaint four-bar mechanism comprised at least one moving rigid link, and in complaint
mechanisms for motion generation, the coupler link was always considered the motion
generation objective that is based on the definition of the task provided for conventional
rigid-link mechanisms. In this paper, we considered a new objective involving the
guidance of a slender flexible segment rather than a rigid link complaint mechanism to
accomplish the objective task. In what follows, we first outline the problem statement and
follow it by a detailed explanation of the synthesis technique. Lastly, we illustrate the
synthesis approach by means of a numerical example.
174
 RTMST

2 Problem Statements:
Motion generation is one of the three customary tasks for kinematic synthesis of rigid-
link mechanism. Erdman and Sandor [8] have defined it is as the task which requires that
an entire rigid body be guided through a prescribed motion sequence. The “body” to be
guided is usually a part of a rigid floating link, and the prescribed “rigid-body motion
sequence” comprise of desired positions and orientations of the floating link. However, it
the body to be guided is flexible, and then it would undergo elastic deformation in
addition to rigid-body motion. Such a case has not been treated so far, and we considered
the same in this paper.
The objective of the synthesis presented in this paper is to guide a given slender flexible
segment with known initial shape, elastic properties, and external loading, at any, to
another prescribed configuration to another specified configuration as illustrated in Fig.
1.

Figure 1: Demonstration of compliant segment motion generation

Such a motion of the segment involves a rigid-body displacement of the segment along
with controlled elastic deformation or change in the shape of the segment. We call such a
task to accomplishing such a motion as “complaint-segment motion generation”, and in
general this task could include multiple prescribed configurations i.e. position and shapes
of the segment, just as in case of conventional “rigid-body motion generation” task. Thus,
complaint-segment motion generation task is task that requires a flexible segment of a
mechanism to be guided through a sequence of discrete prescribed “shape”, or “precision
shapes”.
Many potential application of complaint-segment motion generation can be envisioned.
For instance, a certain segment of a large flexible space structure that functions as a
reflective structure may be required to be oriented in different directions and also shape
into different curvatures for the purpose of modulating the characteristic of reflecting
sound or light waves. A similar application of complaint-segment motion generation is
also practicable at micro level, for example, in micro mirrors for controlled reflection of
light. Another example of potential application of complaint-segment motion generation
is a stamping application where a flexible contour is required to contorm to contoured
rigid surtaces that have differently shaped curvatures. In all such applications, the
175
 RTMST
required task can be efficiently accomplished by devising a suitable compliant four-bar
mechanism.
To accomplish complaint-segment motion generation, we considered a complaint four-
bar topology wherein the two ends of segment are connected to the ends of two other
flexible segments to be synthesized so that the three segments form one continuous
planar link. The two tree ends the link are pinned to ground and actuated by input torques
or rotations as shown in Fig.2. Such a topology may also be referred to as a structurally
binary link with three compound segments in the nomenclature suggested by Midha et al
[9]. In this figure, B-C is the given flexible segment to be guided to the specified final
configuration b-c; A-B and C-D are two segment to be synthesized; A-B-C-D represent
the initial configuration of the mechanism, and A-B-C-D represents the final
configuration of the mechanism. Based on the consideration of this topology, the problem
statement may be explicitly phrased as follows: given the initial configuration, synthesize
a complaint four-bar mechanism comprising all flexible segments and the corresponding
input torques that precisely effect the prescribed complaint-segment motion generation.
We assume that the specified initial and final shapes of the given flexible segment are
“smooth”, and can be adequately represented by second or third order polynomials. We
also assume that the involved shape change is “small”, which means the displacements
and strains are “small” enough for the application of linearized beam theory, and the
beams are “slender” so that axial and shear strains are negligible.

3 Synthesis Procedure:
The basic goal of the synthesis for compliant-segment motion generation can be
separated into two components: the elastic deformation and the rigid-body displacement
of the segment. The first component necessitates an appropriate moment distribution over
the given flexible segment the second component necessitates appropriate kinematic
displacements of the two end of the segments. Since the geometry of the shape change
involve can be represented by either a cubic or a quadratic equation, it is apparent from
the Euler’s elastic curve equation which relates the bending moment distribution on a
segment to its curvature, that the order of the moment distribution function on the
segment has to be at most one. That is, in order to effect quadratic or cubic shape
changes, it is sufficient to generate a linearly varying or constant moment distribution
along the length of the segment, and this in turn can be accomplished by means of only
two unequal or equal end-moments at the two ends of the segment. Based on this
fundamental concept, the synthesis strategy aims to accomplish both the required moment
distribution end-moments on the given segment and the prescribed kinematic
displacements of its two ends by appropriately synthesizing the two segments pinned to
the ground and corresponding input moments. Thus, specifically, the synthesis tasks are:
to determine optimal shapes and sizes of segments A-B and C-D, locations of the pivots
A and D, and the input torques MAin and MDin of the complaint mechanism in Fig.2

176
 RTMST

Figure 2: Demonstration of action of moments on movable links

The approach adopted for this synthesis is conceptually similar to that of rigid-link four-
bar mechanisms in that it involves disjoining the mechanism into various links and
designing each link separately, However, its implementation is very much modified in
view of the fact that pure kinematic motion and elastic deformations cannot be treated
separately in complaint mechanisms. The synthesis strategy is to disjoin all the segments
and impose boundary conditions on each segment so that at the fusing ends,
displacements and rotations are identical and the internal forces and moments are equal in
magnitude but opposite in sign, and each segment is independently in equilibrium. These
two conditions ensure that when the segments are joined, a/ all the internal forces and
moments cancel each other, b/ displacements and rotations are compatible at the fusing
ends, and c/ the resulting structures will be in equilibrium under the action of input
actuation moments. This approach is illustrated in Fig.3.

Figure 3: Limiting conditions for dimensional synthesis of segment

This process of setting up of boundary conditions being with determination of forces and
moments as well as corresponding rotations at the ends “B” and “C” of the coupler

177
 RTMST
segment “B-C” that is to be guided. These loads and the prescribed kinematic
displacements of the ends “B” and “C” are translated into boundary conditions for the
two input segments “A-B” and “C-D”. Thus, the synthesis strategy can be divided into
two main steps: computation of end-loads on the coupler segment, and synthesis of the
input segment.

3.1 Computation of End-Loads on the Coupler Segment

The loads forces and moments at the ends of the floating segment corresponding to the
desired shape change in the segment are determined by the standard method of
mechanics. For this purpose, the rotations at the ends are first calculated from the
equations of the known initial and final deformed configurations; then, the moments
corresponding to the end rotations are computed using the basic slope-deflection equation
of the moment distribution method is not commonly used in modern structural analysis
literature, its derivation, taken from [10], is presented below for the benefit of the readers.
For ease of illustration of the method, and without any loss of generality, let the initial
shape of the member be straight. Figure 4 represents the deflected shape n-f of a straight
member N-F. Clockwise moment or rotation of

Figure 4: a. Positive directions of end-rotation b. End moments caused by a unit rotation

c. End moments caused by a unit rotation and different end

either end of the member is considered positive. The relative translation of the ends
perpendicular to the original direction of the member, (yF, yN), produces bending and the
relative translation along the axis of the member, x, is considered to be zero, that is, it is
178
 RTMST
considered that no change in length occurs. The chord rotation, (yF yN)/l, is considered
positive when clockwise. Further notation is as follows:
N: rotation at the end N near end
F: rotation at the end F far end
SN : rotational stiffness of end N that is, the end-moment at N corresponding to a unit
rotation at N while the displacement at F is restrained. Fig. 4-b
SF: rotational stiffness of the end F that is, the end-moment at F corresponding to a unit
rotation at F while the displacement at N is restrained. Fig. 4-c
t: carryover moment that is, the moment at the fixed end F caused by a unit rotation of
end N Fig. 4-b; also equal to the carry-over moment at the fixed end N caused by a unit
rotation of end F Fig. 4-c by Betti’s theorem.
Consider the curve n-f representing the deformed shape of a member N-F subjected to
external later loads Fig. 4-a. The translation part of the displacement of the member that
is, its translation part of the displacement of the member that is, and its translation as a
rigid-body to the straight dotted line n-f produces no moments. The end-moment MN at
end N can be expressed as the sum of the moment due to the external load on the member
with the end displacements prevented the fixed-end moment Mext and of the moments
induced by rotations (N) and (F) at the ends N and F respectively. Thus, MN SN(N)
tF(SNt)(Mext)N where (Mext)N is the fixed end moment at N, that is, the value of the end-
moment caused by the actual external lateral loads on the beam with displacements at
both ends prevented. This equation can be rearranged as MNSNNtF(SNt) (M ext)N . This is
the slope-deflection equation for a prismatic or non-prismatic beam. The equation can be
used to express the moment at the left or right hand end. When the member has a constant
flexural rigidity EI, S B 4EI/l and t2EI/l. the slope deflection equation then becomes:

MN = 4 +2 − 6∅ + ( ) …(1)

Using this slope-moment equation, moments required at the ends B and C of the segment
B-C in Fig.3 are determined. Once the end moments have been established, the forces
required for the equilibrium of the segment at each of the two ends are determined from
simple statics, and then, the computed force at each end may be resolved into horizontal
and vertical components through the known angle of orientation of the chord with respect
to the horizontal.

3.2 Synthesis of input Segments

This synthesis process begins with setting up the right boundary conditions on the
segments With reference to Fig.3, the boundary conditions on the segments A-B and C-D
are set-up as follows: the forces and moments determined at the ends B and C of segment
B-C are transferred to the ends B and C of the segments A-B and C-D as explained
179
 RTMST
earlier; the rotations required at the ends B and c are computed from the known change in
curvature of the segment B-C; and the displacements of the ends B and C are known
directly from the design specifications. The rotations and input moments at the ends A
and D are also part of the boundary conditions for the synthesis, but their values are
unknown. After establishing the boundary conditions, the objective of the synthesis is to
determine the shapes and sizes of segments A-B and C-D, locations of the pivots A and
D, and the unknown values of the input torques MA in and MD in which satisfy the
equilibrium of the segments. Since the synthesis process is identical for both the segment
A-B and C-D, we consider only one segment in the following.
Consider the segment A-B, for instance Fig.5 illustrates the Known and unknown design
parameters involved in the synthesis of segment A-B. The forces FBx, FBy, the moment
MB, the displacement d at an angle from the horizontal, and the rotation of B at the tip of
B are all known; the unknown quantities are the length and the shape of the segment, the
location of the ground point A, and the input moment Main required at the tip A. All
these known and unknown quantities are related through the equilibrium condition for the
segment, and constitute and an inverse elastic problem as discussed in the following.

Figure 5: Design parameters involved in synthesis of the segment

3.2.1 An atypical Inverse Elastic Problem

The most common problem in the theory of bending is that in which the free shape is
sought. This problem is rather straight-forward to solve. If the expected deflections are
small, the problem can be easily solved by the Euler-Bernoulli linear beam theory; but in
the case of large, non-linear deflections a more sophisticated model such as the elastic is
necessary. However, in many engineering applications there exist problems which are the
inverse of the direct bending problem. One class of such inverse problems involves the
case where the length of the beam and locations of its tips and/or tip loads are known, but

180
 RTMST
the equilibrium geometry of the beam is unknown. A problem comprising this particular
set of known’s and unknowns in the theory of bending is called inverse elastica problem.
Such a problem has been studied by many researchers including Shop and McLarnan[11],
Watson and wang [12], and stack et al[13]. Another class of inverse problems involves
the case where the deflected shape of a beam corresponding to a particular load may be
given and the free unloaded shape of the beam is sought [14]. Expecting the cases where
the beam geometry and loads are very simple, such inverse problems have, in general, no
closed form solutions, and mostly, they required numerical methods for their solutions.
However, as was observed in the works cited, numerical solutions to inverse elastic
problems are complicated by the highly non-linear behavior and non uniqueness of the
solutions.
Typically, in such inverse problems either the initial shape or the deformed shape of the
beam, the locations of tips of the beam, as well as the total length of the beam are known.
However, in the inverse problem under inquiry problem illustrated in Fig.5, both the
initial shape and the deformed shape are not known, nor are the tip locations and length
of the beam known. Further, in this problem, unlike typical inverse problems, both the tip
loads and the tip displacements including rotations are known. This set of Known and
unknowns in elastic bending theory constitute a very atypical inverse problem that has
not been reported in the literature, and the solvability of this problem is discussed below.
Consider the bending of the curved beam segment in Fig.5. The shapes of beam’s axis in
the unloaded undeformed and the loaded deformed states are planar curves represented
by parametric equations (s) and (s) respectively, where s is a parameter identified with
length measured along the axis of the beam from point A. The beam is assumed to be
inextensible, hence s is the same in both equations, and the total arc length of both shapes
is L. Thus, (s) and (s) represented the slopes measured with respect to the horizontal, and
x(s), y(s) and x(s), y(s) represent the Cartesian coordinates at any point s on the
undeformed and the deformed shapes of the beam. It is assumed that one of the principal
axes of inertia of the beam remains to be a planar curve after deformation. The required
forces and the moment at the tip B are known from the boundary conditions at the end B;
this implies that the forces at the end A are also known since they must be equal and
opposite to that at B to satisfy the equilibrium condition. However, the input moments
MAin in unknown since the relative locations of the tip are not known. As assuming a
constant flexural rigidity EI, the in-plane bending of the beam is governed by Euler’s
equation:

− = − ( )+ ( )… (2)

( ) = cos( ( ))

( )
= sin( ( ))

( )
= cos( ( ))
181
 RTMST
( )
= sin( ( ))… (3)

The loaded beam is subject to the following displacement bound- ary conditions at the
end B:

( )− ( ) = ∫ cos[( ( ) − cos ( ) ] = cos( )… (4a)

( )− ( ) = ∫ sin( ( ) − sin( ( )) = sin ( )… (4b)

( )− ( ) = … (4c)

The above set of non-linear equations forms an under constrained system for the solution
of the unknowns EI, (s), (s), x(L), y(L), MAin, and L. One might think of obtaining the
loaded shape of the beam segment, (s), by an inverse elastic method, and then solve for
(s) in the governing equilibrium Eq. 1. However, expression for (s) obtained by
rearraging Eq. 1 has, in general, no closed-form solution. The insufficient information
about the beam to be synthesized, or in other words, this set of large number of unknowns
poses difficulties even for a solution by numerical methods. The set of unknowns in this
system of equations exhibit the existence of infinitely many beam configurations that can
satisfy the load-displacement requirements. However, it is possible to obtain uniquely
“best” possible solutions for such problems by employing constrained optimization
techniques. There is no reported work in literature that has particularly treated the above
set of knowns and unknowns. One closely related work is that of banichuk [15], who
presented an optimization method for determining the optimal shape of a curved beam
loaded by a single concentrated force at the tip of the beam. However, in that problem,
the locations of the beam’s tips, the length and the total volume of the beam were
assumed to be known, and also in that problem the tip deflection was minimized rather
than satisfying a specific value of the tip deflection as is the case in the current problem.
The difficulty associated with insufficient information for a direct solution to the above
set of non-linear equations has been overcome in the following by combining kinematics
and mechanics in a novel way through an optimization scheme. Before elucidating the
optimization scheme, the details of a model for the beam used in the optimization set-up
is in order below.
3.2.2 Modeling a curved Beam Segment
Since differential equations are cumbersome for optimization by “classical” approaches,
a method of discretization of the flexible segment into discrete elements is necessary for
optimization by a numerical approach. For this purpose, a curved flexible member may
be modeled as a series of small, straight, rigid elements connected end to end through
linear torsional springs such that the elements of the

182
 RTMST

Figure 6: Model of curved beam using torsional spring and rigid elements

rigid elements add up to the length of the original curved beam segment see Fig. 6. The
spring constant, K, of the torsional spring is approximately taken as K EI/L, where EI is
the bending stiffness of the beam, and l is the length of the rigid element. Such a model
for a curved beam segment characterizes the static deformation behavior of the
corresponding beam segment with sufficient accuracy when large number of springs and
rigid elements are used in the model. For example,

3.3 Optimization
Figure 7 shows the spring-link models of the initial and the final configurations of the
beam segment. The angels I and I in Fig. 7 are related to the tip loads, the input actuation
rotation A and the boundary-conditions through a simple equilibrium equation in terms of
spring constant K. To arrive at an optimal feasible solution to the problem, the total
weight or volume of the segment is minimized as a cost functional subjected to the
known tip displacement kinematic constraint and the known tip rotation elastic constraint.
The design variables and the associated bounds are: 011 max, 01 min, cross-sectional
width of the member and min at max input rotation at the pinned end A. Thus, the
optimization problem is proposed as:

Minimizing case:

183
 RTMST

Figure 7: Computational scheme of finite link model

Subjected to: Kinematic Constraint

Elastic Constraint
Equilibrium Equation … (5)
Equilibrium Equation: the equilibrium equation is given by:
[( − )−( − )] = … (6)

i= 2..n

Where Mi is the net moment action on ith element;

Elastic Constraint: This is the required rotation at tip B
( − )= … (7)

Kinematic Constraint: For the kinematic constraint an idea similar to the standard loop-
closure equation employed in the synthesis of traditional rigid-link mechanism is used.
By representing the position of each of the rigid links in its initial and final positions as
vectors see Fig.7: Zi =l(cos ), and Wi =l(sin ),respectively, and representing the
specified tip displacement as vector d, a loop-closure equation for the two positions can
be written as:
∑ ( − ) = … (8)

The two scalar components of this vector equation are:

184
 RTMST

(cos − cos ) = cos

∑ (sin − sin ) = sin … (9)

Where is the orientation of the displacement vector d as shown in Fig.7. The optimization
is started with initial “guess” values for the design variables, and at convergence, it yields
a set of optimal values for l, and i, (i = 1…n) that defines the shape, the length the ground
pivot of the flexible segment, as well as, the magnitude of the input actuation moment
MAin, or rotation A. The accuracy with which the converged values will satisfy the
synthesis requirements depends on the convergence tolerance set for the optimization and
number of the elements employed in the model.
The above steps must be repeated to synthesize the remaining segment C-D. Then, the
shapes of all three complaint segments, A-B, B-C, and C-D will b known. The structure
that results by connecting the three segments end-to-end and pinning the ends A and D to
ground pivots is the required complaint mechanism for the specified task.

4 Example
Consider a 200 mm long flexible beam BC which is initially straight and oriented at
14.47O with respect to a horizontal axis of a Cartesian coordinate system is required to be
guided into a new position and bent into a shape where the axis of the beam, identified by
a parameter s, is represented by the cubic equation: 0.8 10 5s3 0.00235s2 0.1s when the
end B is at the origin of the Cartesian frame see Fig.8. Let the required displacements at
the two end be: db 10mm at an angle B 2000 and dc 10mm at an angle c 2000. Let the
specified properties of the beam be: Young’s modulus, E 20 104 N/mm2, and width of its
square cross-section, 5mm (I 52.08 mm4).

Figure 8: Schematic illustration of design specimen

185
 RTMST

Figure 9: Schematic demonstration of computed ends, and end forces

First, the required rotations at the ends are computed from the known change in
geometry: B 0.1 radians and c 0.08 radians. The chord rotation and the length of the beam
are given, viz. 14.470 and L 200 mm, and since no external load is specified on the beam,
we have Mext 0. By substituting the known values viz. E,I,l, B, C, and Mext, in Eq. 1, the
necessary end –moments to effect the desired deformation are computed: MB 29166.5 N-
mm, and Mc 27083.16 N-mm. Then, applying the basic
moment equilibrium principle from statics, the end-forces at the two ends of the segment
BC are obtained, and resolved into horizontal and vertical components in the Cartesian
coordinate
system: FBX 70.31N, FBY 272.32 N, and FCX 70.31 N, FCY 272.32N see Fig.9.
Next, the complaint segments AB and CD are synthesized by applying the optimization
scheme described earlier. Consider the synthesis of segment AB. The compatible loads at
the segment BC and the compatible displacement and rotation at the tip B are known
from the problem specifications. That is, the boundary conditions FBX 70.31N, FBY
272.317 N, and MB 29166.5 N-mm, d BX 9.3 mm, dBY 3.4mm, and B 0.1 radians are
enforced at the tip B. The curved beam with end A pinned to the ground is modeled using
spring-link elements as shown in Fig.8. Choosing the following bounds for the design
variables: square cross-section width, 10 mm h 50 mm; length of each element, 0 mm L
100mm; and initial orientation of the elements , O I radians, the optimization problem
was set-up and solved in the MATLAB® software package for various number of
elements n in the model. The optimized values of the set l and i,( i= 1..n) were used
to construct the shape for segment AB.

5 Conclusions:
A problem is effective for prescribing shape change for average flexible segment along
with, rigid body motion which is said to be compliant segment motion. Synthesis of
closed loop compliant mechanism is illustrated through a numerical example and the
feasibility of the approach is verified through reversal finite element analysis. Multi
configurations are given for analysis of segment motion which is depicting elaborately
four-bar mechanism which is finally reduced to compliant segment generation motion.

186
 RTMST

References
[1] Paros, J.M.,and Weisbord, L., “How to Design Flexure Hinges”, Machine
Design, Nov. 25, 1965, pp. 151-156
[2] Burns, R.h., and Crossley, F.R.E., 1968, “Kinematic Synthesis of Flexible
Link Mechanisms”, ASME Paper No. 68-MECH-36.
[3] Midha, A., 1993, “Chapter 9: Elastic Mechanism”, Modern Kinematics-
The Devlopments in the Last Forty Years, A.G. Erdman Ed., Jhon Wiely
and Sons Inc.
[4] Ananthasuresh, G.K., 1994, “A New Design Paradigm in Micro-
Electromechanical Systems & Investigations on Compliant Mechanism”,
Ph.D. Dissertation, University of Michigan.
[5] Saggere, L., 1998, “Static Shape Control of Smart Structure: A new
Approach Utilizing Compliant Mechanism”, Ph.D. Dissertation,
University of Michigan.
[6] Sevak, N.M., and McLarnan, C.W., 1974, “Optimal Synthesis of Flexible
Link mechanisms with Large Static Deflection”, ASME Paper No. 74-
DET- 83. Also published in J. Eng. Ind., May 1975, pp. 520-526.
[7] Howell, L. L., and Midha, A., 1994, “A Loop-Closure Theory for the
Analysis and Synthesis of Compliant Mechanism”, ASME J. Mech. Des.,
118, No. 1, pp. 121-125.
[8] Erdman, A.G., and Sandor, G.N., 1993, Mechanism Design: Analysis and
Synthesis, Vol. 1, Rrentice Hall, Inc., Englewood Cliffs, new Jersey.
[9] Midha, A., Norton, T.W., and Howell, L.L., 1994, “On the Nomenclature,
Classification, and Abstractions of Compliant Mechanisms”, ASME J.
Mech. Des., 116, No.1 pp. 270-279
[10] Ghali, A., and Nevile, A.M., 1978, Structural Analysis, Chapman and
Hall Ltd., London.
[11] Shoup, T. E., and McLarnan, C. W., 1971, “On the Use of the
Undulating Elastica for the Analysis of Flexible Link Mechnisms”,
ASME J. Ind., 93, Feb. pp. 263-267
[12] Watson, L.T., and wang, C.Y., 1981, “A Homotopy Method Applied
to Elastica Problems”, Int. J. Solids Struct., 17, pp. 29-37.
[13] Stack, K.D., Benson, R.C., and Diehl,T., 1994, “The Inverse Elastica
problem and its Application to Media Handling”, Proceeding of the 1994
International Mechanical Engineering congress and Exposition: Inverse
problems in Mechanics, Proc. R. Soc. London, Ser. A, ASME, AMD-Vol
186, pp. 31-36
[14] Frisch-Fay, R., 1962, Flexible Bars, Butterworths, Washington.
[15] Banichuk, N.V., 1990, Chapter 7, Introduction to Optimization of
Structures, Springer-Verlag, NY.

187
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18 th and 19th April, 2013, NITTTR, Kolkata

Automatic Shape Recognition by Using

Robot Vision Technique

Sudip Kumar MONDAL, Mandakinee BANDYOPADHYAY,

Amit Kumar SHAH, Sagarika PAL

Department of Electrical Engineering

National Institute of Technical Teachers’ Training and Research, Kolkata
[Under MHRD, Govt. of India]
Block – FC, Sector –III, Salt Lake City, Kolkata -700 106
Tel.: +91 33 23374125/0479, Fax: +91 33 23376331
E-mail: sudip.km.89@gmail.com; mom_saltlake@yahoo.co.in;
amitshah.darj@gmail.com; spal922@yahoo.co.in

Abstract: Machine vision is an integral part of any robotic environment. The purpose of
this paper is to establish a method of shape recognition through machine vision technique
using one commonly used webcam and grasping of unknown objects using a robotic arm
with gripper and place the objects in different places according to their shape. The shape
recognition algorithm has been developed and solved by MATLAB software in a
computer and interfaced to a flexible robotic test bed. The shape recognition algorithm
has been used to obtain object identification through position and orientation recognition.
This position and orientation data are then passed to the robot controllers by using a
programmable microcontroller to perform the sorting operation. The shape recognition
robot is a pick and place robot, which is of low cost and automatic. Through the shape
recognition capability it can identify and grasp the desired object. The image processing
technique has been applied to the images to make recognition possible. This includes the
extraction of shapes features from images of the objects and the creation of descriptors
that overcome difficulties of affine transformations. The present article gives an in-depth
account about the desired output and system response. This system can be used in micro-
188
 RTMST
manufacturing industry or mechanical system industry where shape recognition is one of
the important purposes to solve different problems. The experimental result of the present
work gives a brief idea about to create a new approach for object recognition according to
shape which can be a good basis for further development in the manufacturing industry.

Keywords: Shape recognition, Real-time image processing, MATLAB, Webcam, Robot,

Manufacturing industry

1. INTRODUCTION
Human being can easily identify objects in our surroundings, regardless of their
circumstances, whether they are upside down, different of color or texture, partly
occluded, etc. Even objects that appear in many different forms, like vases, or objects that
are subject to considerable shape deviations, such as trees, can easily be generalized by
our brain to one kind of object. Many different theories have been developed on how
humans perform object recognition. Most of them are based on logical reasoning and on
clear abstractions, and sound very plausible. Yet to go from human object recognition to
computerized object recognition is a large step. The theories of human object recognition
often do not touch on the lower levels of vision processing, i.e. they may assume that
extracting an object from its surroundings is a given and that decomposition of this object
into different entities is something that happens naturally.

In the robotic machine vision, the shape is essential information in the object recognition
and cognition process. Many real world applications require real-time Shape detection
and recognition of a moving object. Object detection and recognition according to their
shape is one of the most challenging problems in robotic machine vision; as the
performance of shape recognition system should be fast enough so that moving objects in
video can be detected and processed in real time.

In this thesis we will make an attempt to go all the way from the lowest layers of
processing to allowing the computer to generalize different views of the same object into
one category. The emphasis will be on the machine vision technique as this is an integral
part of any robotic environment. The machine vision technique is under the field of
Digital Image Processing. Digital Image Processing [11] means processing of digital
image on digital hardware usually on computer. Digital image processing is the use of
computer algorithms to perform image processing on digital images. As a subcategory or
field of digital signal processing [10], digital image processing has many advantages over
analog image processing. It allows a much wider range of algorithms to be applied to the
input data and can avoid problems such as the build-up of noise and signal distortion
during processing. Since images are defined over two dimensions (perhaps more) digital
image processing may be modeled in the form of Multidimensional Systems. Digital
image processing allows the use of much more complex algorithms for image processing,
189
 RTMST
and hence, can offer both more sophisticated performance at simple tasks, and the
implementation of methods which would be impossible by analog means.

In the shape detection of image, the shape of objects is described through the edge
characteristic and the object we studied is usually regarded as a close curve. Most
researchers used edge information to recognize the shape. For example, Hankyu Moon,
Ramalingam Chellappa and Azriel Rosenfeld [2] proposed an approach to accurately
detecting 2-D shapes by defining an optimal 1-D operator for shape detection by
extending the DODE filter along the shape’s boundary contour; Min Min Kyaw, Syed
Khaleel Ahmed and Zainul Abidin Md Sharrif [5] addressed one way to identify
agricultural products based on their shapes by feature extraction and edge detection by
using MATLAB and Support Vector Machine (SVM), Chun-Chieh Wang, Kuo-Lan Su
and Chih-Teng Shen [6] developed a Tour Guide Robot via Shape Recognition and Path
Planning. However, there are some difficulties in picking up an integrated and well
closed edge of the object because of the influence of noise, shadow, shelter or other
factors. So Firdaus Zakaria, Hoo Seng Choon, and Shahrel Azmin Suandi [9] proposed a
shape recognition method by thresholding, median filtering and thinning alongwith edge
detection using Sobel operator. Several other techniques on shape-based identifications
have been developed such as Hough Transform [1] [7], volume intersection, shape
measurements, scaling and rotating of a standard image. Sugiyama Seiji, Tsuneo
Yoshikawa, Masanao Koeda and Hiroshi Fujimoto [4] used a simple 2-D version of the
visual volume intersection method for shape recognition. Kitti Suwanratchatamanee,
Mitsuharu Matsumoto, Ryo Saegusa and Shuji Hashimoto [3] proposed a simple tactile
sensing and control method to realize an effective robot controller for the 3-D object
handling. The Hough transform is a popular arithmetic solution technique which has a
better way in dealing with the problems such as noise, shadow, or the shelter of the
object. Also it has been widely used in the object shape detection. JI Yuan, MAO Li,
HUANG Qingqin and GAO Yan [8] proposed a shape recognition method using fuzzy
set theory and generalized Hough Transform. But if the noise is too strong or the
background of the object is too complex, the traditional Generalized Hough Transform
also may not vote to a focus.

In this paper we have carried out of identifying different kinds of shapes, like sphere,
cone, pyramid, cylinder, prism etc. from a conveyor taking one by one by a robotic arm.
The whole process is done by Automatic shape detection technology through machine
vision system.

2. METHODOLOGY

Most image-processing techniques involve treating the image as a two-dimensional (2-D)

signal and applying standard signal-processing techniques to it. In this work also the real
world 3-D shapes are converted into 2-D image form for image processing purpose. The
video captured by the camera is being processed by the MATLAB program that helps in

190
 RTMST
shape detection and color recognition. The steps of shape detection process using
webcam and MATLAB is shown below using a flow diagram in Fig.1:

Fig. 1. Steps of shape detection using camera and MATLAB

2.1. Software Development for Shape Recognition

Recognition of an item involves three levels of processing: data collection, feature

extraction and classification. The data is the set of images of the objects to be identified.
All the images are collected using a digital camera and loaded into MATLAB
individually to extract their features. Feature extraction is based on the edge detection
method. Using this method, we extract the image parameters; i.e., area, major-axis length,
minor-axis length and perimeter. These together with two other dimensionless quantities
form the set of features used. The results of this processing can be used in numerous
security applications such as intrusion detection, Spy robots, fire alarm, person finder etc.
In the present work, MATLAB R2010a platform containing Image Processing Toolbox
and Image Acqision Toolbox Software is used to detect the shape of an object. A brief
description of shape recognition process through the MATLAB software is described
below:
2.1.1. Receiving the image input in MATLAB

To perform the experimental task, at first the webcam is connected to the computer using
the webcam driver. After the connection is made, the interfacing between the webcam
and the MATLAB environment is done. For MATLAB to recognize the video camera,
the camera is created as an object using the command obj=videoinput (`winvideo').

2.1.2. Starting the video input object

Acquire an object by calling the start ( ) function. For example, starting an object locks
the values of certain object properties. Starting an object does not initiate the acquiring of
image frames, however. The initiation of data logging depends on the execution of a
191
 RTMST
trigger. Start ( ) function is called to start the video input object. Objects stop when they
have acquired the requested number of frames. Because in this paper we use continuous
acquisition of video, so the stop ( ) function must be called to stop the object.

2.1.3. Triggering the acquisition

To acquire data, a video input object must execute a trigger. Triggers can occur in several
ways, depending on how the Trigger type property is configured. For example, if an
immediate trigger is specified, the object executes a trigger automatically, immediately
after it starts. If you specify a manual trigger, the object waits for a call to the trigger
function before it initiates data acquisition.

2.1.4. Bringing data into the MATLAB workspace

The toolbox stores acquired data in a memory buffer, a disk file, or both, depending on
the value of the video input object Logging Mode property. To work with this data, it
must be brought into the MATLAB workspace. To bring multiple frames into the
workspace, the getdata ( ) function is used. Once the data is in the MATLAB workspace,
it is ready to be processed and can be manipulated as any other data.

2.1.5. Binary Image Conversion

The shape detection algorithm is applied on the image. By using the MATLAB program,
at first the color image is converted into a binary or black and white image. In the binary
image, the background is black colored and the shapes are white colored. From the binary
image, the edges of the shapes can be detected.

2.1.6. Edge detection

Now, the edge-detection operators, like Sobel, Prewitt, Log, Canny, Roberts, Zero-cross
etc. are applied on the binary image for detecting the edges. The filtering of is done to
eliminate the noises and smoothing the edges. After edge detection, we can get different
shapes with clear-cut edges.

2.1.7. Shape Recognition

Now the area, perimeter, major-axis length, minor-axis length of the obtained shapes are
calculated; and compared with the pre-defined shape parameters and templates. If the
parameters match together, then the shape is identified. For example, if the object is a
spherical one, then its 2-D image will be a circle, and will match the pre-defined
functions and template set for the circle. Thus the object can be identified as a sphere.
Recognition of a circular shape using our shape detection algorithm is shown in Fig.2:

192
 RTMST

Fig. 2. Converting image into binary form, shape detection and showing result in original
image form

3. SYSTEM OVERVIEW
The hardware system consists of the following parts: 1. A robot with a robotic arm and a
gripper with two jaws, 2. A robot controller unit with a programmable microcontroller
and a motor driver circuit, 3. An Omni directional camera attached to the robotic hand
and 4. A Conveyor, 5. Sorting Stations and 6.one Computer for control and data
processing of the fingers and camera. Each hardware module in the system will be
described in the following subsections.

3.1. Robotic Arm with Gripper

The robotic hand which is developed has 3-degrees of freedom. Each joint is connected
with a stepper motor (360 degree, 200 steps, 1.8 ̊ per step). The first joint axis and the
other two joint axes are orthogonal. Hence the fingertip can make three dimensional
motions. The end effecter of the robotic arm is a mechanical gripper. The gripper
contains a gear-pinion mechanism with two jaws attached to two gears, so that it can
easily grasp an object. A stepper motor is connected with the gripper to control the
movement of it. The schematic view of the robotic arm is shown in Fig. 3:

193
 RTMST

Fig.3. Robotic arm with gripper

3.2. Controller Unit

Controller unit consists of two controlling unit such as: 1. Programmable Microcontroller
and 2. Stepper motor driver circuit. Microcontroller used in this system is 89C51. It is a
40 pin microcontroller and it can interface with system software through pin RxD and
TxD Microcontroller can be connected with the stepper motor driver through pin P.0-
P.7.
The stepper motor driver unit consists of IC 4027, which can be used to control the
unipolar stepper motor having 4 coils. The stepper controller circuit can drive for a motor
current of up to about 500 mAmp/Winding by suitable heat sinks for the SL-100. It
consists of 4 power transistors (2N3055). Every time in move the motor put signal the
STEP line of the IC is pulsed, the motor moves one step (i.e. 1.8 ̊).

3.3. Omni-directional Camera

The camera used in the system 2 is a simple webcam ENTER E-12MPIR (12 megapixel).
It can capture the visual information on 360 degree of circumferences around the camera
axis in one image. This camera is attached to the base of the robotic hand. The webcam
is connected to the computer through the USB port using the webcam driver software. It
captures the real-time image of the object and sends that to MATLAB workspace for
further processing.

3.4. Conveyor

There is a conveyor in the system, which is used to present the objects one by one in front
the robot arm. The movement of the conveyor is controlled using a DC gear motor. The
programming is done in such a way that, when an object comes just in front of the
webcam connected to the robot arm, the conveyor stops, shape recognition and pick and
place operation is done, and then conveyor movement starts again.

194
 RTMST
3.5. Sorting Stations

There are three different sorting stations in the system namely, ‘sphere’, ‘cube’, and
‘pyramid’ for different shaped objects. After the shape recognition, the object is picked
by the robot arm and placed in its desired station. If the object shape does not match with
any of the pre-defined shapes, then it is rejected and not placed in any of the stations.

The hardware model of the system is shown in Fig. 4:

Fig. 4. Hardware Model of the shape sorting system

195
 RTMST
4. FLOWCHART OF THE SYSTEM:

Fig.5. Flowchart of Shape Detection Method

196
 RTMST
5. EXPERIMENTAL RESULTS

The total experiment is done by MATLAB R2010a. For recognizing the shape of the
unknown object, robot will capture a picture of the object which is to be moved by the
omni directional camera attached to the arm. Afterwards it will move the picture to the
computer connected with it and check for whether the shape is predefined or not. If the
shape is predefined then computer will send a signal to the robot. Then the robot will
pick-up the object and keeps it into the particular box. When the shape is not predefined
the robotic arm would not pick-up the object and the conveyor belt will be going on. This
dataset can be divided into four groups which are dataset that contains only one object,
three same objects, three different objects and multiple different objects.

The experimental result is shown below:

Fig.6.1. Detection of various shapes in MATLAB software

197
 RTMST

Fig.6.2(a). One object Fig.6.2(b). Three same object

Fig.6.2(c). Three different objects Fig.6.2(d). Multiple different objects

Fig.6.3(a). Input image Fig.6.3(b). Circle

detection

198
 RTMST

Fig.6.3(c). Square detection Fig.6.3(d). Triangle detection

Fig.6.4(a). Input image Fig.6.4(b). Incorrect detection of circles

TABLE I: PROPOSED METHOD ACCURACY

Object Object No. of Successful

Accuracy
Sample Shape Objects Detection

Circle 3 3 100%
One Object Square 7 7 100%
Triangle 4 4 100%

Circle 6 3 50%
Three Same
Square 9 7 77.8%
Objects
Triangle 6 4 66.7%
199
 RTMST

Three Circle
Different Square 18 18 100%
Objects Triangle

Multiple Circle
Different Square 23 19 82.6%
Objects Triangle

6. Discussion on Experimental Work

The various examples of shape detection through our proposed method are shown in the
above figures. Fig. 6.3(b) – 6.3(d) shows the example of successful detection by using the
proposed method. Fig. 6.4(b) demonstrates the example of incorrect detection of the
proposed method. From Table I, we can see that, if there is only one object, then our
algorithm can give 100% of successful detection. But in case of three same objects, the
accuracy reduces. Likewise, if there are three different shaped objects, then it can give
100%. But for multiple different objects, the accuracy reduces to 73.9%.

There are several reasons why the proposed method produced undesirable detection:
a) Due to the input image has uneven intensity, the image is not threshold properly
and thus the shapes cannot be detected.
b) Some of the objects are touching each other which contribute to inaccurate
calculation in the parameter and area estimation.
c) Noises not totally eliminated where these noises will be detected as objects.

The difficulties of object recognition are extensive. Shapes can differ in appearance for
several reasons. The most important reason is the difference in perspective we can have
on a shape, i.e. shapes can be viewed from different angles and positions possibly making
the shape appear bigger, upside down, tilted etc. In this section we will give a quick
overview of the most important difficulties we have to deal with when working with the
visual aspects of object recognition, as shown in Fig.7:

200
 RTMST

Fig.7. Various difficulties in recognition of an object for a simple geometric shape

7. CONCLUSIONS
A method of shape recognition and grasping of unknown objects by a robotic hand has
been proposed in this paper. As being described above, this paper proposed a method for
shape recognition especially for object on the conveyor with simple algorithm with low
computational time. Its main objective is to differentiate basic shape such as circle,
square and triangle in the given input image by using computer vision technique. The
Shape recognition is achieved using an omnidirectional webcam and MATLAB 2010a
software environment. The Hough transformation is used for shape recognition. After the
recognition, a control signal will be sent to the robot arm via the controller unit. The arm
will then move accordingly to different angles to place the object in desired place or
station. We have presented an approach to detecting 2-D shapes. The scheme for shape
detection has been designed from the lowest level of edge detection to guarantee the
detection of shapes and accurate estimation of geometric parameters. From results, we
can see that the accuracy of our system is more than 80%.

8. FUTURE WORK

With our application we have shown some possibilities of using artificial techniques to
perform object recognition in the real world. It should be noted that our approach at the
moment just looks at a cross section of the object, so it may run into the object as it goes
to grasp it since the arm approaches from above. As the algorithm is for 2-D shapes like
circle, triangle, rectangle etc, so that we are converting the 3-D shapes into 2-D images
for image processing purpose. A problem may arise if we try to differentiate between
shapes like sphere, cone and cylinder. As we are taking the top-view image of the object
using the webcam, our algorithm cannot differentiate between these objects, as the top-
view of a sphere, a cone and a cylinder all are seemed to be like a circle. In future, the
algorithm can be developed in such a way that, it can directly differentiate between
201
 RTMST
various 3-D shapes. These can be possible, if the difficulties like Rotation, Mirroring,
Non-uniform Scaling, Translation etc can be overcome. Then we will be able to
recognize all the shapes like cone, cylinder, sphere, pyramid, cube, parallelepiped,
various types of prisms etc. For this purpose, more advanced and strong software
platform than MATLAB can be used in future. We can also develop the artificial
intelligence part in the robot, so that it can detect shapes more accurately in future.

9. REFERENCES

[1]. Loannou, D Huda, W Laine, “Circle recognition through A 2D Hough transform

and radius histogram”, Journal, Image and Vision Computing, 1999.

[2]. Hankyu Moon, Ramalingam Chellappa and Azriel Rosenfeld, “Optimal Edge-Based Shape
Detection”, IEEE Transaction on Image Processing, 2002, Page(s): 1209 – 1227.

[3]. Kitti Suwanratchatamanee, Mitsuharu Matsumoto, Ryo Saegusa and Shuji Hashimoto, “A
simple tactile sensor system for robot manipulatorand object edge shape recognition”,
Industrial Electronics Society, 33 rd annual conference (IECON) of the IEEE, 2007, Page(s):
245 – 250.

[4]. Tsuneo Yoshikawa, Masanao Koeda and Hiroshi Fujimoto, “Measurement of grasp
position by human hands and grasp criterion for two soft-fingered robot hands”, IEEE,
International conference on Robotics and Automation (ICRA), 2009, Page(s): 2487 – 2492.

[5]. Min Min Kyaw, Syed Khaleel Ahmed and Zainul Abidin Md Sharrif, “Shape-based sorting
of agricultural produce using support vector machines in a MATLAB / SIMULINK
environment”, IEEE, 5 th International Colloquium on Signal Processing & Its Applications
(CSPA), 2009, Page(s): 135 – 139.

[6]. Chun-Chieh Wang, Kuo-Lan Su and Chih-Teng Shen, “Implementation of a Tour Guide
Robot via Shape Recognition and Path Planning”, IEEE, 4 th International conference on
Innovative Computing, Information and Control (ICICIC), 2009 , Page(s): 544 – 547.

[7]. D. Shi, L. Zheng, and J. Liu, “Advanced hough transforms using a multilayer fractional
fourier method,” IEEE Transactions on Image Processing, vol. 19, no. 6, 2010, pp. 1558–
1566.

[8]. JI Yuan, MAO Li, HUANG Qingqing, GAO Yan, “Research on Object Shape Detection
from Image with High-level Noise Based on Fuzzy Generalized Hough Transform”, IEEE,

202
 RTMST
International Conference on Multimedia and Signal Processing (CMSP), 2011, Page(s):
209-212.

[9]. Mohd. Firdaus Zakaria, Hoo Seng Choon, and Shahrel Azmin Suandi “Object Shape
Recognition in Image for Machine Vision Application” International Journal of Computer
Theory and Engineering, Vol. 4, No. 1, February 2012, Pages(s): 76 – 80.

[10]. Akanksha Sharma, Parminder Kaur “Motion and Color Detection in Real Time Images”,
International Journal of VLSI & Signal Processing Applications, Vol.2, Issue 1, Feb 2012,
Page(s): 24-28.

[11]. Digital Image Processing Using MATLAB By Rafael C. Gonzalez, Richard E. Woods.

203
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18 th and 19th April, 2013, NITTTR, Kolkata

OBSTACLE AVOIDING WALL FOLLOWING

ROBOT
Arjita DAS, Sanjeev KUMAR, Sagarika PAL
Department of Electrical Engineering
National Institute of Technical Teachers’ Training and Research, Kolkata
[Under MHRD, Govt. of India]

Block – FC, Sector –III, Salt Lake City, Kolkata -700 106
Tel.: +91 33 23374125/0479, Fax: +91 33 23376331

E-mail: arjita.carmel@gmail.com, sanjeev.kumar0501@gmail.com

spal922@yahoo.co.in

Abstract: The paper presents a wall following robot detecting obstacles on its way while
moving ahead. The wall following is a fundamental of autonomous mobile robot. It is
characterised by following the wall on its side and maintaining a constant distance to that
wall. These robots are designed to perform task like exploration and navigation in an
open space by encountering the wall. In this present work, a prototype of wall follower
robot has been set up with a chassis attached with sensors, actuators and controller. An
infrared range finder is fastened on the side of the chassis to measure the distance from
the wall, and an IR sensor (transmitter receiver) in front to detect the obstacles coming on
its way. A microcontroller is interfaced with the other electronics to control the robot’s
position. These IR sensors sensed the object and generated a corresponding output
voltage, which is further fed to the microcontroller. The microcontroller received the
input from the sensor and generated a control signal; these control signals drive the motor
to control the robots movement and position. The developed robot is highly adaptable to
control its action, and it performed the task precisely and accurately. This wall following
robot have a vast application in the manufacturing process plant, it is applied for pick and
place object, carry object from one conveyor belt to other, trace or track particles, even
204
 RTMST
gathers information from a hazardous area where a person cannot reach. Mainly these
types of autonomous mobile robots are placed in the underground mining areas and
nuclear power plant for a typical work done

Keyword: Wall Following Robot, Autonomous Mobile Robot, Sensors, IR range finder,
Actuators, Controller, Microcontroller, Process industries.

1. INTRODUCTION

A wall following robot is an autonomous mobile robots. It is a mechanical device

operates partially in unknown and unpredictable environment, mainly programmed to
execute a predefined action. In most of the application, these self operating robots follow
a predetermined wall and detect obstacles through IR sensing. In present project, a wall
following robot has been designed which follow the wall on its side and maintained a
constant distance to it.
Various researches introduced wall-following missions by ultrasonic sensors to
measure the distance from the wall and the data of the sensor are first used to gain a local
representation of the environment in order to control the robot accordingly [1].
These Autonomous robots spread its various applications in manufacturing
process plant like sensing, obstacle detection and avoidance, and path planning. They are
mainly programmed to achieve several goals like exploration and navigation.
The navigation of wall follower robot is mainly performed by Infrared and is
more appropriate in indoor environments [2, 5]. In the case of indoor environments
autonomous navigation can be accomplished by wall-following [3, 4] where robot
navigates alongside a wall at a pre-determined distance. Wall following can be thought of
as being a low level reflexive behaviour which when coupled with a more intelligent
behaviour can accomplish a useful high level task.
Another wall-following method, presented in [6], uses feedback from a single
passive, sensitized antenna to enable both walls following obstacle avoidance for a
dynamic wheeled mobile robot. They suspect that well-designed “touch probes” such as
antennae and whiskers will provide a compelling alternative to existing proximity
sensors. Presumably, these touch probes are unaffected by surface properties such as high
specularity, and environmental factors such as light and dust. In essence, most obstacles
that a robot can physically hit can be sensed by touch.
In [7], image processing is used to detect perspective lines to guide the robot
along the centre axis of the corridor or wall.

205
 RTMST
The paper [8] describes an incremental evolutionary approach used in the
development of a
Suitable neural controller for achieving robust obstacle avoidance behaviour is then
further fine-tuned towards a wall following one for a simple mobile robot. Obstacles
avoidance is based on data acquired from 8 IR sensors. Thus there are various technique
for accomplishing wall following robot for navigation and exploration., but the papers
which proposes IR sensor for navigation are much more successful than other sensors or
control logics since IR sensors are simple, commonly employed, and relatively low-cost
sensing modalities to perform the wall-following task.
In paper [9] it is described the reactive exploration by eight infrared sensors to
detect the surrounding walls, and a compass was simulated to help the robot turns to the
desired Orientations and travel in a straight line.
In this present paper, a prototype of wall follower mobile robot has been developed, by
combination with IR sensors and microcontroller; the sensor and controller enabled the
robot, to travel along a wall maintaining a constant distance to it and to detect any
obstacles came on its way. The recognition capabilities of IR sensors have been strongly
recommended to detect the presence or absence of objects in the environment (proximity
detection) for applications such as obstacle avoidance, counting or wall-following.

2. METHODLOGY

Fig. 1. Schematic diagram for wall following robot.

Fig.1. shows a schematic diagram of the wall following robot, these autonomous mobile
robot follows a wall on its left or right according to the sensor mounted on the chassis. A
microcontroller is used as the brain of the robot, which is programmed to control the
input and output modules interfaced to it. The controller receives input from an IR sensor
206
 RTMST
(Infrared Range finder), computations are performed on this input and a control signal is
generated to control the robot’s position, and the robot follows a wall and maintains a
constant distance with it. Another IR transmitter –receiver (RX-TX) pair is placed in
front of the chassis which detect the front obstacles.

The Infrared Rangefinder (the distance sensor) gave the input pulse to the
microcontroller’s port, and the pulse is received through another port to measure the
width and calculate the distance. The output from the microcontroller is fed to the motor
driver to drive the motor for a forward or backward motion or a turn. Some computations
are performed at this total and a control signal is generated to control the robot’s position.
The sensor output is calibrated by varying the distance, and a fixed range of distance is
kept as a set value. The robot moves as shown in Fig. 2 and maintained a constant
distance to the wall. But near the corner of wall it cannot maintain a constant distance
because when it reaches near the front wall robot treat it as an obstacle and it turns before
it encounters. So its trajectory near corner is circular in nature.

Fig. 2 Robot maintains the constant distance to the wall

2.1 Sensor (IR Range finder)

Fig. 3. Infrared Range finder

207
 RTMST
The Sharp IR Range Finder shown in Fig.3 is extremely effective sensor, easy to
use, affordable, very small, good range (inches to meters), and has low power
Consumption. This sensor having features, such as high distance measuring range (20-
150cm) within Supply voltage ranges 4.5-5.5V. The circuit design embedded within it, is
shown below in the Fig. 4

Fig. 4. Internal circuit of IR range finder

This Sharp IR Range Finder used in this project has a strong recognition capability, it is
interfaced to the Microcontroller which continuously monitors presence of the obstacles
and intimates to the controller. This works on the process of triangulation shown in fig.
5.

Fig. 5. IR Range Finder works by the process of triangulation

208
 RTMST
A pulse of light (wavelength range of 850nm +/-70nm) is emitted and then
reflected back (or not reflected at all). When the light returns it comes back at an angle
that is dependent on the distance of the reflecting object. Triangulation works by
detecting this reflected beam angle - by knowing the angle, distance can then be
determined.
The IR range finder receiver has a special precision lens that transmits the reflected light
onto an enclosed linear CCD array based on the triangulation angle. The CCD array then
determines the angle and causes the rangefinder to then give a corresponding analog
value to be read by the microcontroller.

2.2 Microcontroller (ATMEGA 8)

The microcontroller used in this present project is ATMEGA8 of the AVR series;
it’s an 8 bit microcontroller, has Advanced RISC architecture and has fully Static
Operation. Its frequency is up to 16MIPS Throughput at 16MHz.Microcontroller is the
controlling device of the whole system, which is programmed to control the input and
output modules interfaced to it. The controller makes use of an IR sensor as input sensor
to sense the distance from the wall and detects obstacles. This data received by the IR
sensors and is fed to the Microcontroller. The Microcontroller processes this data and acts
accordingly on motors of the robot.

2.3 Software description

While working on this project, to run the robot successfully we perform in AVR
software, the programs are done in
1. WinAVR (free GNU compiler)
2. AVR Studio 4.0 (free IDE)
Sensors are input or ‘senses’ of any intelligent system. The input from the sensors
is fed to the microcontroller, It is clear that the output of the sensor is an analog voltage.
But Microcontroller does not accept the analog voltage. So we need to convert the analog
voltage to digital before we feed it to the microcontroller. For this conversion we can use
1. Comparators
2. ADCs.
These comparators &ADCs are internally designed in the ATMEGA8 microcontroller.
So, the outputs of the sensor are fed to the motor controller to drive the motor in the
forward motion. According to program fed in the WinAVR the robot move in the
respective directions.

2.4 L293 MOTOR DRIVER

The L293 and L293D are quadruple high-current half-H drivers. This motor driver is
used in this present project to drive the motor in the respective direction. The L293 is
designed to provide bidirectional drive currents of up to 1 A at voltages from 4.5 V to 36
V. The L293D is designed to provide bidirectional drive currents of up to 600-mA at
voltages from 4.5 V to 36 V. Both devices are designed to drive inductive loads such as

209
 RTMST
relays, solenoids, dc and bipolar stepping motors, as well as other high-current/high-
voltage loads in positive-supply applications. All inputs are TTL compatible. Each output
is a complete totem-pole drive circuit, with a Darlington transistor sink and a pseudo-
Darlington source. Drivers are enabled in pairs, with drivers 1 and 2 enabled by 1,2EN
and drivers 3 and 4 enabled by 3,4EN. When an enable input is high, the associated
drivers goes enabled and their outputs are active and in phase with their inputs. When the
enable input is low, those drivers are disabled, and their outputs are off and in the high-
impedance state. With the proper data inputs, each pair of drivers forms a full-H (or
bridge) reversible drive suitable for solenoid or motor applications.

Fig. 8. Connection with Motor

As shown in fig. 8 there are connectors provided in the development board to connect DC
motors with L293D. Each connector contains 3pins. The centre pin is ground. The left
and right pins should be connected to the two motor terminals.
The motor driver has 3 input pins for each motor. They are EN, A, B. A truth table is
provided below to relate the input conditions with the motor action in table 1.

EN A B Motor Action
0 X X Stopped
1 0 0 Brake
1 1 1 Brake
1 0 1 Clockwise
Rotation
1 1 0 Anti-Clockwise
Rotation

Table 1. Truth table of the input condition of the motor table

3. Hardware Model of the System

210
 RTMST

Fig 9. Chassis Model of the robot

The shape of the chassis is shown in the Fig. 9. Dimensions are given as shown in the
figure. The base of the chassis made up of Bakelite board and two wheels are attached
to its both side of the rear end. The castor wheel is placed in front of the chassis to
balance the entire one.

Fig. 10(a).Top view of the developed robot Fig. 10(b). Bottom view of the developed
robot

211
 RTMST

4. Working Principle of the wall following Robot

The Fig 9 shows the hardware model of the autonomous mobile robot, where a
wheel is placed on either side of the chassis runs independently from each other by the
help of the motor, which is drive by the motor driver L293D. The IR Range finder is
placed on the right side of the chassis which senses the distance from the wall and
gives the corresponding output in voltage in analog form. This output voltage is fed to
the microcontroller Atmega8 (where ADC is internally built up). Now, a signal is
generated by the microcontroller and drives the motor driver L293D to move the
motor further.

212
 RTMST

4.1 Flowchart for the wall following Robot

Fig. 11. Flowchart o

START

Take data from IR sensors.(IR1= data from ir range finder sensor,

Fig. 11. Flowchart of wall following method
IR2= data from ir obstacle detector sensor)

ye
s Is
Obstacle detected
IR2>5
00

Move left no

no no
Move forward Is Is

IR1>6 IR2<5 Follow the wall

20 55

Move right
ye ye Move forward
s s

Too far to wall

Move forward

Move left Move right

213
 RTMST

The value 500, in the above flowchart is the count value of the ADC for the corresponding IR
sensor2
(i.e. The sensor that detects the obstacle in front) is the set value. The count value of the IR range
finder i.e. sensor1, is range between 620 to 555 whose corresponding distance is 16 to 20cm. This
is the set value of the IR sensor1.

4.2 Experimental result and discussion

Table 2
O/P Voltage
Distance Count Value
Serial no. of Sensor in Robot Movement
in cm (AVR ADC O/P)
volt
1. 8 2.02 470 Not used*
2. 10 2.19 490 Move Left
3. 11 2.35 520 Move Left
4. 12 2.75 590 Move Left
5. 13 2.82 630 Move Left
6. 14 2.85 660 Move Left
7. 15 2.84 640 Move Left
8. 16 2.8 620 Move Forward
9. 17 2.75 590 Move Forward
10. 18 2.7 580 Move Forward
11. 19 2.58 570 Move Forward
12. 20 2.49 555 Move Forward
13. 21 2.35 520 Move Right
14. 22 2.21 500 Move Right
15. 23 2.09 480 Move Right
16. 24 2.02 470 Move Right

*That case not used because as IR sensor gives same o/p at very close as well as very far
distance of object. It has been observed that the sensor works properly minimum at a
distance of 13cm away from the wall.

214
 RTMST

Fig. 11. Wall following mechanism by IR Sensor and microcontroller

The above Fig.11 had been obtained following the experimental result as indicated in the
Table-2 of the wall following mechanism. This autonomous mobile robot had
successfully followed the wall and moved parallel with it in an unknown area. It also
detected the obstacles came on its way. The robot while moving forward maintained a
constant distance to that wall, which was attained by the sensor (IR Range finder), the
sensor gave a corresponding output in voltages according to the input distance from the
wall, a certain distance was chosen and the value was fixed, which was further fed to
microcontroller as the set value of the distance. This value was maintained as the constant
distance by the robot while following the wall.
Whenever it detected obstacles came on its way it stopped for a second avoided it by
moving little bit left and again moved in the same manner. The controlling device of the
whole system is a Microcontroller to which IR sensors, and DC motors are interfaced.
The Microcontroller processes this data and acts accordingly on motors of the robot, and
controlled the robot’s position and movement very accurately. By using the infrared
range finder some advantages are obtained rather than ultrasonic sensors or bump
switches.
.

215
 RTMST
5. CONCLUSION

In this paper, it was found that under ideal condition the robot was able to follow the wall
with constant distance to the wall. The method of wall following and obstacle detection
of an object in an unknown area is described above. It proposed a method of following
the wall and avoiding the obstacle while moving ahead. This wall-following task was
based on Distance measurements, which was obtained by a sensor IR rangefinder. The
state of the robot with respect to a near wall is estimated from this sensor measurement.
Obstacle avoidance capability has been added to the task by placing an IR sensor in front
of the robot chassis to detect the obstacles that came on its way and avoided it. The
resultant value is very accurate and precise in nature, then other sensors. It is highly
rugged.
The difficulties of obstacle avoiding or wall following tasks are extensive. The wall may
not be continuous, or the obstacles may not be defined or recognised to its requirement,
but performing it in the favourable condition, the experiment was successfully completed.
This wall following robot have a vast application in the manufacturing process plant, it is
applied for pick and place object, carry object from one conveyor belt to other, trace or
track particles, even gathers information from a hazardous area where a person cannot
reach. Mainly these types of autonomous mobile robots may place in the underground
mining areas and nuclear power plant for a typical work done
Thus under experimental conditions, navigating along the wall and avoiding the obstacles
in front has been successfully set up.
The resulting time for a complete experiment is relatively short. This is one of the added
advantages of this algorithm.

6. REFERNCE

[1]. P. van Turennout, G. Honderd, L. J. Van Schelven, Wall-following control of a Mobile

Robot, IEEE, International Conference onRobotics and Automation, pp. 280-285,
May,1992.

[2] J. Borenstein and Y. Koren. Real-time obstacle avoidance for fast mobile robots. IEEE
Trans. SMC, 19(5): 1179-1 187, September 1989.

[3] R. A. Brooks. Elephants don’t play chess. In Proceedings of the 1990 IROS, North-
Holland, June 1990.

[4] G. Giralt. NATO ASI Series Robotics and Art$cial Intelligence, chapter Mobile Robots.
Springer-Verlag. New York, NY, 1984.

[5] C. Gourley and M. Trivedi. Sensor based obstacle avoidance and mapping for fast mobile
robots. In Proceedings of IEEE ICRA, pages 1306-1311, San Diego, CA, May 1994

216
 RTMST
[6] A. G. Lamperski, Owen Y. Loh, Brett L. Kutscher and Noah J. Cowan, Dynamical Wall-
Following for a Wheeled Robot using a Pasive Tactile Sensors, John Hopkins
University,Baltimore

[7]. R. Frizera Vassallo, H. J. Schneebeli, J. Santos-Victor, Visual navigation: combining

visual servoing and appearance based methods, Proceedings of the International
Symposium on Intelligent Robotic Systems, SIRS”98,Edinburgh, Scotland, July, 1998.

[8]. Dhiraj Bajaj, Marcelo H. Ang Jr., An Incremental Approach in Evolving Robot Behavior,
Proceedings of the Sixth International Conference on Control, Automation, Robotics and
Vision (ICARCV’2000), Singapore, 5-8 Dec,2000

[9] R. Clark, A. El-Osery, K. Wedeward and S.Bruder, A Navigation and Obstacle

Avoidance
Algorithm for Mobile Robots Operating in Unknown, Maze-Type Environments,
Proc.International Test and Evaluation Association

[10] Chris Churavy, Maria Baker, Samarth Mehta, Ishu Pradhan, Nina
Scheidegger,Steven Shanfelt, Rick Rarick, and Dan Simon, Cleveland State
University,Department of Electrical and Computer Engineering, IEEE Potential
(Unpublished), October 2007.

[11] Mehmet Ergezer, Multivariable Control Methods for Wall Tracking Robot,
Cleveland State University, Department of Electrical and Computer Engineering,
Project Paper, 2006.

217
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18 th and 19th April, 2013, NITTTR, Kolkata

HYDROFORMING SETUP AND ITS

EXPERIMENTAL RESULTS

DWAIPAYAN DE 1, ASISH BANDYOPADHYAY 2

1
PG Student (MME), Department of Mechanical Engineering, Jadavpur University, Kolkata - 32
dwaipayan88@gmail.com, 9477417570 (M)

2
Professor, Department of Mechanical Engineering, Jadavpur University, Kolkata - 32
asishbanerjee@yahoo.com, 9433467997(M)

Abstract
In industry several materials are used for making different components. Some materials
are preferred for their lightweights with adequate strength for the desired use. The
formability of these light metals such as aluminium alloy is not as good as that of steel.
Hence various new forming techniques have been introduced. One of the methods is the
hydroforming techniques to manufacture components. Hydroforming is the process that
metal sheets are formed to the desire shape by high hydraulic pressure. It has become
popular in recent years, due to the increasing demands for lightweight parts in various
fields, such as bicycle, automotive, aircraft and aerospace industries. In the present work
a hydroforming setup has been developed with water as working fluid. Experiments have
been carried out with this newly developed hydroforming setup with aluminium as work-
piece material and the results are found to be satisfactory.

Keywords: Hydroforming, hydraulic pressure, aluminium sheets.

218
 RTMST

1. Introduction

Metal forming is a process in which permanent deformation is obtained while working

in the plastic region. This process exploits a remarkable property of metals, which is
their ability to flow plastically in the solid state without deterioration of their
properties. With the application of suitable force, the material is moved to obtain the
desired shape with almost no wastage. The required forces are generally high, and the
tools and equipment needed are quite expensive. Forming is usually very much
common in the field of manufacturing of different types of products including
household products. There are different types of forming process like rolling, forging,
drawing, deep drawing, extrusion, bending etc. These are generally called the
conventional forming operations. Various new and advanced forming techniques are
evolved to meet the purpose of forming of light metal alloys like aluminium alloy,
copper alloy etc. These advanced forming techniques are generally termed as non-
conventional forming operations and they are becoming quite popular in recent times.
Some of them are explosive forming, electrohydraulic forming, electromagnetic
forming, hydroforming etc. In Hydroforming the metal sheets are formed by applying
high fluid pressure. The hydroforming process, in general, is of two types - sheet
hydroforming and tube hydroforming. In the sheet hydroforming (SHD) process, the
sheet blank is formed by hydraulic pressure inside the die cavity as illustrated in Fig.1.
This technique allows a much deeper draw, which is necessary for manufacturing
products with complex geometry. Hydroforming of sheet-metal components requires
high-capacity presses, special deep drawing dies and special high-pressure hydraulic
components. To compensate for these disadvantages, new designs for presses and dies
as well as new optimized hydroforming processes have been developed. All these
developments have made hydroforming sheet metal components more competitive,
especially for low-volume production, but also in some cases for high-volume
production also.

219
 RTMST

Fig. 1. Simple sheet hydroforming operation

Siegert et al. [1] have discussed about the possibilities and advantages of
hydroforming in sheet metal forming as well as in forming tubes and extrusions. They
have shown that for hydroforming tubes it is possible to form with outer and/or inner
pressure with and without axial forces. Their research shows that for hydroforming
sheet metal it is useful to control the metal flow between the binders with multipoint
cushion systems.Ch. Hartl [2] reported about the research and development work
carried out intensively during the past few years in relation to hydroforming
technology. Also the present status for economic mass production of high quality light
weight components using hydroforming techniques has been reported in this research
work. In modern times the main aim of the automotive industries is to reduce the fuel
consumption. With this aim in view, the demand for weight reduction in modern
vehicle has become a driving factor and for that reason the author paved the way to
carry out research work on hydroforming technology. The author also pointed out that
the R & D work carried out on the area of hydroforming technology during the past
few years offers decisive advantages regarding the manufacturing of complex shaped
light weight components. The author also provides information in his article about the
continuous improvements in hydroforming technology regarding the design of process
steps, semi-finish products, hydroforming press, tool technology, hydroforming of
sheet material and the use of heat energy are current measures contributing to further
increase of hydroforming applications. According to Keigler et al. [3] new
requirements of the automotive industry like light weight components and non-
corroding construction, demand for new production methods become inevitable and
for this reason hydroforming process of aluminium alloys plays a vital role. But the
disadvantage of aluminium alloys is the poorer formability compared to steel. For this
reasons, they propose the path in their research that the enhancement of the forming
temperature would increase the formability of the aluminium alloys during the

220
 RTMST
hydroforming process. Parsa and Darbandi [4] have decided to manufacture a
complicated stamping shell fender by sheet hydroforming process to evaluate the
capability of sheet hydroforming process for the production of complex parts, whose
manufacturing through conventional methods required several steps. Their
experimental results and analysis show that sheet hydroforming could be used for the
production of complex parts, through the optimization of blank geometry, fluid
pressure, shape and position of the semi die device or guide. Hama et al. [5] focus on
the outflow characteristics of a pressure medium in a sheet hydroforming process. The
initial measurements of the outflow of a pressure medium through a gap between two
clamped dies have been carried out by them in order to study the basic properties of
the outflow. The variation and distribution of the hydraulic pressure in the flange area,
variation in the gap between the die and the blank holder can affect the fluid-
lubrication effect, and the changes in the hydraulic pressure distribution in the flange
area are the future research areas that can be carried out according to their point of
view on sheet hydroforming process.
With the information available, in the present work it has been attempted to develop a
hydroforming setup for forming of aluminium sheets and to study the effect of
hydraulic pressure on maximum height of the dome.

2. Experimentation

A sheet hydroforming setup has been developed and the photographic view of the
setup is shown in Fig.2. Experiments have been carried out on 0.3 mm aluminium
sheets by varying the hydraulic pressure ranging from 5 bar to 22 bar. The
corresponding dome heights have been measured by using height gauge. The results
are shown in Table 1

221
 RTMST

Fig. 2. Photographic view of the hydroforming setup

3. Results & Discussions

The aluminium sheet of 0.3mm thickness after deformation attains almost a

uniform spherical shape as shown in Fig. 3.

222
 RTMST

Fig. 3. An aluminium sheet after deformation

The shape of the deformed material has been shown in Fig.4.

Fig. 4. Shape of the deformed sheet metal

The results i.e. maximum height of the dome for various hydraulic pressures has
been shown Table 1.

223
 RTMST
Table 1. Variation of maximum height for different applied pressure

Sl No Pressure Maximum height of the dome

(bar) (mm)

1 5 10.00

2 10 10.70

3 15 15.20

4 20 15.30

5 22 15.92

The variation of the maximum height of the dome with variation variation of
hydraulic pressure has been plotted in Fig.5. From the experimental data it is
observed that with the increase of pressure the deformation of aluminium sheet i.e.
the height of the dome also increases. But this increase is not uniform. With the
increase of pressure from 5 bar to 10 bar the rate of increment of the maximum
height of the dome is low as compared to that with change in pressure from 10 to
15 bar. Again from 15 bar to 20 bar the rate of increment is low and the variation
remains almost flat. With increase in pressure from 20 bar to 22 bar the rate of
maximum height of the dome increases significantly. With further increase of
pressure above 22 bar the height of the dome does not increase as the aluminium
sheet is fractured and the pressure automatically falls down. The surface finish of
the aluminium sheet is found to be unaffected by the process.

224
 RTMST

Fig. 5. Variation of maximum height of the dome for different applied pressure
4. Conclusion

From the present work it is concluded that

 The deformation by using hydroforming process can be achieved quite

satisfactorily.
 The advantage of this type of non-conventional hydroforming process is
that the surface finish remains same before and after forming.
 The cost of developing a compact small hydroforming machine is not very
high.
 The different shape and size of the blank material can be achieved only by
changing the die. Thus any shape and size can be achieved in this type of
forming and complex shape can also be attained with ease.

Acknowledgement

The authors sincerely acknowledge The Institution of Engineers (India) and Jadavpur
University Research Grant for the financial support for the fabrication of Sheet
Hydroforming setup .

225
 RTMST
References

[1] Klaus Siegert, Markus Haussermann, Bruno Losch and Ralf Rieger, “Recent
developments in hydroforming technology”, Journal of Materials Processing Technology,
Vol 98, 2000, pp. 251-258.
[2] Ch. Hartl, “Research and advances in fundamentals and industrial applications of
hydroforming”, Journal of Materials Processing Technology,Vol 16, 2005, pp.383-392.
[3] Michael Keigler, Herbert Bauer, David Harrison and Anjali K.M. De Silva, “Enhancing
the formability of aluminium components via temperature controlled hydroforming”,
Journal of Materials Processing Technology, Vol 167, 2005,pp.363–370.
[4] M.H. Parsa and P. Darbandi, “Experimental and numerical analyses of sheet
hydroforming process for production of an automobile body part”, Journal of Materials
Processing Technology, Vol 198, 2008, pp.381-390.
[5] Takayuki Hama, Ken Kurisu, Keita Matsushima, Hitoshi Fujimoto and Hirohiko
Takuda, “Outflow Characteristics of a Pressure Medium during Sheet Hydroforming”,
ISIJ International, Vol 49, No 2, 2009, pp.239–246.

226
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18 th and 19th April, 2013, NITTTR, Kolkata

Parametric Optimization of Laser Microdrilling of SIC-

30BN Nanocomposite.

N.ROY1, A.S.KUAR2, S.MITRA3, B.ACHERJEE4

1
Research Scholar, Department of Production Engineering, Jadavpur University, Kolkata,
India
Phone No. - +919836725305; Email: nilanjan_83@yahoo.co.in
2
Associate Professor, Department of Production Engineering, Jadavpur University,
Kolkata, India
Email: askuar@rediffmail.com
3
Associate Professor, Department of Production Engineering, Jadavpur University,
Kolkata, India
Email: souren.mitra@gmail.com
4
Assistant Professor, Department of Mechanical Engineering, Asansol Engineering
College, Asansol, India
Email: a.bappa@yahoo.com

Abstract:

This paper presents an experimental approach to investigate the relation between process
parameter and machining response during pulsed Nd: YAG laser micro drilling of Silicon
Carbide 30% Boron Nitride (SiC-30BN) nanocomposite material. The setting of
machining parameters was determined by using Taguchi L27 orthogonal array. Single
Objective Optimization also performed to determine optimum level of process parameter.
The orthogonal array, the signal-to noise ratio, and analysis of variance are incorporated
here to study the effect of each process parameter on machining response. It is found that
the quality of hole circularity improved considerably at the optimal condition determined
through proposed optimization method. This indicates the feasibility of proposed
optimization technique to obtain desired product quality during laser microdrilling of
nanocomposite materials.

Keywords:

Laser MicroDrilling, Nanocomposite, Hole Circularity, Taguchi Method.

227
 RTMST

1. Introduction:

Ceramic and ceramic matrix composite materials have an outstanding properties like low
thermal and electric conductivity, low specific gravity, high strength, low deformability,
high wear resistance and chemical inertness which make ceramic materials highly
competitive against other conventional materials and widely used in construction,
aerospace and transportation industries as well as in medical and military applications[1].
In which nanocomposite offers a remarkable potential for new properties and
technologies. However, ceramic materials are typically brittle and difficult to machine.
Application of composite materials has often been hindered due to the difficulty in
machining for their inherent inhomogeneity, abrasive nature of reinforcements [2]. Laser
based micromachining of nanomaterials have some advantages over other conventional
and non conventional manufacturing technique in size distribution, material property,
geometry and purity [3]. Laser micromachining, being a non-contact process, does not
have the problems of mechanical damage and tool wear. Lasers micromachining offer a
wide range of wavelengths, pulse duration and repetition rates. These attributes allow
laser micromachining with high resolution in depth and lateral dimensions. The field of
laser micromachining includes manufacturing methods like drilling, cutting, welding as
well as ablation and material surface texturing [4]. The material is melted and ejected to
form a hole during pulsed Nd:YAG laser microdrilling process, uses a series of pulses at
the same spot with a particular laser parameter setting[5]. Hole Circularity is one of the
important machining response during laser percussion drilling. Modeling of the process is
required to get the desired machining response. Yilbas have investigated the effects of the
laser parameters and the material properties on the hole quality with different types of
materials [6]. Jackson and O‟Neill investigate the interaction phenomena of different
process parameter of Q-switched, diode-pumped Nd:YAG laser using different
wavelengths on M2 tool steel [7]. Bandhopadhayay et al. investigated the influence of the
process variables on hole diameter and taper angle of drilled holes produced on thick
IN718 and Ti-6Al-4V sheets by Nd:YAG laser [8]. French et al. used two level factors in
Nd:YAG laser percussion drilling and it was found that pulse shape, energy, peak power,
focal position, gas pressure and Nd:YAG laser rod were the most significant influences
on the hole taper[9]. some recent researches have been done to control hole-taper via the
development of drilling techniques such as use of an assist gas during laser drilling
process. The influence of processing gas on laser micro drilling and ablation of aluminum
and stainless steel are experimentally studied by Reg et al.[10]. The effect of laser peak
power and pulse width on the hole geometry repeatability in laser percussion drilling
where oxygen is used as an assist gas have been investigated by G.K.L.Ng and Lin. Li.
The results shows that higher peak power, and shorter pulse width gives better hole
geometry repeatability [10]. Reza et al develop a new laser cutting technique with the use
of a controlled mixer of oxygen and nitrogen as an assist gas for cutting of CFRPs to get
the improved quality of the cut whilst keeping the MRR at reasonable levels [11]. Since a
large number of process parameter controls the laser machining process, it is very
complicated to develop a physical model for laser micro drilling to get desired micro
holes. Some prior studies have used statistical modeling to analyze and determine optimal
228
 RTMST
parameter setting of the process. Taguchi method is a systematic application of design
and analysis of experiments for the purpose of designing and improving product quality.
Maghsoodloo et al. examine the concept of Taguchi’s quality loss functions, orthogonal
array, robust design, and signal-to-noise ratio thoroughly for practical applications [12].
Nalbant et al. used the Taguchi method to find the optimal cutting parameters for surface
roughness in turning [13]. Cao et al optimize the laser cladding process of ZE41A-T5
magnesium alloy castings and also find out the most dominating factor by using Taguchi
experimental method [14]. Ng et al is used Taguchi method as a statistical technique to
optimize selected laser processing parameters in laser surface alloying of NiTi shape
memory alloy with Co and Nb [15]. Taguchi method of parameter design is used as a
statistical design of experiment technique to set the optimal process parameters into laser
transmission welding of acrylics by Bappa et.al. [16] .From this literature review, it is
observed that the Taguchi method is widely applied in different applications for
determining the optimal parameters setting in different manufacturing processes to get
the desired product quality.
In the present research work an attempt is made to optimize the process parameter in
laser microdrilling of SiC-30BN nanocomposite to get the desired micro hole. The
Taguchi method of parameter design is used as a statistical tool to plan and analyze the
experimental results. The influence of process parameters on micro hole is also reported.
In addition, the most effective process parameter and the order of importance of the
controllable parameters to the machining response in the laser micro hole drilling process
are determined by signal-to-noise (S/N) ratios and analysis of variance analysis
(ANOVA).

2. The Taguchi Method for Robust Parameter Design:

The Taguchi method is an offline statistical quality control technique in which the level
of controllable factors are chosen so as to nullify the variation in responses due to
uncontrollable factors such as humidity, vibration, and environmental temperature. It uses
orthogonal array (OA) to set up the experiment for the advantage of less number of
experiments, and optimizes the process parameters by the analysis of signal-to- noise
(S/N) ratio response table and response graph. Here, the ‘signal’ represents the desirable
value and the ‘noise’ represents the undesirable value and the S/N ratio expresses the
scatter around the desired value. Usually, there are three categories of the quality
characteristic in the analysis of the S/N ratio, i.e. the larger-the-better, smaller-the-better,
and nominal the- best. The S/N ratio for each level of process parameters is computed
based on the S/N analysis. Larger S/N ratio corresponds to a better quality characteristic.
Therefore, the optimal level of the process parameters is the level with the highest S/N
ratio. The optimal process parameters can improve the robustness of products, so the
Taguchi method is also called parameter design. The optimal combination of process
parameters can then be predicted based on this analysis. ANOVA is employed to estimate
the error variance and determine the significant parameters.

229
 RTMST
3. Experimental Set Up:

A typical CNC pulsed Nd:YAG laser machining system manufactured by M/s

Sahajanand Laser Technology, India, is used for the experimental study. The following
independently controllable process parameters are identified to carry out the experiments:
Lamp Current, Pulse Frequency, Assist gas pressure, Pulse Width and Focal length.
Previous research works reveals that air as an assist gas is unable to produce the desired
quality characteristics for nanocomposite. Because of this, inert argon gas is selected for
this work to use it as an assist gas. Trial runs were conducted by varying one of the
process parameters at a time while keeping the rest of them at constant value. Hole
Circularity is selected as machining response for this study. 1.57 mm thick Silicon
Carbide 30% Boron Nitride (SiC-30BN) nanocomposite has been considered for
experimental study. Properties of SiC-30BN nanocomposite is listed in Table 1.

Table 1. Properties of SiC-30BN Nanocomposite

Property Unit Values

Yield Stress MPa 263.2 ± 6
Young’s Modulus GPa 230±5
Fracture Strength MPa 380±10
Density g/cc 2.906

The methodology of Taguchi for five factors at three levels was used to implement the
planning of the orthogonal array experiments. A L27 orthogonal array with five columns
and 27 rows is employed in this work. The experiments are carried out according to the
arrangement of the orthogonal array. The experimental layout for the laser micro drilling
parameters using the L27 orthogonal array is presented in Table 2.

Table 2. Machining Parameters and Their Levels

Process Parameters Level

1 2 3
Lamp current (L)(amp) 23.5 24.5 25.5
Pulse frequency (f)( kHz) 1 2 3
Pulse width (w) (%) 2 4 6
Argon gas pr. (p) (kg/cm2) 0.4 0.7 1.0
Focal length (F) (μm) -200 0 +200

230
 RTMST
4. Experimental Observation:

Table 3. Experimental Data:

Experiment Circularity S/N Ratio

No.
1 0.6709 -3.46684
2 0.5955 -4.50236
3 0.5354 -5.42643
4 0.7465 -2.53940
5 0.6883 -3.24444
6 0.6543 -3.68446
7 0.4240 -7.45268
8 0.3869 -8.24803
9 0.3376 -9.43195
10 0.7681 -2.29164
11 0.4454 -7.02500
12 0.3813 -8.37466
13 0.6386 -3.89542
14 0.4609 -6.72787
15 0.4309 -7.31247
16 0.3605 -8.86189
17 0.3333 -9.54329
18 0.3419 -9.32202
19 0.8983 -0.93157
20 1.0000 0.00000
21 0.7067 -3.01530
22 0.7465 -2.53940
23 0.5146 -5.77060
24 0.4690 -6.57654
25 0.4953 -6.10263
26 0.3926 -8.12099
27 0.4380 -7.17052

The observed data from the set of experiments have been used as inputs to MINITAB
software for parametric optimization Statistical software MINITAB is used here to get
response table of means of circularity. Statistical analyses (S/N and ANOVA) were
performed at a significance level of 0.05 (i.e. α = 0.05 or a 95 per cent confidence level).
The S/N ratios for each process parameter were calculated to study the influence of these
process parameters on machine response. Means of S/N ration of circularity is
graphically represent here. The signal-to- noise (S/N) ratios of Circularity is considered
as ‘larger the- better’ characteristic and calculated and given in a tabular form as follows,

231
 RTMST
1 n 1
η  10Log10 (  )
i n i 1 y 2
i

(1)

Where, ηi is the S/N ratio at the ith trial or experimental run, yi is the observed response or
quality value at the ith trial or experimental run, and n is the number of trails at same
parameter level.

Table 4: Responses for S/N ratio:

Process parameters lavel Delta Rank

1 2 3
Lamp current (L)(amp) -0.795 -0.898 -0.754 0.145 2
Pulse frequency (f)( kHz) -0.724 -0.753 -0.969 0.245 1
Pulse width (w) (%) -0.848 -0.797 -0.802 0.051 5
Argon gas pr. (p) (kg/cm2) -0.813 -0.776 -0.857 0.082 4
Focal length (F) (μm) -0.740 -0.833 -0.873 0.133 3

5. Results and Discussion:

Lamp current Pulse frequency P ulse width Argon gas pr. Focal length
(L)(amp) (f)( kHz) (w) (%) (p) (kg/cm2) (F) (mu m)

-0.75

-0.80
S / N R a t io

-0.85

-0.90

-0.95
0

00
.5

0. 7
2
3
2

6
0 .4
.5

1 .0
- 20
.5

0
23

+2
24
25

Fig.1. Effects of Nd:YAG laser microdrilling parameters on Circularity.

5.1. S/N Ratio analysis:

Circularity has been found to be a decreasing tendency with increase of lamp current up
to 24.5 amp, and then it increases with increase of lamp current. Generally, higher laser
beam power generates higher thermal energy which produces better circularity. From the
figure.1. it is observed that Increase in the Pulse frequency results in decrease in
232
 RTMST
circularity due to instant melting and vaporization from the top surface at high pulse
frequency and vice versa. Increase in Pulse width results in uniform material removal
from machining zone due to high peak power. For which shape of the micro hole at top
surface become regular results in increase in circularity but further increase in Pulse
width causes higher heating value in top surface than bottom surface results in decrease
in circularity. It is found that, circularity increases with gas pressure upto 0.7 kg/cm2 and
thereafter starts decreasing. Increase in Assist gas pressure causes rapid removal of
molten material as well as excess heat from machining zone results in increase in
circularity but further increase in assist gas pressure cause uneven heating value on the
top surface of the work piece. As a result low circularity is observed. Circularity
decreases with increase in Focal length. The increased beam spot area with increased
focal length causes high energy laser beam which removes large amount of material from
the top surface rapidly results in low circularity.

5.2. ANOVA Analysis:

ANOVA (Analysis of Variance) is a computational technique to estimate the effect of

each process parameter on microhole quality characteristics quantitatively. In ANOVA,
the ratio between the variance of the process parameter and the error variance is called
Fisher’s ratio and it is used to determine the significance of process parameter on
machining response by comparing the F-test value of the parameter with the standard F
table value (F0.05) at the 5 per cent significance (95% confidence) level. The ANOVA
results are listed in Table 5.

Table 5 Results of the ANOVA

Source Degrees of Sum of Mean of F P %

Freedom Square Square Contribution
Lamp Current 2 0.0010899 0.0005449 15.54 0.000 18.03
Pulse Frequency 2 0.0035316 0.0017658 50.36 0.000 58.42
Pulse Width 2 0.0001576 0.0000788 2.25 0.138 2.61
Assist Gas 2 0.0003383 0.0001691 4.82 0.023 5.59
Pressure
Focal Length 2 0.0009283 0.0004641 13.24 0.000 15.36
Error 16 0.0005610 0.0000351
Total 26 0.0066067

From the table it is found that all the process parameters except pulse width are
statistically significant. Percentage contribution column in ANOVA result table indicates
at the degree of influence of all the process parameter on the microhole Circularity.
According to the table it is found that Pulse Frequency (58.42%) has the most dominant
factor followed by Lamp Current (18.03%), Focal Length (15.36%), and Assist Gas
Pressure (5.59%).

233
 RTMST

5.3 Determination of optimal factor levels:

Based on the main effects plot for S/N ratios (Fig. 1), the optimal microdrilling process
parameters are chosen within the range of tested values, as lamp current at level 3 (L=
25.5 amp), Pulse Frequency at level 1 (f= 1kHz), Pulse Width at level 2 (w= 4%), Assist
Gas Pressure at level 2 (p= 0.7kg/cm2), and Focal Length at level 1(F= -200 µm), i.e.
L3f1w2p2F1.

6. Confirmation Test:

After getting the optimal parameter settings for desired micro hole characteristics, the
next step is to verify the feasibility of the proposed Taguchi method. The estimated S/N
ratio, ηopt, using the optimal level of the welding parameters is calculated as

q
η opt  η̂    η j  η̂ 
j1 

(2)

Where, ̂ is total average S/N ratio of all the experimental values, η j is the mean S/N
ratio at the optimal level, and q is the number of laser drilling parameters that
significantly influence the quality characteristics. An experiment is performed with
optimal parameter settings. Table 6 furnishes the result of confirmation test.

Table 6 Result of confirmation experiment

Initial parameter Optimal parameters

setting Prediction Experiment
Level L3f1w3p2f2 L3f1w2p2F1 L3f1w2p2F1
Circularity 0.937 0.941
S/N ratio (dB) -0.56521 -0.54150 -0.52825
Prediction error (%) 2.51
Improvement of S/N ratio 0.03696

The improvement in S/N ratio at the optimum level is found to be 0.03696 dB (6.54%)
against the initial process parameter setting. The micro hole Circularity value at the
optimum level is 0.941 which is an improvement of 0.04 (4.27%).

7. Conclusion:

Taguchi optimization technique is used here to optimize the laser microdrilling process
performance characteristic on SiC-30BN nanocomposite. It is shown from the study that

234
 RTMST
the performance characteristic of the laser microdrilling process i.e. hole Circularity is
improved by using the proposed method. From the ANOVA analysis it has been found
that pulse frequency, applied gas pressure, lamp current and focal length have statistically
significant effects on microhole Circularity. Pulse frequency is found as the most
dominant factor on variations of microhole Circularity whereas Pulse Width is the least
significant factor.The mechanisms for the phenomenon observed during the experimental
study are also discussed. The outcome of the study indicates the justification of proposed
optimization technique to obtain desired product quality during laser microdrilling of
SiC-30BN nanocomposite materials.

References:

[1] M. V. Swain, R.W. Cahn, P. Haasen, and E.J. Kramer, “Structure and properties of ceramics”
(VCH, New York 1994).
[2] Chinmaya R. Dandekar, Yung C. Shin, “Modeling of machining of composite materials: A
review”, International Journal of Machine Tools & Manufacture (Article in press).
[3 ] Lin Li, Minghui Hong, Michael Schmidt, Minlin Zhong, Ajay Malshe, Bert Huis in’tVeld,
Volodymyr Kovalenko : “Laser nanomanufacturing – State of the art and challenges”, CIRP
Annals - Manufacturing Technology 60 (2011) 735–755.
[4] Udo Klotzbach, Andrés Fabián Lasagni, Michael Panzner , Volker Franke: “Laser
Micromachining”,
[5] A. S. Kuar, B. Acherjee, D. Ganguly, and S. Mitra, “Optimization of Nd:YAG Laser
Parameters for Microdrilling of Alumina with Multiquality Characteristics via Grey–Taguchi
Method”, Materials and Manufacturing Processes, 27: 329–336, 2012.
[6] Yilbas.B.S., “Parametric study to improve laser hole drilling process”. Journal of materials
Processing Technology 1997, 70, 264–273.
[7] Jackson MJ, O‟Neill W. 2003, “Laser micro-drilling of tool steel using Nd:YAG lasers.
Journal of Materials Processing Technology”, 142, pp. 517–25.
[8] Bandhopadhayay S, Sarin Sundar JK, Sunderajan G, Joshi SV. 2002, “Geometrical features
and metallurgical characteristics of Nd:YAG laser drilled holes in thick IN718 and Ti-6Al-4V
sheets”, Journal of Materials Processing Technology;127, pp. 83–95.
[9] French PW, Hand DP, Peters C, Shannon GJ, Byrd P, Watkins K, “Investigation of Nd:YAG
laser percussion drilling process using factorial experimental design”. In: Proceedings of
ICALEO 1999. 1999. p. 51–60.
[10] G.K.L. Ng , L. Li, “The e!ect of laser peak power and pulse width on the hole geometry
repeatability in laser percussion drilling”, Optics & Laser Technology 33 (2001) 393–402.
[11] Reza Negarestani , Lin Li , H. K. Sezer , David Whitehead & James Methven; Nano-second
pulsed DPSS Nd:YAG laser cutting of CFRP composites with mixed reactive and inert gases; Int
J Adv Manuf Technol (2010) 49:553–566.
[12] Maghsoodloo, S., Ozdemir, G., Jorden, V., and Huang, C. H. Strength and limitations of
Taguchi’s contributions to quality, manufacturing, process engineering. J. Mfg. Syst., 2004,
23(2), 73–126.

235
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18 th and 19th April, 2013, NITTTR, Kolkata

Effect of Machining Parameters on the Surface

Roughness of Mg/Y2O3 Nano-composite

Sanjib GOGOI, Jitul SAIKIA, Naumeet MEENA and Ashis MALLICK*

Department of Mechanical Engineering, Tezpur University, Napam – 784 028, INDIA

Ph: +91 9577248832, Fax: +91 3712 267005/6, E-mail: mal123_us@yahoo.com

Abstract: The aim of this study is to investigate the effect of machining parameters on the
surface roughness of Mg/Y2O3 nano-copmposite materials. Bulk magnesium nano-composites
(Mg reinforced by 2 % Y2O3 nano-particulates) are used as work materials. The nano-composite
was prepared by powder metallurgy technique incorporating energy efficient microwave
assisted rapid sintering process followed by hot extrusion. Machining experiments were
conducted using Taguchi methodology to ascertain the effects of machining parameters. ANOVA
analysis were carried out to optimize the machining parameters such as cutting speed, depth of
cut, and rake angle for accruing better surface finish. The effects of nano-particle as
reinforcement and Micro structural changes in the machined samples were studied through
SEM micrographs. The surface roughness was measured by portable surface roughness tester
(Surftest SJ-210). The results are compared with that of monolithic Mg sample prepared by the
same route.

Keywords: Nano-composite, Powder metallurgy, machining parameters, surface roughness,

Taguchi method.

1. Introduction

The metal matrix composite (MMC) of magnesium (Mg) and its alloys are novel, advanced class
of materials and getting significant attention in the material science community due to their
attractive properties like light weight (density ~ 1.74 gm/cm3), high specific strength with a
236
 RTMST
density and toughness [1-3]. Among metal matrix composites, magnesium based composites are
being studied intensively because of their potential applications in the automotive and
aerospace industries, where weight savings and resulting increase in fuel efficiency are very
important. However, some drawbacks are observed in Mg based materials due to limited slip
systems for dislocation movement in its hexagonal closed packed (HCP) structure that are often
restricted the use of this metal in engineering design especially for stressed application. Mg also
suffers its inherent deficiencies such as low ductility and poor corrosion resistance. To overcome
these drawbacks, Mg matrix and its alloys are reinforced by intermetallic powder particles of Ti
[4], A2O3 [5], SiC [6] and CNTs [7] or reduced the grain size to nano scale [8]. The concept behind
this is to restrict the dislocation motion either by the large grain boundary network or by hard
nano particles. The softening and corrosion behaviour of these materials pose a considerable
challenge in machining. Moreover, Mg and its alloys are highly temperature sensitive and rapid
degradation of mechanical properties and microstructural changes are observed at the
temperatures above to room temperature [8,9]. Furthermore, due to temperature rise during
the machining process, the chips of Mg materials can be ignited suddenly which leads to fire
problem. Continuous wearing of tool, the breakage of tool and chip disposal from the surface of
the machined material are the problems associated with machining of Mg composites. Thus, the
knowledge of exact machining conditions and parameters are of great interest for the
researcher to achieve better machining quality of these materials.

The machining surface roughness of metal matrix composite significantly differs from the
conventional metals and alloys [10]. During machining, the presence of reinforced hard particles
in the composite may effect in the quality of the surface finish. While, in every industrial and
structural application, the machining of composite is necessitated to produce near net shaped
components. Hence, better surface finish is demanded in the machined components. The
surface finish can be improved by controlling the machining parameters like rack angle of the
tool, cutting speed and cutting depth. If the quality of surface finish is not good then the work
materials wear more frequently which effect on the friction on the friction coefficient. Due to
bad surface finish irregularities in the surface may occur from the nucleation sites that crack and
corrosion may happen easily. Hence, better surface finish is demanded in the machined
components.

The aim of the present study is to synthesize the Mg/Y2O3 nano-composite and thoroughly
investigate the effect of machining parameters on the surface roughness of this composite.
Machining experiments were conducted using Taguchi methodology to ascertain the effects of
machining parameters. ANOVA analysis were carried out to optimize the machining parameters
such as cutting speed, depth of cut, and rake angle for accruing better surface finish. The results
are compared with the monolithic Mg sample prepared by the same route.

237
 RTMST

2. Experimental procedure

2.1 Materials

Pure commercial elemental powders of magnesium of 98.5% purity with an average particle size
of 60–300 m (Merck, Germany) was used as the matrix material. The Mg matrix was reinforced
by nanosized (30–50 nm) yttrium oxide (Y2O3), supplied by Inframat Advanced Materials, USA)
powder. Y 2O3 was chosen due to its potential to strengthen the Mg along with high thermal
stability at elevated temperatures. Good thermal stability of Y2O 3 with superior resistance to
aggressive chemical reaction, corrosion and thermal shock provides minimal reaction between
matrix and reinforcement which leads to good interfacial integrity [11]. In this study, we only
considered 2 wt% Y2O3 as reinforcement.

2.1 Processing

Bulk sample of pure Mg and Mg/Y2O 3 composite containing 2.0 wt.% of yttria powder were
synthesized by powder metallurgy technique. Air controlled weighing machine was used to
measure the respective wt.% of Mg powder and nano sized Y2O 3 powder. The mixture of
weighed powders was then shaken carefully inside the sealed plastic container followed by
blending to obtain a homogeneous mixture. The blending was performed in a RETSCH PM-400
mechanical alloying machine operated at 200 rpm in both clockwise and anticlockwise direction
for 1 hr (Fig.1a). It is to be noted that during blending no balls or process control agent was
used. After blending, the mixture of Mg and Y2O3 powders were then consolidate using an
isostatic pressure of 97 bar (50 tons) and produced a cylindrical billet of 35-mm diameter and
40-mm height (Fig.1b). The compacted billets were sintered using an innovative hybrid
microwave sintering technique [12 -14]. The sintering was required to fuse the adjacent
particles together. It is to be noted that before the sintering processes start, the colloidal
graphite sprayed on the compacted billets to protect the Mg from oxidation during sintering.
Finally, 7-mm diameter cylindrical rods were extruded from the billet through hot extrusion
process with an extrusion ration 25:1 using 150 tons hydraulic press. By adapting the similar
procedure (except that blending step was not used) bulk pure Mg 7-mm diameter rods were
produced.

238
 RTMST

(a) (b)

Fig.1. (a) Milling machine for the preparation the mixture of matrix and the hard ceramic particle
(Y2O3) and (b) compaction machine to compact the powdered into a cylindrical billet.

2.2 Microstructural studies

Microstructural studies were carried out on polished reinforced and monolithic pure Mg
samples to study the grain size, grain morphology, presence of porosity and the presence and
distribution of reinforcements. The analyses were performed using Hitachi FE-4300 field
emission scanning electron microscope (FESEM) equipped for energy dispersive X-ray
spectroscopy (EDS), an Olympus metallographic optical microscope and Scion Image Analyser
software. The constituent of phases and the texture changes were studied by X-ray diffraction
(XRD) analysis. The XRDs were performed on the polished extruded composite and monolithic
samples. An automated Shimadzu LAB-X XRD-6000 diffractometer was used. The machine was
operated at 30 kV and 40 A with Cu K radiation ( = 1.54056 Å) and scanning speed of 2
degree/min.

239
 RTMST

2.3 Machining process

In this study a 7-mm diameter Mg/Y2O3 composite and pure Mg rod prepared by powder
metallurgy route were used for machining in a high precision lathe machine, model umber NH
26 manufactured by HMT. The technical specifications of the lathe machine are (a) type of bed is
straight, (b) swing over bed is 575 mm, (c) swing over carriage wings is 545 mm, (d) swing over
cross slide is 355 mm, (e) Swing in gap is 800 mm, (f) distance between centers is 2000 mm, and
(g) power of main monitor is 11 kW. Various cutting speeds and feeds were employed for
machining. Machining processes were performed at different rack angle. The rack angles were
changed by the grinding of cutting tool. The rack angles were measured from the image of
grinded cutting tool with the help of plot-digitizer software.

2.4 Measuring of surface roughness

The surface roughness of the machined sample of composite and pure Mg was measured by the
SJ-210 Surftest machine. Surface roughness (Ra) is the measure of finer surface irregularities and
is measured by the high frequency and short wavelength component of the measured surface.
A diamond stylus was employed in the machine which moved at a constant rate across the
surface, perpendicular to the lay pattern. The rise and fall of the stylus is detected electronically
that are amplified and recorded on a strip chart, or processed electronically to produce readings
for a meter. To measure the surface roughness, the Surfest machine was connected in lathe
machine in such a way that the diamond stylus was always touched with the machined sample.
The machined sample rotate and the Surfest machine measured the roughness of the sample.
Fig.2 shows the complete attachment of Surfest machine, lathe and the machined sample.

240
Fig.2. Measuring of surface roughness by SJ210.
 RTMST

3. Taguchi Technique

For estimating the correct parameters in designing procedure, the conventional approach is
complicated, laborious and cumbersome as the number of processed parameters is increased.
Taguchi technique is a systematic optimization approach to optimize the various parameters
with regard to best performance, quality and cost effective [14–15]. Genichi Taguchi, a Japanese
scientist first introduced this technique to improve the quality of manufactured products [16],
and more recently this technique is also applied in the various field of engineering,
biotechnology, marketing and advertising. The philosophy of Taguchi gives two tenets: (a) the
reduction in variation of a product or process that represents lower loss to the society, and (b)
the proper development strategy that can intentionally reduce variation. By adopting Taguchi
technique, we can study the machining parameters and the surface roughness with minimum
number of experiments. In machining process, this technique offers a systematic way for
acquiring data to obtain the information about the character of the process. This technique is
very effective over the other optimization technique because of their robustness in reduction of
the effort for conducting experiments. This technique considerably reduced experimental time
with least production cost and discovers significant factors in machining operation. Teguchi
technique suggests to analysis the mean response for each run in the inner array and the
variation with the help of signal – to – noise (S/N) ratio. S/N ratio, the statistical analysis of
variance (ANOVA) was carried out for the analysis of the influence of process parameters and
the level of a desired signal to the level of background noise. ANOVA is very demanding because
of industrial need for the comparison of multiple samples, and it is also computationally very
elegant and robust against the violation to its assumption. It is defined as the ratio of signal
power to the noise power. A ratio higher than 1:1 indicates more signal than noise.
Confirmation tests with the optimal levels of machining parameters were considered. During the
study of ANOVA with certain test error, the estimation of the variance of error is very important.
In most common interest for optimising the problem, the S/N ratio characteristics can be
divided into three categories as follows:

241
 RTMST

3.1 Nominal is best characteristic

In Taguchi’s S/N ratio for a nominal-the-best characteristic is pre-processed as,

 y 
S / NT  10 log 
 s 2 y 
 
(1)
where y is the observed data, y is the average data, s2y is the variance of y, and n is the number of
observations.

3.2 Smaller is best characteristic

In smaller-the-best, the data sequences for S/N ratio is pre-processed as,

1 n 2
S / N L  10 log  yi
n i 1
(2)

3.2 Larger is best characteristic

In this category, data sequence for S/N ratio is expressed as,

1 n 1
S / N L  10log 
n i 1 yin
(1)

The analysis of the data was performed by analysis of Mean S/N Ratio using MINITAB software
to study the contribution of the factor and interactions and to explore the effects of each
process on the observed value.

4. Results and discussion

4.1 Density measurement

The densities of composite and pure Mg of the extruded samples were estimated as 1.756 and
1.736 respectively. These results gives almost same of that theoretical density. However,
composite exhibit little higher density as compared to that of pure Mg sample prepared by the
242
 RTMST
same powder metallurgy route. For both the sample, the volume percentage of porosity was
observed less than 1%.

4.2 Microstructural study

In the composite sample, the distribution of Y2O3 particles are mostly uniformed as discussed in
our published paper [12]. The identity of the Y2O3 particulates was confirmed through XRD
analysis. Few nanopores were observed in all the composite samples. Fig. 3(a) and Fig. 3(b)
represent the optical image of extruded sample of composite and pure processed by mirror
polishing followed by proper etching. The grain sizes were calculated about 18 m and 20 m
for composite and pure Mg respectively. This result reveals that the present of nano particulates
reduce the grain in the composite. The presence of nano reinforcement particulates act as an
obstacle to the grain growth process during preparation of bulk composite sample. Near-
equiaxed grain morphology was observed for both monolithic and reinforced samples.

(a) (b)

Fig. 3. Microstructure of polished and etched extruded samples obtained from optical
microscope: (a) pure Mg and (b) nanocopmosite.

243
 RTMST

4.3 Surface analysis

Laser or stylus type profile-meters are normally used to measure surface roughness i.e the
texture of the surface. This is calculated by the vertical deviations of a real surface from its
actual surface before machining. Fig.4 represents the amplitude distribution curve for (a) pure
Mg and (b) Mg/Y2O3 nano-composite after machining. The cutting speed for the machining was
taken at 192 r.pm and the rake angle of cutting tool was +8 for both the cases. It can be
observed that the peak value of amplitude is lower for composite than that of pure Mg.

(a) (b)

Fig.4. Amplitude distribution curve (ADC) of the surface of (a) pure MG and (b) Mg/Y2O 3 nanocomposite
after machining.

Table 1 represents the comparison of S/N ratio of pure Mg and Y2O3 nano-composite. S/N ratios
were estimated at different rake angle and cutting speed. It can be observed that the rake angle
significantly effect on the S/N ratio. Positive (+) rake angle gives higher value of S/N ratios for
both the Mg and composite. Furthermore, higher S/N value can be seen in nano-composite as
compare to pure Mg in all the cases except the for rake angle of 8. S/N ratio also depends on
the cutting speed. For +4 rake angle, the higher value of S/N can be seen either cutting speed at

244
 RTMST
68 r.p.m (lower value) or at 325 r.pm (maximum value). However, at 8 rake angle the S/N ratio
increased with the increase of cutting speed.

Table 1. S/N ratio of pure Mg and composite at different rake angle and cutting speed.

Speed S/N ratio of pure Mg S/N ratio of Mg/Y2O3

(r.p.m) +4 4 +8 8 +4 4 +8 8
68 -51.6692 -48.8125 -54.848 -50.5409 -52.7196 -49.1636 -55.9308 -49.5282
114 -46.0871 -50.3782 -54.4586 -51.1101 -47.5257 -49.7592 -56.3264 -49.5112
192 -44.9799 -49.0613 -53.8907 -54.2354 -47.0873 -52.9146 -55.9312 -51.2484
325 -49.2921 -49.0252 -53.5188 -54.9466 -53.9192 -50.6008 -56.5745 -54.2843

The mean values of S/N ratios are shown in Fig. 5 at different rake angles (+4, 4, +8, and
8): the top row is for pure Mg and the bottom row is for nano-composite. It can be observed
that for both the material (pure Mg and composite of Mg/Y2O 3) the mean value of S/N ratio of
the machined surface is smaller when the rake angle is negative and 4. However, exactly
opposite behaviour can be seen when the rake angle is 8. The presence of hard particles of Y2O3
in metal matrix composites results in higher value in the mean of S/N ratio except machining by
the tool with 8 rake angle.

(I)

(II)

(a) (b)

Fig. 5. Mean value of S/N ratio at different rake angles: (a)  4 and (b)  8. The top row
245
represents for pure Mg (I) and the bottom one is for nanocomposite (II).
 RTMST

5 Conclusions

We systematically synthesised the Mg/Y2O3 nano-composite using powder metallurgy

technique. The microstructural analysis reveals that the presence of nano Y2O3 particles reduces
the grain size in the extruded nano-composite of Mg/Y2O3. Machining of Mg and its composite
composites is quite difficult as compare to conventional materials, and lead to the damage of
surface and subsurface. However the damage can be reduced and the surface integrity can be
improved significantly by optimising the rake angle of the cutting tool, cutting speed and depth
of cut. The volume fraction of the hard reinforcement Y2O3 particulates has a great influence on
the quality of surface finish in the machined region. It has been observed that the composite
exhibit the better surface finish when compared with surface finish of matrix metal for same
cutting conditions and parameters. The mean S/N gives higher value for Mg/Y2O3 nano-
composite when compared with pure Mg except machining by the tool with 8 rake angle.

Acknowledgements

The bulk Mg and nano-composites were prepared in the Department of Mechanical Engineering,
National University of Singapore. The authors thank Dr. Srikanth Vedantam, IIT Chennai and Dr.
Gupta Manoj, National University of Singapore for providing the laboratory facilities of
Mechanical Engineering Department, NUS for preparing the Mg/Y2O3 nano-composite material.

References

[1]. A. Evans, C.S. Marchi, A. Mortensen, Metal matrix composites in industry: an introduction
and a survey, Boston: Kluwer Academic Publishers, 2003.
[2]. L. Lu, K.K. Thong, M. Gupta, Mg-based Composite Reinforced by Mg2Si, Compos sci Technol.,
Vol.63, 2003, pp. 627-623.
[3]. P. Lukac, Z. Trojanova. Magnesium-based Nanocomposites, Int. J. Materials and Product
Technology, 23, 2005, 121-137.
[4]. P. Perez, G. Garces, P. Adeva, Mechanical Properties of a Mg-10 (vol.%)Ti
Composite, Comp Sci Technol Vol. 64, 2004,145-151.
[5]. S.F. Hassan, M. Gupta, Development of High Performance Magnesium Nanocomposites
Using Solidification Processing Route, Mater Sci Technol., Vol. 20, 2004, 1383-1388.
[6]. L. Lu, C.Y.H. Lim, W.M. Yeong, Effect of Reinforcements on Strength of Mg9%Al
Composites. Comp. Struct., Vol. 66, 2004, 41-45.

246
 RTMST
[7]. C.S. Goh, J. Wei, L.C. Lee, M. Gupta, Development of Novel Carbon Nanotube Reinforced
Magnesium Nanocomposites Using the Powder Metallurgy Technique, Nanotechnology,
Vol. 17, 2006, 7-12.
[8]. A. Mallick, S. Vedantam, L. Lu, Grain Size Dependent Tensile Behavior of Mg-3%Al Alloy
at Elevated Temperatures, Mater. Sci. Eng. A, Vol 515, 2009, 14-18.
[9]. M.M. Avedesian, H. Baker, Magnesium and Magnesium alloys, ASM International,
1999.
[10]. R. Komandhuri, Machining fibre-reinforced composite, Mechanical Engineering,
Vol. 115. 1993, 58-66.
[11]. F. Thummler and R. Oberacker, An Introduction to powder metallurgy, Ins. of
Materials,London, 1993.
[12]. A. Mallick, K. S. Tun, S. Vedantam, M. Gupta, Mechanical characteristics of pure Mg and a
Mg/Y2O3 nanocomposite in the 25–250 C temperature range, J. Mater Sci., Vol. 45, 210, 3058–
3066.
[13]. A. Mallick, K.S. Tun and M. Gupta, Deformation behaviour of Mg/Y2O3 nanocomposite at
elevated temperatures’ Mat. Sci. & Engg. – A, Vol. 551, 2012, 222 – 230.
[14]. P. J. Roos, Taguchi techniques for quality engineering, McGraw Hil, New York,
1995.
[15]. S. H. Park, Robust design and analysis for quality engineering, Chapman & Hall,
London, 1996.
[16]. G. Taguchi, Quality engineering (Taguchi methods) for the development of
electronic circuit technology. IEEE Trans. on Reliability, Vol. 44, 1995, 225 – 229.

247
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18 th and 19th April, 2013, NITTTR, Kolkata

MICRO-HOLE DRILLING ON THIN SHEET METALS

BY MICRO-EDM

Trisha BISWAS*, Amit Kumar SINGH, P. K. PATOWARI

Department of Mechanical Engineering
National Institute of Technology Silchar
P.O. Silchar 788010, Assam, India
[e-mails: biswastrisha39@gmail.com, amit.kumar965@yahoo.com, ppatowari@yahoo.com]
(*Corresponding author)

Abstract: Micro Electrical Discharge Machining (MEDM) is one of the basic technologies for the
production of miniaturized components. It can be used to create micro holes with good accuracy
and less over cut. In this work, micro holes have been drilled on thin sheet of brass using micro
tools in the form of tungsten rods with diameter 300 m. A detailed study of machining
parameters and their effects have been carried out. A micro hole of diameter 309.2 m has
been successfully drilled with a tool of 300 m diameter. It has been observed that the over cut
increases with an increase in the voltage and capacitance.

Keywords: Micro-electrical discharge Machining, Electrode, Micro hole

248
 RTMST

1. Introduction

The increasing requirement for smaller components in various industrial domains has given rise
to a need to formulate sub-micrometer and even nanometer level machining technologies.
Miniaturization is demanding in the variety of manufacturing industries because of the growing
needs of micro devices
for mechanical, electronics and medical applications. Therefore, many methods for micro-
machining have been adapted such as traditional machining, laser beam machining, ion beam
machining, electron beam machining, ultrasonic machining, electrochemical machining, and
electrical discharge machining based on their principle. Electrical discharge machining (EDM) is
a well-known technique for micro fabrication with exceptional ability to maintain high level of
precision and accuracy. Especially, electrical discharge machining (EDM) is a non-contact
thermal process, which has advantages such as high-precision machining of conductive
materials regardless of material hardness. The basic mechanism of the micro-EDM process is
essentially similar to that of the EDM process with the main difference being in the size of the
tool used, the power supply of discharge energy and the resolution of the X-, Y- and Z-axes
movement [1]. The material removal process of micro-EDM is mainly based on the erosion of
electrically conductive materials through the series of spatially discrete high-frequency electrical
discharges (sparks) between the tool and the workpiece at the presence of dielectric. Micro-
EDM has been found to be a suitable option to satisfy the increasing demand of miniaturization
and found to be a cost-effective process for fabrication of micro-tools, micrcomponents and
micro-features with good dimensional accuracy and repeatability.

Micro-holes are widely used in micro-electromechanical systems (MEMS) serving as channels or

nozzles to connect two micro-features, delivering media in MEMS, or exchanging media outside
of MEMS. In products of aerospace, automobile and biomedical science industries, it is essential
for materials used in fuel nozzles and micro-molds to resist wear, high temperature and high
pressure. These requirements cause difficulty in processing by most micro manufacturing
methods. Micro-EDM, as one of the micromachining processes, has demonstrated its ability to
drill micro-holes in metallic alloys.

Asad et al. used a micro electro discharge & micro turning fabricated electrode for boring
controlled holes on 50µm thick stainless steel & four controlled holes of diameter 6.5, 8.5, 10.5
and 12.5 µm were machined [2]. Weng et al. investigated that a graphite microelectrode can be
machined to approximately 50 μm diameter using supersonic agitation of the anode by
electrolysis anodic etching through which micro parts of different materials with arrays of micro
holes on the coaxial circles can be manufactured [3]. Nakaokua et al. investigated round Ø 50µm
micro-holes can be machinable in sintered diamonds with the diamond particles of 1–20 µm [4].

249
 RTMST
Bamberg et al. showed that by orbiting of the electrode is helpful in generating more uniform
hole geometries and greatly improved bottom qualities for blind holes [5]. Jahan et al. observed
that cemented tungsten carbide (WC-Co) was found to be better capable of producing good
quality micro holes compared to austenitic stainless steel (SUS 304) [6]. Yahagi et al.
investigated in micro drilling EDM that the influence of high spindle using the electrostatic
induction feeding method, high aspect ratio holes can be machined with lower tool wear ratio
with 50000rpm spindle speed [7].

2. Experimental Work

The experiments have been conducted on Hyper-15 Micro Electrical Discharge Machine, it is a
table-top type miniature Machine Tool of Sinergy nano systems, Navi Mumbai, India. The
workpiece used for the experiments have been brass sheet having thickness 40 µm and
tungsten as a tool material having 300 µm diameter. The hydrocarbon oil was used as a
dielectric fluid and other parameter are shown in Table 1.

Table 1. Experimental parameters for MEDM

Sl. No. Parameters Values

1 Dielectric Medium Hydrocarbon Oil

2 Work piece Material Brass Thin Sheet

3 Work piece Thickness 40 µm

3 Electrode Material Tungsten

4 Electrode Size 300 µm Circular

5 Voltage (volt) 75, 100, 125, 150

6 Capacitance (pF) 33, 100, 1000, 10000

7 Polarity N, R

Before experimentation, the bottom of the tool electrode is polished using a very fine grade
emery sheet before every experiment. The workpiece was kept on the machine table using a
fixture shown in Fig 1.

250
 RTMST

(a) (b)

Fig. 1. (a) Photograph of Multipurpose miniature tool, (b) Schematic diagram

A number of experiments have been carried out in order to study the effects of process
parameters on output measure i.e. hole diameter. During this work, two process parameters,
voltage and capacitance are varied in 4 different levels, and experiments have been
conducted for all possible combinations at each polarity. Hence thirty two numbers of
experimental runs are conducted as shown in Table 2.

After completion of all experiments the diameter of hole is measured using Leica DM 2500M
optical microscope. Hole diameter was measured at two places by making a circumscribed
circle and inscribed circle and its mean value is considered as ‘actual diameter’ of that
particular parameter conditions. Table 2 shows the experimental result of the different
parameter conditions.

251
 RTMST
3. Effect of process parameters on the response

The effect of Micro-EDM parameters such as voltage, capacitor and polarity are analyzed
through two-dimensional plots. These plots are useful to visualize the effect of experimental
process parameters such as voltage, capacitor and polarity on the response measure, hole
diameter. From the results as shown in Fig. 2, it is observed that capacitance is the most
significant parameter for hole diameter followed by voltage. Polarity has least effect on hole
diameter. Micro-hole diameter increases due to over cut and it depends on capacitance value.

Figure 2 (a) & (b) illustrates the effect of voltage on hole diameter at different capacitance value
and the variation of hole diameter with regard to capacitance value for various voltage is
depicted in Fig. 2 (a) & (b).

252
 RTMST

Table 2. Experimental Results

Expt. Operating Parameters

Hole Diameter (µm)
No. Voltage (volt) Capacitor (pF) Polarity

1 75 33 N 317.2
2 100 33 N 327.8
3 125 33 N 328.5
4 150 33 N 313.8
5 75 100 N 316.9
6 100 100 N 320.4
7 125 100 N 318.9
8 150 100 N 322.7
9 75 1000 N 321.6
10 100 1000 N 327.1
11 125 1000 N 329.4
12 150 1000 N 328.7
13 75 10000 N 335.0
14 100 10000 N 341.2
15 125 10000 N 343.6
16 150 10000 N 347.7
17 75 33 R 309.8
18 100 33 R 309.2
19 125 33 R 312.2
20 150 33 R 315.3
21 75 100 R 316.1
22 100 100 R 321.6
23 125 100 R 320.8
24 150 100 R 318.7
25 75 1000 R 330.4
26 100 1000 R 330.7
27 125 1000 R 333.4

253
 RTMST
28 150 1000 R 334.0
29 75 10000 R 344.2
30 100 10000 R 349.1
31 125 10000 R 355.4
32 150 10000 R 347.5

370
33pF
Hole Diameter (micro-meter)

360 100pF
1000pF
10000pF
350

340

330

320

310

300
70 80 90 100 110 120 130 140 150
Voltage (volt)
(a)
370
33pF
Hole Diameter (micro-meter)

360 100pF
1000pF
10000pF
350

340

330

320

310

300
70 80 90 100 110 120 130 140 150
Voltage (volt)
(b)

254
 RTMST
350
75 volt

Hole Diameter (micro-meter)

345 100 volt
125 volt
340 150 volt

335

330

325

320

315

310 1 2 3 4
10 10 10 10
Capacitor Value (pF)
(c)
360
75 volt
Hole Diameter (micro-meter)

100 volt
350 125 volt
150 volt
340

330

320

310

300 1 2 3 4
10 10 10 10
Capacitor Value (pF)
(d)
Fig.2. (a) & (c) Normal polarity, (b) & (d) Reverse polarity
Comparison of the size of micro hole has been analyzed, and it’s found that the size of micro
hole increases with the increases of voltage while both capacitor and polarity (Normal or
Reverse) are kept constant. It is shown in Fig 3.

Voltage (volt)
Capacitance
=1000 pF
75 100 125 150
Normal
Polarity

321.6 µm 327.1 µm 329.4 µm 328.7 µm

255
 RTMST
Reverse

330.4 µm 330.7 µm 333.4 µm 334.0 µm

(a)

340
Normal polarity
Hole Diameter (micro-meter)

Reverse polarity

335

330

325

320
70 80 90 100 110 120 130 140 150
Voltage (volt)
(b)
Fig. 3. (a) Optical micrograph of micro hole, (b) at 1000pF its effect are shown in graph.

After completion of all experiments, the minimum micro hole of Ø 309.2 is achieved at
particular parameters i.e. voltage 100 volt, capacitance 33pF, and reverse polarity. The optical
micrograph of the minimum size micro hole and the images of tool before and after machining
are shown in Fig. 4.

256
 RTMST

(a) (b) (c)

Fig.4. (a) Micro-hole of 309.2 µm diameter (b) Tool before machining (c) Tool after machining

4. Conclusions

From the experimental results and analysis the following conclusions are drawn and
summarized as follows:

1. It has been observed that the over cut increases with an increase in the voltage and
capacitance.
2. Size of the micro-hole is highly depends on capacitance value (pF) followed by voltage
(volt). Polarity have least effect on hole diameter.
3. A micro hole of diameter 309.2 m has been successfully drilled with a tool of 300 m
diameter.

References

[1]. T. Masuzawa. State of the art micromachining, Annals of CIRP2000, Vol. 49, Issue. 2,
2000, pp. 473–88.
[2]. A. B. M. A. Asad, Takeshi Masaki, M. Rahman, H.S. Lim, Y.S. Wong. Tool-based
micro-machining, Journal of Materials Processing Technology, Vol. 192–193, 2007, pp.
204–211.
[3]. Feng-TsaiWeng. Electrodischarge machining of a coaxial array of micro holes using a
graphite-copper electrode, International Journal of Advanced Manufacturing
Technolology, Vol. 27, 2006, pp. 1097–1100.
[4]. H. Nakaokua, T. Masuzawa, M. Fujino. Micro-EDM of sintered diamond, Journal of
Materials Processing Technology, Vol. 187–188, 2007, pp. 274–278.
[5]. Eberhard Bamberg, Sumet Heamawatanachai. Orbital electrode actuation to improve
efficiency of drilling micro-holes by micro-EDM, Journal of materials processing
technology, Vol. 209, 2009, pp. 1826–1834.

257
 RTMST
[6]. Muhammad Pervej Jahan, Yoke San Wong, Mustafizur Rahman. A comparative
experimental investigation of deep-hole micro-EDM drilling capability for cemented
carbide (WC-Co) against austenitic stainless steel (SUS 304), International Journal of
Advanced Manufacturing Technology, Vol. 46, 2010, pp. 1145–1160.
[7]. Yuna Yahagi, Tomohiro Koyano, Masanori Kunieda, Xiaodong Yang. Micro Drilling
EDM with High Rotation Speed of Tool Electrode Using the Electrostatic Induction
Feeding Method, Procedia CIRP 1, 2012, pp. 162 – 165.

_____________

258
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18 th and 19th April, 2013, NITTTR, Kolkata

A policy making framework for resilient supply chain

systems

Anupam HALDAR1, Amitava RAY2, Debamalya BANERJEE3, Surojit GHOSH4

1 Mechanical Engineering Department, Netaji Subhash Engineering College, Kolkata 700152,

India

2 Dept. of Mechanical Engineering, National Institute of Technology, Silchar, Assam–788010,

India

3 Production Engineering Department, Jadavpur University, Kolkata 700032, India

4 The Institution of Engineers (India), Kolkata 700020, India

Abstract: The aim of this research is to develop a Risk Management-based Decision Analysis
(RMDA) framework on the basis of common fundamental elements that define the nature of
disruption that threaten Supply Chain Systems (SCS). Considering a systems perspective, the
roots of uncertainty in supply chain system is categorized into two major groups such as a)
Natural disasters, and b) Man-made disasters. Initially, a classic risk assessment methodology is
applied in this research work, to understand the nature of uncertainties through identifying,
analyzing and prioritizing risks in the SCS. Most of the important design aspects of SCS address
cost-effectiveness and functionality. But, there is a need to assess the value of decisions as
incorporating resilience is a prominent long-term attribute of a system. The RMDA guides the
decision-makers to define ways for risk mitigation and plan for contingencies. The structure of
system is categorized into three major phases: shock prevention, vulnerability reduction, and
response preparedness. From a systemic point of view, the responsiveness of SCS is considered
in two distinctive phases i.e., before and after of facing the disruption which may be referred as
prevention and recovery phases of a system. Secondly, cause-and-effect diagram is developed

259
 RTMST
to create a tree of events and their effects, which can be used to model the devising strategies
for resiliency. Finally, the adopted resilience investment plans are evaluated utilizing a Decision
Tree Analysis (DTA) methodology for assessing the cost-effectiveness. The proposed concept has
a huge impact on the approach to project investment and can be incorporated in analysis of
uncertainties that threaten the functionality of an infrastructure. This research work is an effort
to develop a systematic process for making strategic and investment decisions. The proposed
RMDA framework addresses the process of decision-making for the prevention phase based on
likelihood of disruption and its consequences.

Keyword: resiliency, RMDA, decision tree analysis, fishbone diagram

1. Introduction
Profitability of company is often determined by how it reacts to unforeseen disruptions. The
ability of organizations to measure and track the impacts of random disruptions, as well as
changes in trends over time are important to effectively manage and control supply chain
disruptions, Bosman [1]. Supply chain face a variety of operational and environmental
uncertainties that can disrupt their manufacturing, operation, delivery, transportation etc. that
potentially resulting in billions of dollars of direct and indirect financial losses. These threats can
range from natural disasters such as hurricanes, cyclones and floods to man-made disasters like
terrorist attacks and strikes, Nilchiani et. al. [2]. In our study the domain of supply chain
disruptions starts from receipt of an order, order placement, production until a part is delivered
to the customer – along the value chain”. Companies do not have the right raw materials at the
right time and location when required, subsequent workstations come to a standstill. Loss of
productivity is the key outcome for most supply chain disruptions. Leading sources of
disruptions vary from industry to industry. Resiliency is the physical property of a material to
return to its original shape or position after deformation that does not exceed its elasticity limit,
or a rebound back to its initial condition. Haldar et. al. [3, 4, 5] developed several supplier
selection approach by integrating AHP, TOPSIS, QFD and Entropy method to recover from
disruption in a supply chain network (SCN). The RMDA framework [6, 7] has not been used
earlier for strategic decision to invest in developing a supply chain resiliency.

2. RMDA framework in practice of a SCN:

A framework is a basic conceptual structure, which applies systems thinking, logic, and a variety
of tools to frame and potentially solve complex issues. The proposed RMDA framework
addresses the process of decision-making for the prevention phase based on likelihood of
disruption and its effects. The application of Decision Tree Analysis (DTA) methodology is

260
 RTMST
proposed for assessing cost-effectiveness of alternative strategies [6]. The proposed RMDA
framework consists of three phases, which include: assessing vulnerabilities, devising resilience
strategies, and valuing investment strategies.

2.1. Assessing vulnerabilities at the SCN

In the manufacturing supply chain systems, the process of assessing vulnerabilities includes the
identification of threats associated with the infrastructure; and development of prevention,
contingency, and emergency plans that enables systems to maintain functionality. Based on a
systems perspective, we categorize the roots of uncertainty in SC systems into two major groups
as:

 Natural disasters,
 Man-made disasters.
Some major threats of supply chain under these categories are identified. The result of our
general investigation on this matter is illustrated in a cause-and-effect diagram in Fig. 1.

Natural Hazards

Atmospheric Hazards Hydrologic Hazards Geologic and Seismic Hazards

Cyclones Earthquake
Tsunami
Hurricanes Floods

Disruption in
Supply Chain
System
Sabotage
Absenteeism
Strikes

Terrorist Attacks Industry actions

Man-made Hazards

Fig.1: Fishbone diagram

Some key vulnerability in both of the man-made and natural category are identified and are
enlisted in Table 1, only based on subjective opinion of the authors. Moreover the likelihood and
impact of each threat are rated in a 1–5 scale (in which 1 is the lowest and 5 refers to the
highest intensities) on the basis of a risk management approach. The risk factor represented in
Table 1 is simply calculated by multiplication of likelihood and impact of each threat. The

261
 RTMST
estimated costs associated with the selected threats in million dollars is calculated in Table 2,
where other values are estimated according to the subjective judgment of the authors to
present an example on applications of the proposed RMDA framework in a real life case. The
subject matter experts at SCN must prepare a similar table based on relevant data and accurate
information following the approach that is presented here.

Table1. Analysis of disruption:

Category of Threat Likelihood Impact Risk

vulnerabilities (1-5 scale) (1-5 factor,
scale) RF=L*I

Natural N1-Hurricanes affecting the infrastructure 1 5 5

N2-Cyclones affecting the infrastructure 2 3 6

N3-Flood affecting the infrastructure 4 3 12

N4-Tsunami affecting the infrastructure 2 5 10

N5-Earthquake affecting the

262
 RTMST
infrastructure 1 2 2

Man-made MM1- Sabotage 1 4 4

MM2-Strikes 2 4 8

MM3-Absenteesm 3 2 6

Table2. Monetised loss for different types of disruption

Threats Estimated Estimated Estimated Total monetized

infrastructure capital operational day operational loss per loss
loss loss day

N1 $300 90 $4.5 $705

N2 $200 90 $4.5 $605

N3 $650 20 $4.5 $740

N4 $800 60 $4.5 $1070

N5 $12 20 $4.5 $102

MM1 $350 90 $4.5 $755

MM2 $35 6 $4.5 $62

MM3 $16 3 $4.5 $29.5

2.2. Devising resilience strategies for the SCN

Three possible strategies are identified by the decision makers. One strategy might be to
transfer Work-In-Progress (WIP) from the affected site/s to non-effect sites to complete or
partially complete the production operations. But this may not be possible if the intensity of
disruption is very high. Another strategy might be to hold strategic inventory stocks to other
non-affected nodes and utilize a virtual manufacturing cell to maintain the pace of market
supply in the event of an excursion. Such a strategy would require the investment in capacity
buffers in the supply-chain. Yet another strategy, such as in the extreme condition of natural
disaster is to outsource the items to non-affected zones which may be more expensive means.
Last two resilience strategies are identified for our study of policy making framework of a SCN
263
 RTMST
which is shown in Table 3. All the numbers are represented in Table 3 are in million dollars.
Calculating the cost of each resilience strategy requires a complicated process. The authorities
and experts within the industry must calculate the amounts of investment for each resilience
strategy using time value of money. The authors of this paper use their subjective judgment to
estimate the costs associated with each proposed resilience strategy that is presented in this
phase only to exemplify the application of the proposed RMDA framework in a real life supply
chain system.

Table3. Devising resilience strategies

Threats Resilience Strategies Estimated cost of

strategy

N1

N2 R1-Outsourcing $235

N3

N4

N5

MM1

MM2 R2- Virtual cellular Manufacturing $56

MM3

264
 RTMST

2.3. Valuing investment strategies at the SCN

Decision tree analysis is used to calculate the expected value of cost facing disruptive events
explained in Table 1 and make investment decisions based on estimated costs of resilience
strategies. A structured approach of DTA allow us to run sensitivity analysis by changing
estimated costs, probabilities of occurrence, and all other variables as well as scenario analysis
by making changes in premises and suppositions. Calculations in DTA are based on the concept
of expected value, which is the cost multiplied by the probability of occurrence for each one of
the options. There are two kinds of nodes used in DTA: decision nodes that are shown with a
square, and probability nodes that are presented with a circle. While the decision tree should be
developed from a decision point (from left to right), calculations fold back from the farthest tips
of each branch (from right to left). Branches coming out of a decision node are representative of
alternatives available to decision-makers and those emanated from chance nodes are
representing all the possible outcomes. Specifying the chances for different outcomes requires
probabilities of occurrence. Probabilities used in these decision trees are chosen based on
subjective judgments of the authors. The numbers at the farthest right side of each branch are
inserted from the columns of ‘‘Total Monetized Loss’’ and ‘‘Estimated Cost of Strategy,’’
presented in Tables 2 and 3, respectively. The numbers represented on top of chance nodes in
Figs. 2 & 3 are the expected value of branches coming out of the decision node, which is the
sum of calculated cost times its estimated probability. In better words, the expected value of
not investing in resilience strategies to avoid risks for natural disaster and Man-made disaster
are calculated using the following equations (1) and (2).

n
EMVN   pi * Ni
i 1

(1)

n
and EMVMM   pi *MM i
i 1
(2)

where, EMV=expected monetary value

265
 RTMST
Calculated expected monetised value (EMVN) for natural disaster is $98,208,000 and that
(EMVMM) for Man-made disaster is $80,050,000 which is required for not investing in resilience
strategies. After doing the folding back calculations for all of these decision trees, the next step
in DTA approach is to compare the calculated expected value with the amount that is estimated
for R1 and R2 proposed resilience strategies for natural and man-made categories of risks within
a SCN.

The last step of the DTA approach and as depicted in Figs. 2 & 3. Based on the decision tree
approach, investing in resilience strategies of adopting R2 is less costly than accepting the
expected value of risks caused by man-made disasters while investing in resilience strategy of
adopting R1 is more costly than accepting the expected value of risks caused by natural disaster.
From cost-effectiveness point of view, outsourcing is not the suitable option in case of natural
disaster while virtual cellular manufacturing is the better option in case of man-made disruption.
Therefore, a simple analysis of DTA indicates that

N1, p1 = 10-5, $705

N2, p2 = 10-5, $605

$98.208
-4
N3, p3 = 10 , $740
Not to Invest in
-5
Resilience Strategy N4, p4 = 10 , $1070

Chance N5, p5 = 4*10-5, $102

Node
Decision
P6 = (1- p1- p2 - p3 - p4 - p5), $0
Node
Investment in
Resilience Strategy R1, $235

Fig 2. DTA of natural disaster

266

$80.05 MM1, p1 = 2*10-5, $755

Not to Invest in
Resilience Strategy MM2, p2 = 10-3, $62

Chance MM3, p3 = 10-4, $29.5

 RTMST

adopting R2 is financially more favorable to the system in long run. The expected value of the
risk caused by natural-disaster is less than the required investment budget to adopt R1. As a
result, policy-makers might not want to invest in R1 by outsourcing. But the expected value of
the risk caused by man-made disaster is more than the required investment budget to adopt R2.
So, policy-makers might want to invest in virtual manufacturing cell, with only a portion of their
investment budget, which will be more cost-effective to the industry.

4. Conclusion

Disruptions move a system away from its optimal settings so it increase supply chain costs.
There is a lack of an effective framework that incorporates a risk management-based approach
to supply chain systems in the process of policy-making on investment strategies. The proposed

267
 RTMST
risk management-based decision analysis framework identifies the common elements of
uncertainty in SCN through data analysis. The RMDA framework also helps to compare the cost
associated with probable failures caused by disruption as well as the cost of investing in
resiliency strategies. The framework is developed based on risk analysis and management
methodologies, which help to understand the nature of uncertainty in the system and
consequently enables to devise resilience strategies in regards to the known vulnerabilities of
the system. RMDA framework then uses decision-making analysis tools to choose the best policy
in regards to the devised resilience strategies.

5. References:

[1] Bosman, R. The New Supply Chain Challenge: Risk Management in a Global Economy, United
Kingdom, 2006.

[2] Nilchiani R, Mostashari A. An introduction to resilience in port infrastructure systems,

Hoboken: Stevens Institute of Technology; 2008.

[3] Haldar A., Ray A., Banerjee D. and Ghosh S., A Hybrid MCDM Model for Resilient Supplier
Selection, International Journal of Management Science and Engineering Management, Vol. 7
Issue 4, 2012, pp. 284-292. ISSN: 1750-9653.

[4] Haldar A., Banerjee D., Ray A. and Ghosh S., An integrated approach for supplier selection,
Elsevier Procedia Engineering, Vol. 38, 2012, pp.2087-2102. ISSN: 1877-7058.

[5] Haldar A., Banerjee D., Ray A. and Ghosh S., Supplier Selection in the Intuitionistic
Environment under Requirement Perspective, in 4th International and 25th AIMTDR Conference,
Kolkata, India, 14-16 December 2012, Vol.2, pp.1287-1292.

[6] Mo Mansouri, Roshanak Nilchiani and Ali Mostashari, A policy making framework for resilient
port infrastructure systems, Marine Policy Vol. 34, 2010, pp.1125–1134.

[7] Mansouri M, Nilchiani R, Mostashari A., A risk management-based decision analysis

framework for resilience in maritime infrastructure and transportation systems in 2009 Third
annual IEEE systems conference; Vancouver, BC, Canada, 23-26 March 2009, pp.35–41.

268
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18 th and 19th April, 2013, NITTTR, Kolkata

Development of A New Shakeout Machine By Applying

Value Engineering: A Case Study

Aditya1, Samiran MANDAL2

1
Mechanical Engineering Discipline, PDPM
IIITDM, Dumna Airport Road, Jabalpur, PIN-
482005 (M.P.) India,
Tel.: +91-9425363337, E-mail:
aditya.me02@gmail.com
269
 RTMST

2
Professor, Department of Mechanical Engineering,
NITTTR, Block-FC, Sector-III, Salt Lake City, Kolkata,
PIN- 700106 (W.B.), India,
Tel.: +91-33-23370099, E-mail:
samiran_m04@yahoo.com

Abstract:

Proper conveying of the materials is one of the important concerns of materials handling
in today’s competitive environment. For this purpose the application of vibration in
conveying the materials has a significant role. By introducing the vibration in conveying
the materials, the productivity of the materials handling system can be improved
significantly. In this paper two foundry machines, the shakeout machine and vibrating
(oscillating) conveyor are considered. The shakeout machine is excited by an unbalanced
vibromotor and the oscillating conveyor is driven through a separate driving mechanism.
Due to the complexity of operating two machines separately, different alternatives of
combining the functions of these machines, by applying the methodology of value
engineering, are presented. Also, a design of machine is proposed which consists of the
function of both the machines. The results show a very little deviation of average
conveying velocity from the standard values.

Keywords: Shakeout machine; oscillating conveyor; materials handling; value

engineering

1. Introduction:
1.1 Materials Handling and Vibration:
Metal casting is a major manufacturing process, where intricate shapes can be produced
easily which cannot be produced by machining. Sand casting is a very old process and
extensively used in manufacturing industries, where large shaped parts are being
manufactured. For sand casting a lot of processes are to be involved like carrying the
mould box, sand, patterns, products and etc. For the smooth movement of these materials,
a modern material handling system is required for the effective utilization of time and
space and to minimize the production cost. Now-a-days automated material handling
270
 RTMST
systems like conveyor, feeders, cranes, vibrating screens etc. are required for this
purpose.
The use of vibration in materials handling (MH) has been used from many decades.
Vibrating mechanisms, particularly devices used for the sizing of aggregates or the
bolting of cereal grains, have been a part of human scene for centuries. The earliest
devices, which were probably simple wooden boxes with perforated bottoms, were
manually shaken to achieve desired separation results. Evidence exists that such devices
were in common use in earlier civilization. However, it has remained for this century to
bring any large degree of technical enlightenment of this field. Today, of course vibrating
equipment is widely accepted for industrial applications [1].

T. R. Vijayaram [14] presented the importance of selecting the appropriate material

handling equipment in engineering industries. He suggest that by selecting the proper
material handling system, it reduces rework, time and increase production effectiveness
by increasing the profit and application of expert system in the manufacturing
engineering industries. The conveyor may be defined as gravity or powered equipment
commonly used for moving the materials continuously or intermittently, uni-directionally
from one point to another over fixed path, where primary function is conveying of
material [18]. Vibratory conveyors are used to convey a wide variety of materials ranging
from dry powders to heavy block castings. They offer many unique advantages such as
conveying of materials at elevated temperatures, simple construction, combining
operations like cooling, drying, blending and screening with conveying and their
suitability to handle abrasive materials and their applicability as dosing equipment [13,
14, 16]. The most common vibrating equipments, used in industry are vibrating screens,
feeders, shakeouts and conveyors. In addition to transport, vibration can be utilized to
screen, separate, compact or loosen product [14].

M. A. Parameswaran and S. Ganapathy [8] list out some important formulations for
different types of vibratory conveyors. The design of various aspects of vibratory
conveying is a challenging task and it is still an art. They suggested that the jump
conveying is very economical mode of conveying materials, but for powdery materials
like cement, slide conveying is to be better suited. E. M. Sloot and N. P. Kruyt [10] first
represent the design parameters for the vibratory conveyor in their research. They list out
the design parameters for vibrating conveyor as; vibration angle, amplitude, frequency of
oscillation. E. M. Sloot and N. P. Kruyt utilized the combined action of vertical and
horizontal vibration of the support for the controlled transport of bulk cargoes in a whole
variety of industrial process. They also consider the layer of granular material as a point
mass on the conveyor as Rademacher and Brog and present a point mass theory for the
rest, slide and flight phases that occur during transport of granular materials by inclined

271
 RTMST
vibratory conveyors. They also suggested that if τ<1, sliding phase will occur, while if
τ>1 flight phase will occur.

The basic objective of MH activity, which operates in parallel with production, is to

maximize the production with effective utilization of space. There are several design and
conveying principles of vibrating conveyor, depending upon the type of material to be
conveyed. For a non-dusty and unit load like grains, stones stone chips and etc., it will be
an open trough type, while for a dusty and loose material, like cement, it will be a closed
vessel type.
Christof A. Kruelle [15] present a granular flow and pattern formation on a vibratory
conveyor. They investigate the transport behavior of granular material on three principle
oscillation forms- linear, circular and elliptic and suggested that under certain conditions
the direction of flow can be predicted. In an another research, F. Al-Qura’n et al. [19]
present some mathematical formulas and calculation procedure to select a vibratory. They
also suggested the range of some parameter to choose for different purposes. Three
different conveying principles, i.e. sliding, throwing and ratcheting are described by
Christof A. Kruelle et. al. [20].

As a methodology, Value Engineering (VE) is very important to improve the value of the
product by finding and eliminating the unnecessary costs. The concept of Value
Engineering was originally suggested by L.D. Miles [4], who defines Value Engineering
as "The objective of concepts and techniques of value analysis is to make possible a
degree of effectiveness in identifying and removing unnecessary cost that approaches the
effectiveness of identifying performance improvements for the product or service and
incorporating them into it. Value analysis is the name applied to this discipline, step-by-
step thinking system, with its specific approaches for mind setting, problem setting and
problem solving. According to Miles "A problem stated in a solvable manner is half
solved". In this step functions of the product under study are identified and the facts are
separated from what may be called as "honest assumptions".

Arthur E. Mudge [2] said that the purpose of VE systematic approach is to provide each
individual with means of skillfully, deliberately and systematically analyzing and
controlling the total cost of a product. This is accomplished in area of both function and
cost of a product. He also suggested the meaning of value as- the lowest cost to reliably
provide the required functions or service at the desired time and place with the estimated
quality. W. L. Gage [3] provided a frame work, that how to apply the concept of VE to a
system and describes systematic procedure to apply the VE in the different departments
of a company, i.e. manufacturing department, purchase department, maintenance
department etc. Both Mudge and Gage provide the lists that what may be the reasons of
unnecessary costs.

272
 RTMST
Irving Dlugatch [5] suggested that before applying Value Engineering to an product, first
the product should be broken down to subsets such as electrical circuitry, mechanical
device, enclosures and manufacturing process. One can readily establish which of these is
the most expensive and on this basis make the decision to apply Value Analysis to the
expensive item.

In an another research, N. I. Sinisuka [12] said that by applying Value Engineering and
some Reliability models, we can predict the value and life of electrical products in power
transmission more precisely. He says that by using Reliability models in Value
Engineering we can reduce the occurrence of accident in power transmission and can
change the equipments before their failure. SAVE [23] also describe the different phases
of VE job plan and suggests the tool for every job plan. Qian Shi et al [17] proposed that
Value Engineering methodology can be used in all phases to minimize the cost without
reducing the green performance in green building construction with the improved AHP
approach determining the weight of the criteria of building’s green performance. Fenling
et al [21] combines the idea of Value Based Management (VBM) and Value Engineering
in his research of a four-dimensional model and propose a value event path model based
Value Engineering analysis methodology.

1.2 Shakeout Machine and Oscillating Conveyor:

There are many materials handling equipments used in the foundry shops, the two most
important machines are shakeout machine and oscillating conveyor. These machines are
commonly employed for the knocking and conveying of sand. As shown in fig. 1, the
shakeout machine is a simple and robust machine designed for continuous duty to
separate the mould and sand from the mould box, castings from sand and cores from
castings. The top frame of the shakeout is made to shake violently by a vibromotor
system, as shown in fig. 2.

The shaked out sand comes down to the oscillating conveyor and then conveyed as per
requirements. The oscillating conveyor is shown in fig. 3. When oscillating conveyor is
installed under a shakeout machine, it eliminates direct handling of hot sand by the belt
conveyor. Front view of the oscillating conveyor is shown in fig. 4. A manufacturer of
these machines wants to develop a machine, by combining these two functions into a
single machine, for the customers who are using both of the machines. So it is required
to develop a machine which will combine the functions of shakeout machine and
oscillating conveyor. The design of new machine is developed by using value
engineering method.

273
 RTMST

Fig. 1. Shakeout machine consider for the study

(a) (b)

Fig. 2. Shakeout with vibro-motor; (a) front view and (b) side view

274
 RTMST

Fig. 3. Oscillating conveyor consider for the study

Fig. 4. Front view of oscillating conveyor

In this paper the functions of two machines, i.e. shakeout machine and oscillating
conveyor are analyzed individually and an attempt is tried to combine these two functions
of two machines in to a single machine and to find out the best overall combination of
these machines from implementation as well as from Value Engineering point of view,
especially for the industries who are using both machines. And that combination should
also be meaningful for both of the manufacturer and the customers.

2 Methodology:

275
 RTMST
The design of machine is done by value engineering technique where the value of the
machine is increasing by the utility and decreasing the cost of the proposed machine in
comparison to the basic machine i.e. shakeout machine and oscillating conveyor.

There have been many definitions of Value Engineering (VE) put forward by many
persons and bodies since its introduction as discussed in previous section. The value
methodology is applied using a process known as the “Job Plan”. A study of the job plans
indicates that the steps to be followed for Value Engineering study. The Value
engineering Job Plan suggested by Mudge [2] is shown in figure 5.

Fig. 5. VE job plan

3. Oscillating Conveyor and its Performance Parameters:

As other conveying equipments, oscillating or vibratory conveyors are used to convey the
materials. They are similar in application to vibrating feeders; both transport material by
successive series of “throws and catches” and depend on friction for the movement of
materials. Systematic studies on the phenomenon of vibratory conveying are conducted
for last many years [1]. The forward movement of the material takes place mainly during
the “throwing or jump” phase due to the velocity imparted to it by the trough during the
preceding contact phase. At the end of the “jump” phase the material particles strike the
vibrating trough. During the contact phase, that is to say between two successive “jump”
phases, the material particles may slide or not with respect to the trough depending upon
the nature of the impact, the relative velocity after impact, trough acceleration function,
frictional characteristics etc.. Several other factors, not necessarily related to the
parameters of the trough and its drive, make a strict analysis of the material movement
very difficult [8].

276
 RTMST
There are number of parameters involved in predicting the performance of vibratory
conveyor. The performance of a vibratory conveyor depends upon both, the vibratory
conveyor and the material to be conveyed [6, 7, 8, 9, 15]. Two most important parameters
are discussed as follows-

3.1. Throw Number (τ):

The throw number is a dimensionless number and is defined as the ratio of the
perpendicular component of the machine body vibration accelerations to the acceleration
of gravity and is expressed as [7, 9, 15]-

= (1)

( )
or = (2)

where, A= vibration amplitude (mm), = frequency of oscillation (Hz), = inclination

angle of body vibration versus the horizontal line or vibration angle (degree), g=
acceleration of gravity (m/sec.²).

The throw number has very much important from both the practical and theoretical point
of view. It decides the type of conveying, i.e. whether throwing or sliding. The vibrating
conveying will be referred as a sliding conveying if the throw number does not exceed
unity, i.e. τ<1 and the material will always be in contact with the trough surface. On the
other hand, the jump conveying will occur if the throw number exceeds unity, i.e. τ>1
and the material will loose the contact with the trough between two successive jumps.

3.2 Average Conveying velocity (V):

Since the velocity varies within a period, for practical purposes the transport velocity is
introduced, which is defined as the average conveying velocity during a full period and is
expressed as follows [8, 9]-

=

(3)

where, = Correction factor. It is equal to 1 for an ideal material of large porosity which
behaves as a single mass and sticks to the trough during the contact phase. For material
like gravel, cement clinker, seeds etc. varies from .075 to 1.1 and for powdery materials
like cement it is about 0.33, g= acceleration due to gravity, = the factor ( -1) is the
number of trough vibrations over which the material flies ( =1 for 1<τ<3.3), n= ratio of
277
 RTMST
the duration of flight to the periodic time, = impressed frequency of excitation, =
inclination of the trough to the x axis.

The range of V in m/sec., recommended by M. A. Parameswaran et al [8], is 0.05-0.4

m/sec. for rotating unbalance driven conveyor, 0.3-0.7 m/sec. for crank driven conveyors,
and 0.01-0.15 m/sec. for a. c. electromagnet driven conveyors.

Except the above mentioned performance parameters, following are also important
parameters involved for predicting the performance of the vibrating conveyor [9], as
follows-

(a) Parameters related to conveyor: Vibration mode, frequency, amplitude, inclination

or declination of the conveyor, smoothness of the trough surface, geometry of the trough
cross-section, modules of elastic of the trough inner surface etc.

(b) Parameters related to material: Particle shape, particle size distribution, inner
particle friction, friction between particles and trough (wall friction), density of the
material, cohesion, layer thickness, and permeability to air and etc.

4. Proposed Model & its analysis:

Several alternatives were generated after gathering the information about the products
and their subassemblies. But all the ideas were not feasible due to many reasons of VE
point of view and also from customer point of view. The brief outline of the newly
develop idea is shown in figure 6.

278
 RTMST

(1) leg, (2) shakeout hopper, (3) perforated plate, (4) spring with spring support, (5) spring, (6)
vibromotor, (7) basic structure, (8) rectangular plate with slant angle, (9) chute
Fig. 6. Brief Outline of the proposed design

4.1. Design Considerations in the Proposed Model:

The oscillating conveyor, as a separate machine in the earlier design is now eliminated,
and the function of shakeout machine and oscillating conveyor is integrated in to a single
machine. Having many parameters involved in the design of an oscillating conveyor, here
is only two considerations, i.e. the dimensionless throw number and the mean conveying
velocity are considered in this work. The developed design is evaluated from VE point of
view and also compared with other standard values, recommended by the researchers,
which is discussed as follows.

(a) Throw Number (τ):

The outlet side of the shakeout machine hopper is fitted with a rectangular plate, having a
slant angle equal to 24⁰, which is also equal to dynamic angle of repose of sand, i.e.

= = 24⁰
279
 RTMST
= 0.7φ
(4)

The frequency of vibration for a three phase induction motor is 1500 oscillations per
minute. According to IS Code No. IS 8723:1978 [22], at 1500 oscillations per minute the
vibration amplitude = 1.2 mm, = 25 Hz, 2 = 157, and g= 9.81X10³ mm/sec.².

So, from equation (1) the throw number (τ) becomes 1.226. As τ>1, there will be jump
conveying.

(b) Average Conveying Velocity (V):

In equation (3), by taking, correction factor = 1, g= 9.81m/sec.², factor = 1, ratio n= 1

(assume), impressed frequency = 25 Hz, V becomes 0.44 m/sec. As V= 0.44 m/sec.,
which is near the range 0.05-0.4 m/sec., for rotating unbalance driven conveyors, is also
near the standard value.

5. Results and Discussion:

A design of machine is developed by combining the functions of the shakeout machine
and the oscillating conveyor, by applying the methodology of value engineering. The
oscillating conveyor and its driving mechanism are eliminated, as shown in Figure 6, as a
separate machine in earlier design. For the conveying of the sand, a rectangular plate
(conveyor) is fixed at the outlet side of the shakeout hopper, having a slant angle of 24⁰.
It has been found that when the vibration angle is equal to the inclination of the
rectangular plate, i.e. = 24⁰, the throw number becomes 1.226 and which is greater than
unity, which will be called as “Jump Conveying” (as τ>1). In other words, the sand
particles are conveyed when these are thrown in the air at an angle and travel a path of
parabola. At the same time, the trough physically comes backward and the sand particles
come in contact with the trough. Therefore during each cycle, the incremental length of
conveying is equal to the horizontal movement of the sand particles, following a path of
parabola. Moreover the trough physically moves backward the above mentioned distance.
The both of these distances depend upon the, extent and angle of vibration generated by
the vibromotors, and the angle of through, which determine the angle of throwing
(projectile) of the sand particles.

Also, as a case of jump conveying, the hot sand does not remain in contact with the
conveyor, throughout its conveying. Because the sand losses its contact with the
conveyor between two successive jumps and thus the less wear and tear of the conveyor
will take place.

280
 RTMST
The mean conveying velocity (V) was calculated as 0.44 m/sec., which was also near the
standard value, as suggested by M. A. Parameswaran et al [8]. The several advantages of
the proposed design are also list out in the previous section.

Value Engineering, as a methodology has been applied in this work which plays a very
significant role in finding and eliminating the unnecessary cost of oscillating conveyor, as
a separate machine earlier. But the basic function of the conveyor and also the function of
shakeout machine were kept on priority, to fulfill the basic requirements of the
customers.

6. Conclusions:
The functions of two foundry machines, i.e. shakeout machine and oscillating conveyor,
were analyzed separately from value engineering point of view. As the customer wants
only those functions in the products, which support his needs. These are basic functions
or prime functions, and secondary functions, which support the basic functions. For this
purpose the methodology of value engineering has been applied to these machines. A
design of machine is proposed, which combines the functions of two machines. This
proposed design has several advantages over the existing machines. The design
considerations were also discussed and found the value of design parameters were near
about the suggested values by the researchers. Only the average conveying velocity is
very little deviated from the suggested value. It was also found that the “Jump
Conveying” is very economical mode of conveying the hot sand lumps over the vibrating
conveyor.

Acknowledgement:
Authors would like to extend thanks to Dr. B. K. Basak, Executive Director (Foundry
Equipment Group), Wesman Engineering Co. Pvt. Ltd., Kolkata, for his valuable
suggestions and directions for the completion of this work and for his kind permission to
publish this article.

References:
[1] J. H. Booth and H. McCallion, “On Predicting The Mean Conveying Velocity of a Vibratory
Conveyor”, Proc. Instn. Mech. Engrs., Vol. 178, Issue 20, 1963-64, pp. 521-538.

[2] Arthur E. Mudge, “Value Engineering A Systematic Approach”, Mc Graw-Hill America, 2 nd

edition, 1971.

[3] W. L. Gage, “Value Analysis”, Mc Graw-Hill London, 1st edition, 1971.

[4] Lawrence D. Miles, “Techniques of Value Analysis and Engineering”, Mc Graw-Hill

America, 2 nd edition, 1972.

281
 RTMST
[5] Irving Dlugatch, “Methodology For Value Engineering”, IEEE Transaction On Reliability,
Vol. R-22, Issue 1, April 1973.

[6] M. Paz, J. M. Morris, “The Use of Vibration for Material Handling”, Journal of Engineering
for Industry, ASME publication, August 1974, pp. 735-740.

[7] G. Winkler, “Analysing The Vibrating Conveyor”, Int. J. Mech. Sci, Vol. 20, 1978, pp. 561-
570.

[8] M. A. Parameswaran, S. Ganapathy, “Vibratory Conveying-Analysis and Design: A Review”,

Mechanisms and Machine Theory, Vol.4, 1979, pp. 89-97.

[9] Friedrich Johann Carl Rademacher and Lambertus ter Brog, “On the Theoretical and
Experimental Conveying Speed of Granular Bulk Solids on Vibratory Conveyors”, Engineering
Research Bd. 60, 1994, pp. 261-283.

[10] E. M. Sloot, N. P. Kruyt, “Theoretical and Experimental Study of the Transport of Granular
Materials by Inclined Vibratory Conveyors”, Powder Technology 87, 1996, pp. 203-210.

[11] Thoguluva Raghavan Vijayaram, “Materials Handling Technology and Significance of

Expert Systems to Select Appropriate Handling Equipments in Engineering Industries: A
Review”, Journal of Scientific and Industrial Research. Volume 65, August 2006, pp. 619-624.

[12] N. I. Sinisuka, “The Use Of Value Engineering Techniques To Study The Reliability Of
High Voltage Transmission System In Tropical Country”, IEEE Transactions, 2002, ISBN-0-
7803-4/02.

[13] Hamid EL hor and Stefan J Linz, “Model for Transport of Granular Matter on an Annular
Vibratory Conveyor”, Journal of Statistical Mechanics: Theory and Experiment, 2005, pp. 1-8.

[14] Thoguluva Raghavan Vijayaram, “Materials Handling Technology and Significance of

Expert Systems to Select Appropriate Handling Equipments in Engineering Industries: A
Review”, Journal of Scientific and Industrial Research. Volume 65, August 2006, pp. 619-624.

[15] Christof A. Kruelle, Andreas Götzendorfer, Rafal Grochowski, Ingo Rehberg, Mustapha
Rouijaa and Peter Walzel, “Granular Flow and Pattern Formation on a Vibratory Conveyor”,
Traffic and Flow’05, 2007, pp. 111-128.

[16] Erdem Sismek, Siegmar Wirtz, Viktor Scherer, Harald Kruggel-Emden, Rafal Grochowski
and Peter Walzel, “An Experimental and Numerical Study of Transversal Dispersion of Granular
Material on a Vibrating Conveyor”, Particulate Science and Technology, 26, 2008, pp. 177-196.

[17] Qian Shi, Shuoufeng Chen, “Evaluation of Green Construction Alternatives Based on Value
Engineering”, IEEE Transactions, 2008, ISBN-978-1-4244-2108-4/08.

[18] Siddhartha Ray, “Introduction To Materials Handling”, New Age International Publishers
Ltd, India, 1st edition, 2008.

[19] F. Al-Qura’n, J. S. Haddad and M. Dweirj, “Calculation Procedure to Choose the Design
Parameters of Drying Machines”, Journal of Engineering Thermophysics, Volume 18, Issue 4,
2009, pp. 315-320.

282
 RTMST
[20] Christof A. Kruelle, “Physics of Granular Matter: Pattern Formation and Applications”, Rev.
Adv. Mater. Sci., 2009, pp. 113-124.

[21] Fenglin Peng, Xuping Jiang, “A Value-Event Path Model Based Value Engineering Analysis
Methodology”, IEEE Third International Symposium on Intelligent Information Technology
Application, 2009, ISBN 978-7695-3859-4/09.

[22] IS 8723:1978, Dimensions For Vibrating Conveyors And Feeders With Rectangular Or
Trapezoidal Trough, Bureau of Indian Standards.

[23] http://www.value-eng.org, SAVE International website.

283
 RTMST
National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)
Date: 18 th and 19th April, 2013, NITTTR, Kolkata

IMAGE BASED QUALITY CHECKING OF

INDUSTRIAL PRODUCT USING APPLIED
ROBOTICS

Avirup DAS1, Rohit KUMAR2, Biplab ROY3 & Ganesh RAM4

Electrical Engineering Department
National Institute Of Technical Teachers’ Training And Research, Kolkata
(Established by the MHRD, Govt. of India)
Block-FC, Sector-3, Salt Lake City, Kolkata- 700106
Tel: (033)2337-0937/4125/0479 Fax: (033) 23376331
www.nitttrkol.ac.in
1. avirup1987@gmail.com 2.rohit.nitttr@gmail.com 3.roy_biplab_89@yahoo.co.in 4.gnesh.ram@gmail.com

ABSTRACT
The rejection process is done here using image processing technique. System
compares the user defined image with objects. The computer takes snapshots using
camera. The image of the object taken is compared with the standard specimen provided
by the user. The object recognition is done using cross correlation technique of image
processing by Matlab software. Here the object is placed on the conveyor belt. Moving
objects of the conveyor belt is detected by IR sensors and automatic snapshot is then
generated by the system and compared with specimen image. The rejection action is done
using user defined threshold value. The user can set threshold value to measure
acceptable defects of objects. The motor for conveyor belt and rejection action is
controlled using microcontroller. The counter is used here to count total number of object
and rejected object to display rejection ratio in GUI panel. This application can be used in
manufacturing & production units. By using image processing techniques better accuracy
& precision can be obtained during the process of sorting.

284
 RTMST
Keywords: Image processing, cross correlation, GUI, IR Sensor, Embedded system.

1. Introduction:
Image processing techniques are widely being used in many areas of Science and
Engineering for the last few decades and a large volume of literature are now
available with enormous applications leading to a mature field of study in system
theory. The journey of image processing date back to 1920s when cable picture
transmission system was introduced, which reduced the time required to transport
a picture across the Atlantic from more than a week to less than an hour. This was
the first digital coded picture transmitted through cable transmission and
reconstructs them at the receiving end [1]. The application of image processing in
quality control of the manufacturing goods in the product line was attempted in
[2], in which a new method was introduced to achieve precise length measurement of
products on conveyor belt with digital images based on real time tracking of object from
length and time calculation. In [3] the problem of visual robot guidance in tracking
objects moving on conveyor belts was addressed. The instantaneous location of the
moving object was evaluated by a vision system consisting of a stationary, down looking
monochrome video camera. A controller-integrated image processor with a vision
extension feature was developed under structured programming environment. The issue
of quality checking of manufacturing goods in the product line is attempted in
this paper, in which image processing tools were used to compare the images
obtained from the product line to that of the predefined image as components. The
defects in the products are detected using MATLAB processing. The required
detection of difference between the acquired and predefined image is based on
correlation algorithm technique of image processing. Objects moving constantly
over the belt are photographed by camera from different viewing angle. Here, the
first object passing over the belt is taken as the specimen image. Any object can
also be considered as specimen by using option in GUI panel. Each image of the
object is then processed and compared with the predefined images stored. Based
on the average value of correlation coefficient of processed images, defected
objects are sorted. User defined accuracy level determines the required quality of
the product. Sensors and other hardware components are controlled by
microcontroller. The proto type system under development can scan 3 objects in a
minute.

2. Problem statement

2.1 Flowchart
285
 RTMST

START

Object placed on conveyor belt

Object detected by IR sensor

MCU sends signal to MATLAB to

capture images of the object.

Camera activated and takes required

no of snapshot

Object detected by IR sensor and

camera is made off

Image processed by MATLAB & send

decision to microcontroller

NO
Is error more
than predefine
error

YES
Object reaches
IR sensor detects object acceptance tray
and sends signal to
microcontroller

Rejecter rejects the

object

STOP

286
 RTMST

2.2. Hardware development

Mechanical structure of proposed work

Fig 2.1

 ARM

It places objects on the conveyor belt at regular intervals. Having 6 degree of

freedom, objects are placed easily at required position.

 Supporting Frame

It supports the entire structure rigidly.

287
 RTMST
 Pulley

Pulley of radius 3.5cm is used here at both end of the belt. One of the two
pulleys is attached with the motor, which rotates at 10 rpm.

 Conveyor Belt

As shown in figure it’s a long, flat structure which carries the objects for defect
detection and sorting. The length of the belt used is 220cm.The upper surface
length is 99cm. It is mounted over two pulleys which move it with a constant
speed of 3.66 cm/sec.

 Conveyor belt timing diagram

 IR Sensor

Infrared sensor is used here to perform the work of detection of object on

conveyor belt and passes the signal to microcontroller. It synchronises different
288
 RTMST
actuators. This sensor detects the objects as soon as it placed on the conveyor belt. It
informs the microcontroller to make the camera on to take snapshot of the object
placed.This sensor detects the object as it crosses the viewing limit of the camera.The
microntroller reads the output of sensor signal and makes the camera off.This sensor
detects the object on the belt and measure the length of the object to calculate the
required time for rejection action.It is placed just before the rejector. As soon it
detects the object, microcontroller sends signal to rejector for rejection action.

 Rejecter

The image of object taken by the camera is processed and compared with
the standard image. Object having more defectness than accepted is rejected by
the rejecter. The rejecter is 8cm long flat structure, which moves perpendicularly
to the conveyor belt. It strikes the object to make it fall to the rejection tray.The
rejecter hits the object in time.

t = time required by rejecter to hit object after crossing the IR sensor

l = length of the object

 Acceptance tray

After examining and processing of the acquired image, object having acceptable
quality, travels the entire belt and falls in the tray placed at the end position of the
conveyor belt.

 Rejection tray

Objects having the quality below the acceptance error limit are rejected by
the rejecter and object is collected in the rejection tray.

289
 RTMST
 Camera

It plays an important role in the sorting process. Products are sorted on the
basis of their shape with the help of image taken by the camera. It takes snapshot
of the object moving on the belt, and sends it to the computer to process it through
MATLAB. The height of the camera with reference to conveyor belt is
53cm.From this height it can scan the required area of conveyor belt.
 Object

The object is placed on the belt by arm at regular interval. The image of the
object is compared with the standard image using MATLAB.

 Motherboard

It contains all the circuitry components required for the functioning of the
system. Signals from the sensors are fetched here and output signals are given to the
sensor, rejecter.

 Atmega(Microcontroller)

It belongs to the AVR family of microcontroller, having 28 pins. It stores

the required program for sequential working of the system. Sensors are attached
to the ADC pins. I/O functions are carried through 23 I/O pins available.

 Matlab

In this numerical computing environment, task of image processing is

carried out. Using correlation technique, image is processed and compared in
Matlab. Input image is compared with the previously given standard image.

 Computer

The processing of the images using Matlab is done in computer attached

with the system. Computer supporting proper running of Matlab software is
brought in action here. Moreover, it is also used in displaying GUI panel.

2.3. Software development

In the development of this system, programming played an important role there is two
programming section, image processing and microcontroller programming. Image
290
 RTMST
processing program runs in MATLAB platform. ATMEGA8 microcontroller is used here
which belongs to AVR family.

 Image processing

Image processing is used here to compare the specimen object with

other object and give a decision whether it is accurate or not. There is an accuracy limit
which can be defined by user before processing. This image processing based on 2-D.In
this system for comparison, cross-correlation algorithm is used. At first camera take 12
image of specimen object from different angel and store it, when other object runs from
conveyor belt, then also camera take 12 images and compare with specimen images and
result is the average of 12 compared images. If result is higher than user defined accuracy
level then it is accepted otherwise rejected. This is developed in MATLAB platform and
also system has GUI mode for user friendly operation. The main portion of image
processing program is cross-correlation which gives the accuracy level.

 Cross-correlation

It has its maximum value when the two matrices are aligned so that they are
shaped as similarly as possible. If matrix A has dimensions (Ma, Na) and matrix B has
dimensions (Mb, Nb), the equation for the two-dimensional discrete cross-correlation is

Where and .

 GUI mode

GUI mode also designed in MATLAB. GUI mode is used here for simple
operation of the system. User can control full system from a computer using GUI mode.
It has buttons e.g. Start, stopped. User can define acceptance percentage limit in GUI
window. Also it shows the coming object, total no. of object scanned and rejection ratio.

291
 RTMST

 Microcontroller programming

Microcontroller controls the entire hardware portion

mainly conveyor belt motor, object placing arm, rejection arm IR sensors etc. It is
placed in main circuit board or mother board. System is developed using
microcontroller ATMEGA8 of AVR family, which has internal clock frequency
of 8 MHz, A/D converter pins.

 Algorithm

Step1:- Give the power supply to mechanical structure, microcontroller board & other related
components and make sure that every part related to system is working properly.

Step2:-One D.C. Motor is attached with conveyor belt of 10 r.p.m. Receives the specimen object
which is detected by IR Sensor located at position 5.

Step3:- IR Sensor sends the signal to microcontroller; microcontroller gives the commands to
Matlab to processes the signal & takes snapshots of the object through the camera after few
milliseconds.

292
 RTMST
Step4:- Total no of snapshot at particular orientation taken by the camera is compared in Matlab
with the predefined specimens of particular orientation.

Step5:- This will decide whether the object is defected or not and take decisions for rejection on
this basis.

Step6:- Now object passes through IR Sensor located at position 6, this position gives command
via microcontroller to Matlab to turn off camera.

Step7:- IR Sensor located at position 7 calculates the length of object with velocity and time
mechanism.

Step8:- Last sensor located at position 8 will reverse calculate the time which should be equal to
time taken by the motor located at position 9 for shoot out of defected object.

Step9:- If object is not defected then it is carried to box located at 10.

 Experimental result: This is experimental result of image processing program by

comparing object with predefined object image.

--------------------------------Specimen image, matched

percentage=100

--------------------------------Image 1 matched percentage=99.682

------------------------------- Image 2 matched percentage=

99.2945

------------------------------ Image 3 matched percentage= 98.4526

------------------------------ Image 4 matched percentage= 97.4841

293
 RTMST

------------------------------ Image 5 matched percentage= 95.8479

----------------------------- Image 6 matched percentage= 94.6911

------------------------------ Image 7 matched percentage= 93.4599

3. Conclusion
This project can be implemented in industries, requiring sorting of
the product with defected shape. Quality checking via this technique deals with only the
upper surface of the product. Our project can be successfully implemented where there is
no need of internal detailed structure of the object. By employing the technique of cross-
correlation method, image processing finds beneficial application. Objects of varying
shape & colors can be taken as specimen to compare the corresponding objects. User can
define the error acceptance limit of the product.

4. Future scope
 Image processing using Matlab lags real-time operation; this can be
overcome if a dedicated image processing embedded unit is
incorporated.

 By incorporating 3D image processing techniques the overall

performance of the system can be increased.

294
 RTMST
 From this technique we can point out the exact location of error and in
future we can rectify the error of particular object with optimization
technique.

5. References
[1] Digital Image processing using Matlab by Gonzalez.

[2] In 2007, Akihiro Watanbe, Takanori Yamazaki, Hideo Ohnishi of Utsunomiya University,
Japan

[3] Th. Borangiu, "Intelligent Image Processing in Robotics and Manufacturing", Romanian
Academy press, Bucharest, ISBN 973-27-1103-5, 2004

[4] T Yamazaki, et al.: Development of mass and length Measurement system on conveyor belt,
The Journal Of JASME, pp. 513-514, 2006

[5] Zahiri Azar, O. Goksel, and S. E. Salcudean, “Sub-sample displacement estimation from
digitized ultrasound RF signals using multi-dimensional polynomial fitting of the cross-
correlation function,”IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 57, no.11, pp. 2403–
2420, 2010

[6] Xian-Guo Li，Chang-Yun Miao，Ji Wang of Tianjin Polytechnic University presented a

paper “Automatic Defect Detection Method for the Steel Cord Conveyor Belt Based on Its X-ray
Images”.

[7] www.Mathworks.com

[8] www.avrfreaks.net

295
 RTMST

National Conference on Recent Trends in Manufacturing Science and Technology (RTMST-2013)

Date: 18 th and 19th April, 2013, NITTTR, Kolkata

Reliability Analysis of a Paint-Manufacturing Plant

Anujit Chanda1 and Sukanta Kumar Naskar2

1
Assistant Professor, Dr.Sudhir Chandra Sur degree

engineering college-Kolkata, India

Email: chanda.anujit@gmail.com

2
Associate Professor, National Institute of Technical

Teachers Training and Research-Kolkata, India

Email: sukanta_naskar@yahoo.com

Abstract- The Present investigation deals with the analytical evaluation of reliability aspects of Paint
Manufacturing Plant. The collection of the past data is the most vital precondition to carry out such analysis.
The initial stage of the present study has revolved around the collection of well defined data regarding the
breakdowns and maintenance of the Paint Manufacturing Plant. It has been observed that worker
inefficiency, degradation of the components with time is the main factors of the failures of the Paint
Manufacturing Plant component. Reliability of various component of Paint-Manufacturing Plant is estimated
from the breakdown data. The Maintenance policy is necessary to decide the frequency of maintenance. To
determine how frequently maintenance should be done so that the equipment are highly reliable when
needed. Therefore detail and continuous study is required for analyzing the component of a Paint
Manufacturing Plant on reliability aspects.

Key Words: RCM – Reliability Centered Maintenance, PM – Preventive Maintenance, MTBF – Mean time
between failures, FMEA – Failure Modes Effects Analysis, FMECA – Failure Modes Effects and Criticality
Analysis, FTA – Fault Tree Analysis

296
 RTMST
1. INTRODUCTION

The initial stage of reliability analysis is to predict the hazard model of the failures and to choose a distribution
among various probability distributions like Normal distribution, Exponential distribution, poison distribution,
Weibull distribution etc. About all the distributions available for reliability calculations, the Weibull distribution is
the only unique to the field, probably the most widely used distribution in reliability engineering, the Weibull
distribution can model failures caused by fatigue, corrosion, mechanical abrasion, diffusion and other degradation
processes. Reliability analysis of the selected components of the Paint-Manufacturing Plant has been carried out
with due consideration of failure as an indication of the underlying cause of failure has been summarized in term
of weibull shape parameter. Weibull shape parameter beta also indicates whether the failure rate is constant or
increasing or decreasing if beta=1.0, beta>1.0, beta<1.0 respectively. The cumulative % failures versus
Operating time data are plotted on Weibull graph and the values of the parameters are obtained using the
statistical software ‘MINITAB 15’.

Fault Tree Analysis (FTA) was originally developed in 1962 at Bell Laboratories by H.A. Watson, under a U.S. Air
Force Ballistics Systems Division contract to evaluate the Minuteman I Intercontinental Ballistic Missile (ICBM)
Launch Control System.(1) The use of fault trees has since gained widespread support and is often used as a failure
analysis tool by reliability experts.(2) FTA received extensive coverage at a 1965 System Safety Symposium in
Seattle sponsored by Boeing and the University of Washington.(3) Boeing began using FTA for civil aircraft design
around 1966.(4) This change adopted failure probability criteria for aircraft systems and equipment and led to
widespread use of FTA in civil aviation. Within the nuclear power industry, the U.S. Nuclear Regulatory Commission
began using probabilistic risk assessment (PRA) methods including FTA in 1975, and significantly expanded PRA
research following the 1979 incident at Three Mile Island.(5) This eventually led to the 1981 publication of the NRC
Fault Tree Handbook NUREG–0492,(6) and mandatory use of PRA under the NRC's regulatory authority. Barlow
and Proschan in 1975(7) define availability of a repairable system as "the probability that the system is operating
at a specified time t."

2. PROCEDURE

About all the distributions available for reliability calculations, the Weibull distribution is the only unique to the
field. Professor Wallodi Weibull (1887-1979) pointed out that normal distributions are not applicable for
characterizing initial metallurgical strengths during his study on metallurgical failures. He then introduced a
function that could embrace a great variety of distributions and used seven different case studies to demonstrate
how this probably the most widely used distribution in reliability engineering, the Weibull distribution can model
failures caused by fatigue, corrosion, mechanical abrasion, diffusion and other degradation processes. The two
parameter Weibull distribution requires characteristic life (η) and shape factor (β) values. Beta (β) determines the
shape of the distribution. If β is greater than 1, the failure rate is increasing. If β is less than 1, the failure rate is
decreasing. If β is equal to 1, the failure rate is constant. There are several ways to check whether data follows a
297
 RTMST
Weibull distribution, the best choice is to use a Weibull analysis software product. If such a tool is not available,
data can be manually plotted on a Weibull probability plot to determine if it follows a straight line. A straight line
on the probability plot indicates that the data is following a Weibull distribution. There are generally two ways for
graphical evaluation -Exponential plot and Weibull Plot. When the failure rate is constant, the distribution follows
exponential probability law and when failure rate is not constant i.e. non-linear hazard model follows Weibull
distribution.

The observed data regarding the failures of different components of the Paint-Manufacturing Plant shows that the
failure rates of the components are not constant. So the Weibull distribution model can be adopted. Moreover the
technique of linear regression analysis confirms the suitability to use Weibull distribution for the different
components of the Paint-Manufacturing Plant. The analysis determines the best-fit line in

The least square sense. The least square test has been performed to obtain the increasing/decreasing rate of
failures. Failure data of the different components of the Paint-Manufacturing Plant is used for determining the
correlation co-efficient. From the concept of probability, we know that the correlation coefficient must be in
between -1.0 to +1.0. If the correlation coefficient is positive, then the failure rate is increasing, otherwise the rate
is decreasing. The correlation coefficients of the different components of the Paint-Manufacturing Plant under
study show either increasing or decreasing rate of failures. Weibull distribution can be applied for the estimation
of reliability of the different components of the Paint-Manufacturing Plant. The primary advantage of Weibull
analysis is the ability to provide reasonably accurate failure analysis and failure forecast. Another advantage of
Weibull analysis is that it provides a simple and useful graphical plot of the failure data. The data plot is extremely
important to the engineer and to the manager. The Weibull data plot is particularly informative as Weibull pointed
out in his 1951 paper. The horizontal scale is a measure of life or aging. Start/stop cycles, mileage, operating time,
landings or mission cycles are examples of aging parameters. The vertical scale is the cumulative percentage failed.
The two defining parameters of the Weibull line are the slope, beta, and the characteristic life. The slope of the
line, β is particularly significant and may provide a clue to the physics of the failure. The characteristic life, η is the
typical time to failure in Weibull analysis. It is related to the mean time to failure.

The observed data regarding the failures of different components of the Paint-Manufacturing Plant shows that the
failure rates of the components are not constant. So the Weibull distribution model can be adopted. Failure data
of the different components of the Paint-Manufacturing Plant is used for determining the correlation co-efficient.
From the concept of probability, we know that the correlation coefficient must be in between -1.0 to +1.0. If the
correlation coefficient is positive, then the failure rate is increasing, otherwise the rate is decreasing. The
correlation coefficients of the different components of the Paint-Manufacturing Plant under study show either
increasing or decreasing rate of failures. Weibull distribution can be applied for the estimation of reliability of the
different components of the Paint-Manufacturing Plant.

3. EQUATIONS AND SYMBOLS

Linear regression analysis has been carried out to find out the Correlation coefficient of different parts of the
pain manufacturing plants. The Probability equation is-

298
 RTMST

∑[x f(x)] − ∑x [∑f x] ( )

Rx, f(x) =
N ………… 1
2 2
(∑x)
[∑ (x 2) – N
] [∑ f(x2) − ∑f(x)
N ]

Where x = Breakdown hours

Y = f(x) = Cumulative % fail

N = number of trials (number of months)

Rx, f(x) = Correlation coefficient

4. RESULTS AND DISCUSSION

Table 4.1 Weibull Plot for Attritor

SL MONTH MONTHLY NO OF PERCENTAGE CUMULATIVE

NO OPERATING FAILURES FAILURE % FAILURES
TIME(Hr)

1 April’06 327 0 0 0

2 306 0 0 0

3 228 0 0 0

4 200 0 0 0

5 240 1 7.69 7.69

6 513 0 0 7.69

7 573 0 0 7.69

8 535 0 0 7.69

9 420 1 7.69 15.38

10 418 0 0 15.38

11 462 0 0 15.38

12 March’07 330 0 0 15.38

13 April’07 338 2 15.38 30.76

14 366 0 0 30.76

15 240 0 0 30.76

16 234 0 0 30.76

17 215 0 0 30.76

18 505 0 0 30.76

19 585 0 0 30.76

20 505 1 7.69 38.45

21 220 0 0 38.45

22 237 0 0 38.45

23 220 2 15.38 53.83

24 March’08 540 0 0 53.83

25 Apral’08 518 0 0 53.83

299
 RTMST
26 499 0 0 53.83

27 414 0 0 53.83

28 410 2 15.38 69.21

29 440 0 0 69.21

30 370 0 0 69.21

The various data on maintenance actions under taken for the Paint -Manufacturing plant has been collected from
daily breakdown record book and log book supplied by maintenance section. From the daily maintenance record
book, the failure hours, frequency of failures has been estimated to get an overview about the performance of the
Paint-Manufacturing plant during the period under study (From April’06 to March’11).

From the available data the co-rrelation co-efficient is found out. For Attritor it is (.472012). For Ball Mill it is (-
0.086263). Weibull distribution can be applied for the estimation of reliability of the Attritor and Ball Mill.

SL MONTH MONTHLY NO OF PERCENTAGE CUMULATIVE

NO OPERATING FAILURES FAILURE % FAILURES
TIME(Hr)

31 258 0 0 69.21

32 599 0 0 69.21

33 575 0 0 69.21

34 510 0 0 69.21

35 486 0 0 69.21

36 March’09 445 0 0 69.21

37 April’09 298 0 0 69.21

38 340 0 0 69.21

39 360 0 0 69.21

40 478 0 0 69.21

41 424 0 0 69.21

42 448 0 0 69.21

43 360 0 0 69.21

44 350 0 0 69.21

45 340 0 0 69.21

46 225 0 0 69.21

47 310 0 0 69.21

48 March’10 285 0 0 69.21

49 April’10 511 0 0 69.21

50 329 0 0 69.21

51 537 0 0 69.21

52 533 0 0 69.21

53 452 0 0 69.21

54 478 1 7.69 76.9

55 339 0 0 76.9

56 325 0 0 76.9

57 378 3 23.07 99.97=100

58 362 0 0 100

300
 RTMST
59 199 0 0 100

60 March’11 185 0 0 100

Table 4.2 Weibull Plot for BALL MILL

SL MONTH MONTHLY NO OF PERCENTAGE CUMULATIVE

NO OPERATING FAILURES FAILURE % FAILURES
TIME(Hr)
1 April’06 742 0 0 0

2 690 0 0 0

3 664 0 0 0

4 746 0 0 0

5 716 0 0 0

6 726 0 0 0

7 719 0 0 0

8 744 2 25 25

9 744 0 0 25

10 724 0 0 25

11 616 0 0 25

12 March’07 698 0 0 25

13 April’07 723 0 0 25

14 743 0 0 25

15 741 0 0 25

16 590 0 0 25

17 695 0 0 25

18 628 0 0 25

19 744 1 12.5 37.50

20 430 0 0 37.50

21 733 0 0 37.50

22 739 0 0 37.50

23 687 0 0 37.50

24 March’08 745 0 0 37.50

25 April’08 714 0 0 37.50

301
 RTMST
26 739 0 0 37.50

27 694 0 0 37.50

28 637 1 12.5 50

29 742 0 0 50

30 744 1 12.5 62.50

SL MONTH MONTHLY NO OF PERCENTAGE CUMULATIVE

NO OPERATING FAILURES FAILURE % FAILURES
TIME(Hr)
31 128 1 12.5 75

32 705 0 0 75

33 720 0 0 75

34 687 0 0 75

35 744 0 0 75

36 March’09 720 0 0 75

37 April’09 746 0 0 75

38 718 0 0 75

39 739 0 0 75

40 740 0 0 75

41 675 0 0 75

42 631 0 0 75

43 578 0 0 75

44 720 0 0 75

45 571 0 0 75

46 405 0 0 75

47 744 0 0 75

302
 RTMST
48 March’10 722 0 0 75

49 April’10 646 0 0 75

50 739 0 0 75

51 675 0 0 75

52 503 0 0 75

53 457 0 0 75

54 470 2 25 100

55 120 0 0 100

56 700 0 0 100

57 716 0 0 100

58 746 0 0 100

59 683 0 0 100

60 March’11 715 0 0 100

Weibull graph is done in ‘MINITAB 15’.

WEIBULL PLOT FOR ATTRITOR

Weibull - 95% CI
99.9
Shape 3.974
99
Scale 405.0
N 60
90
80 AD 2.287
70 P-Value 0.068
Cmulative%Failure

60
50
40
30
20

10

3
2

90 1 00 15
0
20
0
30
0
40
0
50
0 0
60 70
0

Monthly Operating Time(Hours)

Figure 4.1 Weibull graph for ATTRITOR

Failure Probability of ATTRITOR at mean hour 387.12 hr is= 47.51 %

Reliability of ATTRITOR at mean time 367.00 hr= (100-47.51) %=52.49%

303
 RTMST
WEIBULL PLOT FOR BALL MILL
Weibull - 95% CI
99.9
Shape 6
99
Scale 698.1
N 60
90
80 AD 8.053
Cumulative% Failure

70 P-Value <0.001
60
50
40
30
20

10

3
2

1
0 0 0 0 0 0 0 0 0 0
15 20 30 40 50 60 70 8 0 9 0 100
Mothly Operating Time(Hours)

Figure 4.2 Weibull graph for BALL MILL

Failure Probability of BALL MILL at mean time 664.33 hr is= 24.33 %

Reliability of BALL MILL-1 at 660 hr= (100-24.33) %=75.67%

5. CONCLUSION

The case study to the Paint-Manufacturing Plant is the real application of Reliability Analysis and Maintainability
Aspects of a Paint-Manufacturing Plant. The real data collection is the vital pre-condition to carry out such analysis.

In-depth and continuous study is required for analyzing the effects of the risk assessment and the recommended
preventive maintenance actions for the paint Manufacturing Plant on reliability aspects. The present investigation
can be extended for further study in the following areas:

 Development of basic model for system availability and dependability.

 Reliability of the other machines like TSD, HSD can be evaluated.

REFERENCES

[1] Neal, R.A. (1962) Modes of Failure Analysis Summary for the Nerva B-2 Reactor. Westinghouse Electric
Corporation Astronuclear Laboratory.

[2] Experimenters' Reference Based Upon Skylab Experiment Management. National Aeronautics and Space
Administration George C. Marshall Space Flight Center.

[3] Design Analysis Procedure For Failure Modes, Effects and Criticality Analysis (FMECA). Society for Automotive
Engineers. 1967.
304
 RTMST
[4] Dyer, Morris K.; Dewey G. Little, Earl G. Hoard, Alfred C. Taylor, Rayford Campbell (1972). Applicability of NASA
Contract Quality Management and Failure Mode Effect Analysis Procedures to the USFS Outer Continental Shelf Oil
and Gas Lease Management Program. National Aeronautics and Space Administration George C. Marshall Space
Flight Center.

[5] Sperber, William H. (December 2009-January 2010). "Happy 50th Birthday to HACCP: Retrospective and
Prospective" Food Safety magazine.

[6] Fadlovich, Erik (December 31, 2007). "Performing Failure Mode and Effect Analysis". Embedded Technology.

[7] Richard F.Design Review Based on Failure Mode (DRBFM) approach.

305
Optimal Precision Measuring Instrument Selection Using Analytical
Hierarchy Process

Dr. S. K. DEB@, Mr. M. BHUYAN#

@ Professor, Mechanical Engineering Department, Assam Engineering Colleg, Guwahati-13, Assam,
India (sudipkumardeb@gmail.com)
# M.E 2nd semester, Mechanical Engineering Department, Jorhat Engineering College, Jorhat-07,
Assam, India (manash800@gmail.com)

Abstract

The accurate decision making about the selection of a precision measuring instrument is a very
sensitive issue. It depends totally on the ranking of the instruments from a list of possible alternatives
available. The different opinions put forwarded by the experts (in production system) relating the
instrument ranking process make the problem more complicated. The field of multi-criteria decision
making is now mature, with several well-developed methodologies. In this paper, a relatively new
concept of multiple criteria decision-making procedure popularly called analytical hierarchy process
(AHP) under multiple expert systems is suggested. A computer programming in C++ is developed to
demonstrate the potential application of the procedure with the help of a case study regarding the
ranking of precision measuring instrument (mechanical comparators). For the sake of simplicity in data
collection and its analysis for information recovery, the ranking problem was framed under five
criteria, ten experts and four different mechanical comparators as alternatives. The final ranking value
of alternatives was computed by finding the average priority value of criteria and alternative priority
value under each criterion. The results obtained were validated with the help of popularity index value
based on real life survey in precision measuring in production system plane. The suggested computer
aided technique will definitely help the decision manager as an automated tool in selecting highly
precise mechanical comparators.

Keywords: Multi-criteria, decision making, alternatives, AHP and ranking value.

 1. INTRODUCTION

In modern production processes, precision measuring instrument plays a great role as it concern
with interchangeable products, precise dimensional control, etc. in industry. Precision measuring
instruments achieved these demands arise in production industry. They are sensible, accurate and free
from inertia defects which made them capable enough to maintain the precision and hence, it makes
the production system better. Comparators are the precision measuring instruments calibrated by
means of end standards to measure unknown dimensions. The main purpose of such instruments is to
detect and display the small differences between the unknown linear dimensions and the length of the
standard. Thus, precision measuring instrument are very much essential from production point of view.

The accuracy and precision of any precision measuring instruments depends on the order of
magnification and rate of accuracy of selecting right instrument as per ranking of alternatives available.
Thus, the concept of multiple criteria decision making under multiple expert system has been a
prominent research area for a long period. The field of Multi-criteria decision-making is now mature,
with several well-developed methodologies. If necessary supporting facilities are properly selected and
used, they can enhance usability of the industrial product, thereby, boosting up to achieve better trade
off and competitive posture. Selection of right precision measuring instruments give better accuracy
and the selection should be carefully made considering all influencing subjective as well as objective
parameters.

Lot of research work has already been carried out in this direction for the last few decades. The
multiple criteria selection usually depends on (i) technical factors that deal with the appropriateness of
the technology; (ii) structural factors that deal with the coordination, information and control and (iii)
strategic factors that deal with the management of resources and uncertainty [2]. Several research
papers are available on the ranking of alternatives based on the cost accounting technique like net
present value, internal rate of return etc. These approaches have considered only the quantitative
factors. Many very important subjective factors such as competitive advantage and social issues have
not been included in the approach. Owing to the unstructured nature of the problem, many researchers
have proposed various approaches, which have not been very successful to deal with the qualitative
factors associated with the problem. Dweri [3] has suggested fuzzy decision support system while
finding the attribute weight by AHP for finding the facility selection routine. Deb and Bhatacharyya
[4], [5] have proposed methodology considering the rating attitudes of decision-makers and trade-off
among various selection criteria in the aggregation process to assure an accurate selection policy.
Multiple criteria decision-making methods were used with the support of fuzzy system to find the
selection order of machines for placement in open continual plane considering zero base area
allocation [6]. In this paper, the authors have applied a relatively new and simplified version of the
analytical hierarchy process suggested by T.L. Satty [9] on their previous work [7] relating multi-
criteria decision making technique suggested by Deab B.V and Nishry, J.J. [8]. AHP is a decision
making tool for dealing with complex, ill structured and multiple attributes decision problem. It helps
in evaluating multiple attribute alternatives when subjective assessments of qualitative factors are
integrated with quantitative factors. AHP has been used in a wide variety of decision areas, including
manufacturing and production systems. A computer programming in Borland C++ was developed to
demonstrate the potential application of the procedure with the help of a case study regarding the
ranking of mechanical comparators precision measuring instruments. For the sake of simplicity in data
collection and its analysis, the selection problem was framed under five criteria, ten experts and four
different mechanical comparators as alternatives.

2. Multi-criteria Decision Making

Multi-criteria decision making (MCDM) is a technique for decision support which aims at providing
transparent and coherent support for complex decision situations taking into account subjective
preferences of the decision makers. The field of MCDM is a full-grown branch of operations research,
concerned with the mathematical and computational tools to support the subjective evaluation of a
finite number of decision alternatives under a finite number of performance criteria. It aims at
supporting decision makers who are faced with making numerous and conflicting evaluations. MCDM
aims at highlighting these conflicts and deriving a way to come to a compromise in a transparent
process. Unlike methods that assume the availability of measurements, measurements in MCDM are
derived or interpreted subjectively as indicators of the strength of various preferences. Preferences
differ from decision maker to decision maker and the outcome depends on the goals and preferences of
experts.

A decision is a choice out of a number of alternatives, and the choice is made in such a way that the
preferred alternative is the "best" among the possible candidates. Usually, there are several criteria to
judge the alternatives and there is no alternative, which outranks all the others under each of the
performance criteria. Hence, the decision maker does not only judge the performance of the
alternatives under each criterion. He/she also weighs the relative importance of the criteria in order to
arrive at a global judgment. Moreover, in a group of decision makers each member faces the question
of how to judge the quality of the other members and their relative power positions before an
acceptable compromise solution emerges.

3. ANALYTICAL HIERARCHY PROCESS

The AHP is a decision making tool for dealing with complex, ill structured and multiple attributes
decision problem. It helps in evaluating multiple attribute alternatives when subjective assessments of
qualitative factors are integrated with quantitative factors. T.L. Saaty developed it during 1970s. Since
its initial development, AHP has been used in a wide variety of decision areas, including
manufacturing and production systems. The traditional decision making approaches consider only the
quantitative factors, failing to recognize the many importance qualitative factors such as environmental
link, supervision link etc in a manufacturing system. Moreover traditional layout decision overlooks
the problem of inconsistencies of designers. The AHP uses a nine-point scale defined to get intensity
importance factor (aij) as: 1- equal importance, 3-moderate importance, 5-strong importance, 7-very
strong importance, and 9-extreme importance. The even numbers 2, 4, 6 and 8 are for compromise, and
the reciprocals show the inverse pair wise comparisons. These numbers represent the weight factors
(priorities) of the reasons involved in the decision making process. The intensity importance of factor i
over factor j is equal to reciprocal of intensity importance of factor j over factor i.

Thus, it can be written as

aij x aji= 1 (1)

3.1. Procedure for the Use of AHP

Since AHP involves a comparison of two factors, the matrix that contains the weight assignments
must be a square matrix. Let A represents the matrix and its size as nn. AHP uses a process known as
systemization. The procedure is as follows:
Step 1. Form the importance intensity matrix A=[aij] by using Saaty’s 9 point scale.

Step 2. Find geometric mean of aij for all the participants to get groups numerical assignment when
factor i is compared with factor j with popular 9 point scale of Saaty.
Step 3. Obtain column total (Sj) by adding the weight assignments in each of the column j

(2)

Step 4. Divide each element of matrix A by its column total (Sj) to get normalized pair wise
comparison matrix:

(3)
Step 5. Estimate relative priorities of each factor by computing the average of the normalized weights
in each of i row. Let pi represents the relative priority of factor i. Then,
 (4)

(5)

3.2. Consistency ratio estimation

Following steps are performed to compute the consistency ratio estimation.

Step1. Multiply each of the columns of comparison matrix A by the relative priorities corresponding to
that column and add to obtain an n1 matrix called B. Thus the new matrix can be expressed as:

 b1   p1a11  p2 a12  - - -  pn a1n 

   
 b2   p1a21  p2 a22  - - -  pn a2 n 
B  -    - - - - - - - - - - - - - - - - - - - - - - - - - -
-   - - - - - - - - - - - - - - - - - - - - - - - - - - (6)
   
 bn   p1an1  p2 an 2  - - -  pn ann 
   

Step2. Compute consistency index (CI) [10] as follows

max  n 1 n bi
CI  where max   (7)
n 1 n 1 pi
i

Step3. Obtain consistency ratio as CR=CI/RI where RI is a random index which represents the
consistency index of a randomly generated comparison matrix taken from standard random index table
which is given by Saaty for the number of factors used in the decision-making. Any value of CR  0.1
is considered as an acceptable ratio of consistency.
4. FORMULATION AND MODEL GENERATION

In this paper, a multiple criteria decision making problem based on simplified version of AHP is
proposed. The subjective (qualitative) as well as objective (quantitative) criteria associated with the
problem are assigned by the experts within approximated information value available to determine the
ranking value of the alternatives. The concept of hierarchical structure analysis with three distinct
levels is used in this paper. The first level (level zero) is to set objectives or goal, the second level (i.e.
level 1) is set to evaluate the priority vector of the decision criteria. In the level 2, the priority vector of
alternatives under each criterion is evaluated. The hierarchical structure of the different criteria and
alternatives under analytical hierarchical process is shown in figure 1. A group of experts is assigned to
employ rating sets within Satty’s nine point scale to evaluate preference information by pair wise
comparison methodology. Similarly, the experts assess the suitability of each alternative under each
criterion and assign a preference value within [0-9] scale depending upon the importance of the
alternatives under that criterion.

Let, m = total number of criteria, n = total number of alternatives, E = total number of experts and

APCi = Average priority value of criteria ‘i’ under E number of experts,

APAij = Average priority value of alternative ‘i’ for criteria ‘j’ under E number of experts,

RVAi= Ranking value of alternative ‘i’, then

RVAi= APCj X APAij

where i=1,2,3,………m and j=1,2,3,…….n.

Goals

Expert 1 Expert 2

Criteria 1 Criteria 2 Criteria 3 …..Criteria n

Instrument1 Instrument 2 Instrument n

Ranking values of alternative

instruments

Fig. 1.Hierarchical structure with three distinct levels

5. COMPUTER SIMULATION AND CASE STUDY

In line with analytical discussion made in the above section, a computer programming in
Borland C++ was developed to handle a case study relating ranking of precision measuring instrument
(mechanical comparators) by applying proposed simplified AHP based multiple criteria decision
making methodology. The steps of algorithm are mentioned below:

5.1. Steps of Algorithms

(i) Develop paired comparison matrix for level 1 with respect to goal for determining priority
value of criteria for each expert, and then find the average priority value of criteria (APCi).
Also get adjusted priority value of criteria (AAPC) if any APC value is less than 15%. If all
APC values are more than 15% then APC=AAPC.
(ii) Develop paired comparison matrix for level 2 with respect to alternatives for each criterion
under each expert.
(iii)Get average priority value of alternatives for each criterion under all experts i.e. APAij
(iv) Get ranking value of alternatives i.e. RAVi and select Maximum RAVi as the best alternative.

5.2. Design of Case Study

In order to evaluate the applicability of the proposed methodology, a case study relating the
ranking or popularity of mechanical comparators were tested under 5 criteria, 4 alternatives and 10
groups of experts chosen randomly from the production plane community. A preliminary survey was
conducted in order to reflect on the criteria commonly adopted in the production system in using of
mechanical comparators. The criteria chosen are (i) range of usability (ROU) , (ii) user friendly (UF),
(iii) rate of accuracy (ROA), (iv) order of magnification (OOM), and (v) Service and maintenance cost
(SMC). The aim of the experimentation was to ascertain which mechanical comparators out of 4
alternatives namely Dial Indicator (DI), Johansson Mikrokator (JM), Reed Comparator (RC) and
Sigma Comparator (SC) are having the maximum ranking value.

6. RESULTS AND DISCUSSION

Survey was conducted to get the preference values of experts primarily limited to production
system plane. Ten most reliable sets of data after vigorous screening were used in the computer
programming to get the values of APC, APA and RVA of four alternatives. The final paired wise
comparison matrix of level 1 with the average priority values of criteria (APC) is shown in table 1.
Neglecting the priority values of criteria less than 15% in table 1, the adjusted average priority values
of criteria are shown in table 2. Similarly, the paired wise comparison matrix of level 2 was developed
and the average priority values of alternatives under respective criteria including the value of APA are
shown in table [3 – 5]. The calculated value of RAV is shown in table 6. It is observed from the results
that the maximum value of RVA is 47.9 and it is scored by sigma mechanical comparator (SC). In
order to validate the computer simulation results with the real condition, a popularity index value (PIV)
was determined by surveying in production system in production plane. It is found that the PIV of
sigma comparators is highest, thus validating the authenticity.

Table 1. Paired comparison matrix of level 1 with APC

Criteria SMC ROU UF OOM ROA APC

SMC 1 9 3 5 7 04.2

ROU 0.11 1 3 7 5 10.8

UF 0.33 0.3 1 5 3 15.8

OOM 0.2 0.14 0.2 1 3 29.5

ROA 0.14 0.2 0.3 0.3 1 39.7

Sum 1.78 10.64 7.5 18.3 19 100

Table 2. Adjusted average priority values of criteria (AAPC)

Criteria SMC ROU UF OOM ROA

AAPC 0 0 0.18 0.35 0.47

Table 3. Paired comparison matrix of level 2 with APA Under criteria UF

Alternative DI JM SC RC APA

DI 1 7 3 5 6.44

JM 0.14 1 5 3 19.04

SC 0.20 0.33 0.33 1 45.36

RC 0.33 0.20 1 3 27.35

 Table 4. Paired comparison matrix of level 2 with APA Under criteria OOM

Alternative DI JM SC RC APA

DI 1 3 5 7 5.78

JM 0.33 1 3 5 12.64

SC 0.33 0.20 0.20 1 50.96

RC 0.20 0.14 1 5 30.66

Table 5. Paired comparison matrix of level 2 with APA Under criteria ROA

Alternative DI JM SC RC APA

DI 1 3 5 7 5.63

JM 0.33 1 5 3 13.30

SC 0.20 0.33 0.33 1 46.41

RC 0.20 0.14 1 3 34.66

Table 6. Ranking Values of alternatives (RAV)

.Alternative DI JM SC RC

RAV 5.84 14.13 47.9 31.9

PIV 2 3 3 2

7. CONCLUSION

The paper has proposed and explained the potential applicability of a simplified MCDM based on
analytical hierarchy process under multiple experts. The computer programming developed by the
authors can easily handle the preference values assigned by different experts. The methodology
presented is easy to understand and able to handle any unstructured problem which needs systematic
analysis through qualitative parameters. The suggested methodology and programme developed will be
a readymade tool for selecting the optimal precision measuring instrument i.e. mechanical comparator
within different mechanical comparators. The proposed methodology can be applied on a wide
interdisciplinary area of engineering and management by systematic handling data to make automated
decision.

8. REFERENCES

[1] Madu, C. N. and Georgantzas, N. C. (1991), "Strategic thrust of manufacturing automation decisions: a
conceptual framework", IIE Transactions 23(2). pp 138-48.
[2] Noori, H. (1990), "Managing the dynamics of new technology", Englewood cliffs, NJ, Prentice Hall.
[3] Dweri, F. (1999), "Fuzzy development of crisp activity relationship charts for facilities layout", Int. J of
Computer and Industrial Engg, 36, pp.1-16.
[4] Deb, S.K. and Bhattacharyya, B. (2002), “Material handling equipment selection by fuzzy multiple criteria
decision making methods”, Artificial Intelligence series, Springer-Verlag, Berlin, No. 2275, pp. 99-105.
[5] Deb, S.K. and Bhattacharyya, B. (2004), “Facility layout and material handling system selection planning
using hybrid methodology”, International Journal of Industrial Engineering, 2004, Vol No. 11(4), 377-
386).
[6] Deb, S.K. and Bhattacharyya, B. (2003), “Facility layout planning based on fuzzy multiple criteria decision
making methodology”. International Journal of Production Research, 41(18), 4487-4504.
[7] S.K.Deb, D.Sarma, M.Bhuyan, R.Saikia and B.J.Kalita, 2010, “Multi-criteria decision making for family
car selection under multiple expert system”, Proc. In National Conference on Recent Advances in
Technology and Management (NCRATM-2010), Jadavpur University, Kolkata.

[8] Deab B.V and Nishry, J.J. (1965), “Scoring and profitability models for evaluating and selecting
engineering products”, Journal of operation Research, Society of America, vol 13, July-Aug, pp 550-569.
[9] Saaty T.L. (2001), Decision Making for Leaders – The Analytic Hierarchy Process for Decisions in a
Complex World, 3rd Edition, RWS Publications, USA, 2001.

[10] S.K.Deb, D.Sarma, M.Bhuyan, R.Saikia and B.J.Kalita, 2010, “Computer aided multi-criteria decision
making using Analytical Hierarchy Process”, Proc. In National Conference on Emerging Trends and
Applications in Computer Science (NCETACS-2010), St. Anthony’s College, Shillong.

***************************************