PREDICTION OF DEPTH OF CUT OF ABRASIVE

WATERJET MACHINING OF TI-6AL-4V

The Project Report submitted in partial fulfillment

of the requirements of the degree of

Bachelor of Mechanical Engineering

By

Mr. Dnyaneshwar Sanap

Mr. Shubham Phalke
Mr. Prakash Gharti
Mr. Dinesh Choudhary

Under the guidance of

Prof. Paramjit Thakur

Department of Mechanical Engineering

Saraswati College of Engineering, Kharghar
University of Mumbai
(2017-2018)
I
 Saraswati Education Society’s

SARASWATI COLLEGE OF ENGINEERING, KHARGHAR

Vision

“To become a leading research organization in the world and make youth India globally
competent to take up the challenges of ever expanding fields of technology.”

Mission

“To create dedicated knowledgeable human resource, technically competent to serve the nation
and guide with its leadership qualities to uplift the socio-economic condition of deprived sections
of our society still miles away from progress.”

Department of Mechanical Engineering

Vision
“To emerge as a centre of excellence in academics and research in the field of Mechanical
Engineering.”

Mission
“To create professionally competent engineers with analytical and research skills for promoting
an environment of continuous learning.”

II
 Program Educational Objectives (PEO’s)

1. To create excellent mechanical engineers with core competency in mathematics, science and
engineering enabling development of problem solving skills.

2. To strengthen themselves professionally and personally to accept responsibilities and pursue

higher education in engineering and other professional fields.

3. To use modern tools and techniques necessary for mechanical engineering and
Allied disciplines leading to research and development.

4. To inculcate the students to work effectively in cross-functional teams and cultivate the
principles of professional ethics and social responsibilities.

5. To develop the students adaptable to change in environment and application of appropriate

methodology for value addition so as to support economic growth.

III
 Program Outcomes (PO’s)

At the end of the program a student will be able to:

1. Apply knowledge of Mathematics, Science and fundamental of Mechanical Engineering.

2. Design and conduct experiments, as well as to analyze and interpret the data.

3. Use of techniques, skills and modern engineering tools necessary for engineering practice.

4. Identify, formulate and solve engineering problems.

5. Commitment to professional, social and ethical responsibility to function effectively as a team

leader in engineering practices.

6. Inculcate self-discipline, self-confidence and to make the students internally driven to set
personal goals.

7. Engage in life –long learning in order to maintain sustainable development even during the
crisis.

8. Solve the contemporary issues with broad understanding of the impact of engineering and
technology on society.

9. Understand environmental and contemporary issues.

10. Possess the skills to communicate effectively with a wide range of both engineering andnon-
engineering community.

11. Apply knowledge of management principles to handle projects in multidisciplinary

environment and work as a team and as an individual effectively.

IV
 Acknowledgement

We wish to express our gratitude to Dr. Manjusha Deshmukh(Principal), Saraswati

College of engineering for providing an opportunity and necessary facilities to carry out this
project.

We also wish to convey our sincere thanks to Dr. Fauzia Siddiqui (Head of the
Department), Mechanical Engineering for his encouragement and support towards this project.
We owe our deep sense of gratitude to our Project Coordinator, having extended his fullest
support in completing the project work.
We would like to thank our guide Mr. Paramjit Thakur, Professor, Department of
Mechanical Engineering for providing the technical support and guidance during various stages
of project experimentation.

We also like to express our thanks to all teaching and non-teaching staff members in the
Department of Mechanical Engineering.

The acknowledgement would be incomplete if we do not mention the emotional support

and blessings provided by our friends and family and also our thanks to the entire Mechanical
department faculty and supporting staff for their good support for completing my project in
successful manner. We had a pleasant enjoyable and fruitful company with them. Last but not
the least we would like to thank all the people who directly or indirectly helped us.

V
 CERTIFICATE

This is to certify that the project entitled “Prediction of Depth of Cut of Abrasive
Waterjet Machining of Ti-6Al-4V” is a bonafide work of

Dnyaneshwar Sanap
Shubham Phalke
Prakash Gharti
Dinesh Choudhary

Submitted to the University of Mumbai in partial fulfilment of the requirement for the award

of the degree of “Undergraduate” in “Mechanical Engineering”.

_____________________ _____________________

Prof. Paramjit Thakur Prof. Mitali Mhatre

Project Guide Project Co-ordinator

_________________ ______________________
Dr. Fauzia Siddiqui Dr. Manjusha Deshmukh
Head of Department Principal

VI
 Dissertation Approval for B.E.
This dissertation entitled “Prediction of depth of cut of abrasive waterjet machining of Ti-6Al-4V” by
Mr. Dnyaneshwar Sanap, Mr. Shubham Phalke, Mr. Prakash Gharti, Mr. Dinesh Choudhary
is approved for the degree of B.E. in Mechanical engineering of Mumbai University.

Examiners

1.____________________________

2.____________________________

Guide
Prof. Paramjit Thakur

_____________________________

Principal
Dr. Manjusha P. Deshmukh

_____________________________

Date:
Place:

VII
 Declaration

We declare that this written submission represents our ideas and where others’ ideas or
words have been included. We have adequately cited and referenced the original sources. We
also declare that we have adhered to all principles of academic honesty and integrity and have
not misrepresented or fabricated or falsified any idea /data/fact/source in my submission. We
understand that any violation of above will be cause for disciplinary action by the Institute and
can also evoke penal action from the sources which have thus not been properly cited or from
whom proper permission has not been taken when needed.

__________________ ___________________
Mr. Dnyaneshwar Sanap Mr. Shubham Phalke

__________________ ___________________
Mr. Prakash Gharti Mr. Dinesh Choudhary

Date:

Place:

VIII
 Abstract

Water jet machining is one of the adaptive unconventional type of machining with
significant advantages over the conventional type machining. The measurement of depth of cut is
a key aspect in development of fully automated abrasive waterjet (AWJ) machining system.
While cutting Ti-6Al-4V under waterjet, it is critical to measure the depth of cut of a material at
the instant of cutting. Although having more benefits of water jet machine this may be one of the
drawback which may increase machining time. The paper reports on the Prediction of Depth of
cut by using different Artificial Intelligence techniques and first comparison of hybrid A.I
techniques with conventional A.I techniques. After cutting of Ti-6Al-4V alloy with AWJ
machine measurement of a response (D.O.C) is done by more sophisticated technique of Image
Processing in MATLAB. Effect of different process parameters such as abrasive flow rate
(A.F.R), stand-off-distance (S.O.D) and traverse rate (T.R) on depth of cut (D.O.C) found
considerably. Effect of this process parameters on the depth of cut considered for Prediction of
D.O.C by using various Artificial Intelligence (A.I) techniques like ANN (Artificial Neural
Network), ANFIS (Artificial Neuro-fuzzy Inference System), Fuzzy Logic, GA (Genetic
Algorithm) etc. The results were compared by comparing AI technique in prediction of response
to get most accurate AI technique.

Keywords: Abrasive jet machining; Machining of Ti alloy; ANN; ANFIS; Parameters

optimization.

IX
 CONTENTS

Acknowledgement
Certificates
Abstract
Contents
List of Tables
List of Figures
List of Abbreviation
Chapter 1 Introduction

1.1 Background 2-4

1.2 Motivation 4
1.3 Need of Project 4
1.4 Limitations of Project 5

Chapter 2 Literature Survey

2.1 Literature Survey 6-12

2.2 Summary of Literature Survey 13-17
2.3 Gap Analysis 18

Chapter 3 Problem Definition and Objectives

19
3.1 Problem Definition
19
3.2 Objectives
20-21
3.3 Methodology

Chapter 4 Experimentation and Measurement of Response

4.1 Test Run 22
4.2 Design of Experimentation 22-23
4.2.1 Measurement of Response 23
4.3 Experimental Setup 23-24

X
 4.4 Experimentation 25
4.5 Image Processing 26-28

Chapter 5 Effect of Parameters on Response by using AI

Techniques
5.1 Artificial Neural Network (ANN) 29
5.1.1 Structure of Neural Network 29-30
5.1.2 ANN Results 30-31
5.1.3 ANN Plots 32-33
5.2 ANFIS 34
5.2.1 The ANFIS Architecture 34-35
5.2.2 ANFIS Plots 36
5.2.3 Fuzzy Inference System 37-38
5.3 ANN vs ANFIS Results 39

Chapter 6 Results and Conclusion

6.1 Results 40
6.2 Conclusion 41
6.3 Future Scope 41

References 42-45

XI
LIST OF TABLES

Table no. Description Page No.

2.1.1 Summary of Abrasive water jet cutting of Titanium alloy 13

2.1.2 Summary of Online Process Monitoring 14
2.1.3 Summary of application of Artificial Intelligence in process monitoring 16
4.2 Input Parameters and their Levels 23
4.4 Measurement of Depth of Cut through Image Processing 27
5.1 Results of ANN Model 30
5.1.1 Prediction of DOC by ANN 31
5.3 Prediction of DOC by ANFIS 38
6.1 Comparison of Experimental and Predicted Values 41

XII
LIST OF FIGURES

Figure no. Description Page no.

1.1 Process Parameters Influencing the AWJ Cutting Process 1

1.2 Classification of AI Techniques 3

1.3 Types of Neural Networks 3

3.3 Methodology Chart 21

4.1 Test Run on Ti-Alloy 22

4.3 (a) and (b) Experimental Setup 24

4.4 Material Ti-6Al-4V with 64 Cuts 25

4.5 Two Halves of Material with help of Wire EDM Machine 25

4.6 Image Processing 26

5.1 3-2-1 Neural network 30

5.1.3 ANN Regression Plot 32

5.1.4 ANN Performance Plot 33

5.2.1 The Architecture of ANFIS 35

5.2.2 ANFIS Result (I/P vs O/P) 36

5.2.3 Fuzzy Inference System (FIS) 37

5.3 ANN vs ANFIS Result 39

XIII
LIST OF ABBREVIATION
P Water pressure (MPa)

V traverse speed (mm/s)

H stand-off distance (mm)

F abrasive flow rate

G garnet embedment

α impingement angel (degree)

t material thickness

T Time domain

F Frequency domain

W Wavelet domain

Ra Surface Roughness

Fc Cutting Force

Tc Cutting Temperature

L-M Lavenberg-Marquardt

SCG Scaled Conjugate Gradient

NSGA Non-dominated Sorting Genetic Algorithm

FMPI Fuzzy Multi response Performance Index

gbellmf generalized bell membership function

MSE Mean Square Error

MP Mean Prediction Error

RMSE Root Mean Square Error

PPE Percentage Prediction Error

MLP Multilayer Perceptron

XIV
 CHAPTER 1
INTRODUCTION
Abrasive water jet (AWJ) is an unconventional method used in machining of difficult-to-
machine materials like titanium, tool steels, super-alloys, hardened steels, glass, composites,
metal matrix composites, laminates and advanced ceramics by high pressure, high velocity water
jet. This list of materials expands daily as people apply the unique properties of fluid jets to
industrial problems.

Fig. 1.1: Process Parameters Influencing the AWJ Cutting Process

This machining process is now a day's replacing the conventional machining processes due to
their significant advantages in present industrial scenario. AWJ machining is a process in which
a very high velocity stream of water is used to accelerate particles of an abrasive material, which
in turn cuts the workpiece material by erosion principle. It is critical to measure the depth of cut
of a material while working on the water-jet machine.

Although having more benefits of unconventional water jet machine this may be one of the
drawback which may increase machining time. In this project we are going to predict the depth
of cut of Titanium alloy while working under water jet machine using signal processing and then
Artificial Intelligence and its hybrid techniques.

1
1.1 Background:

The various modern machining processes getting widely used in the industries are:
electric discharge machining (EDM), abrasive jet machining (AJM), ultrasonic machining
(USM), electrochemical machining (ECM) and laser beam machining (LBM) including various
modified versions of these processes. These processes work on a particular principle by making
use of certain properties of materials which makes them most suitable for some applications and
at the same time put some limitations on their use. These processes involve large number of
respective process variables (also called as process parameters) and selection of exact parameters
setting is very crucial for these highly advanced machining processes which may affect the
performance of any process considerably. Due to involvement of large number of process
parameters, random selection of these process parameters within the range will not serve the
purpose. The situation becomes more severe in case if more number of objectives are involved in
the process. Such situations can be tackled conveniently by making use of optimization
techniques for the parameters optimization of these processes.
From last few decades, many researchers are working on the process monitoring and
optimizing the process parameters in AWJ machining with different materials. Many different
Artificial Intelligence(AI) techniques like Artificial Neural Networks (ANN), Fuzzy logic and
Adaptive Neuro-fuzzy Inference System (ANFIS) are being used in predicting and optimizing
the process parameters. AWJM processes these AI techniques giving the details of formulation
of optimization models, solution methodology used and optimization results. Use of these
techniques are enhancing the AI revolution and automation. The artificial intelligence techniques
are divided into two parts, one is symbolic knowledge representation and another is
computational knowledge representation as shown in fig. 1.2. Symbolic representation is again
divided into exact reasoning i.e. expert system and approximate reasoning i.e. fuzzy logic.
Computational knowledge representation is nothing but ANN. Neural networks is classified into
feedforward networks and recurrent/feedback networks mentioned in fig. 1.3. Single-layer
perceptron, multilayer perceptron, radial basis functional nets comes under the category of
feedforward networks. Similarly, feedback networks consist of competitive networks, kohonen’s
SOM, Hopefield network, ART models.

2
 Fig. 1.2: Classification of AI Techniques
However, these approaches require a large amount of data for training them so as to
achieve a reasonable accuracy in the prediction of process parameters. By using time domain,
frequency domain and wavelet domain, process parameters can be predicted in online process
monitoring. For online process monitoring acoustic emission, cutting force dynamics, sound
detection, infrared thermography and vibration sensing techniques can be used. These techniques
are used for investigating different phenomena associated with manufacturing process such as
tool wear, tool breakage, non-destructive evaluation and surface roughness.

Fig. 1.3: Types of Neural networks

3
 Titanium alloys are among the hardest materials which can sustain maximum temperature
at cutting zone. Ti-6Al-4V (UNS designation R56400) is an alpha-beta titanium alloy featuring
high strength, low weight ratio and excellent corrosion resistance. It is one of the most commonly
used titanium alloys and is applied in a wide range of applications where low density and
excellent corrosion resistance is necessary such as e.g. aerospace industry and biomechanical
applications (implants and prostheses). It is known as the “workhorse” of the titanium alloys, Ti
6Al-4V or Grade 5 titanium, is the most commonly used of all titanium alloys. It accounts for 50
percent of total titanium usage the world over. As it is hard to machine this material with
conventional we can machine it with AWJ machining.

1.2 Motivation:
In every field of engineering there is a continuous demand and need of improvement. So,
we being engineers, tend to achieve and fulfill the requirement of various fields. Ti grade 5 has
got vast range of application in aerospace industries, medical, marine and chemical processing
industries. Cutting of this material by conventional CNC milling is too costly due to frequency
tool replacement. Therefore, non-conventional process like AWJM is used to cut this material.
The major issue in AWJ cutting of this material is improper depth of cut.
Due to some special characteristics of Ti-alloy such as high strength, low weight ratio and
excellent corrosion resistance, result in them difficult to machine. If the depth of cut is
incomplete, the number of pass need to be increased. Hence this works deals with online process
monitoring of depth of cut by application of AI techniques to get better performance in minimum
cost.

1.3 Need of Project:

To produce a cavity with controlled depth is a basic problem in AWJ milling operation.
To predict the depth of cut or penetration is the major task in AWJ machining. Literature survey
shows work of various researchers on predicting and optimizing the various process parameters
on different materials. Many AI tools like ANN, Fuzzy logic and ANFIS can be used in
predicting and optimizing the process parameters, as per literature survey there is no work found
on use of ANFIS for predicting the depth of cut while working on AWJ machining. Predicting

4
the depth of cut of Ti-6Al-4V alloy while cutting in AWJ machining using ANFIS is the basic
aim of the project.

1.1 Limitations of Project:

 The project deals with the application of only three process parameters (stand-off-
distance, abrasive flow rate, traverse rate) and left other parameters (like water
pressure, surface roughness, garnet embedment, impingement angel, surface finish,
kerf geometry, MRR, waviness, etc.) due to high cost of machining.
 Due to high costing and non-availability of signal capturing devices we are unable to
do online process monitoring by using signal processing.

5
 CHAPTER 2
LITERATURE SURVEY
This chapter includes study of research papers and gap analysis. Different papers published
on Elsevier and certain journals involving the various experiments pertaining cutting of Ti-alloy
have been reviewed below. These papers show effect of different process parameters on respective
response, different online process monitoring techniques with the help of various sensors and also
use of AI techniques to predict the response

2.1 Literature Survey

The survey considered in this work are categorized in the following subsections separately as
abrasive water jet cutting of Titanium alloy, online process monitoring using various techniques and
application of Artificial Intelligence in process monitoring.

2.1.1 Literature Survey on Abrasive water jet cutting of Titanium alloy:

The following literature review is based on the Machinability of AWJM on various

materials such as Titanium alloys, Ceramics etc. Li and Wang [1] presented an experimental study
of abrasive waterjet machining of Ti-6Al-4V. They have conducted two types of AWJ machining
operation drilling & slotting. In hole drilling the relation between drilling and water pressure were
investigated, it was found that the diameter and hole depth were increased as the time of drilling
were increased but in a decreasing rate. And in the slot cutting the relation between the water
pressure and traverse speed were examined, where they found out that slower traversed speed
resulted in deeper depth cut. Hascalik et al. [2] experimented on the effect of variation in traverse
speed of AWJ on Ti-6AL-4V. In this study, they have investigated the profiles of machined
surfaces, micro-structural features of the machined surfaces and the kerf geometries using the
traverse speed profilometry and scanning electron morphology (SEM). The results shown that the
traverse speed of the jet is a significant parameter on surface morphology and width changes
according to it. Boud et al. [3] shown the influence of the abrasive morphology and mechanical
properties on workpiece grit embedment. In this experiment, the degree of embedment of five-types

6
of garnet abrasive was measured following the through-cutting of Ti-6Al-4V plate, both on the TSI
region and the cut surface. TSI region had much higher degree of embedment (between 12% and
17% depending on abrasive type and cutting conditions) than on the cut face (between 4% and 9%
depending on the abrasive type and cutting conditions), due to higher impacts forces in the former
area. They concluded that when impacts forces are likely to result in particle fracture, grit shape is
not critical parameters in the choice of abrasives for waterjet cutting, both in terms of the cutting
efficiency and in the terms of the tendency to embed the grit.
Hlavac [4] investigated the jet trajectory curvature inside the kerf of the water jet. Two types
of experiments were done one was to determine a theoretical limit value of material thickness for
the traverse speed used for performance of the experimental cut in the given material. The other
type of experiment was based on the theoretical calculation of the traverse speed limit for a given
material thickness. They derived a function describing the curvature of the jet trajectory inside the
cut material and their dependence upon traverse spend, material properties and the jet parameters
were established. Fowler et al. [5] investigated the role of jet-workpiece traverse speed, two
different grit sizes were used for AWJ cutting. Process parameters under investigation were traverse
speed and size of the garnet abrasive particles. All other process variables such as water pressure,
nozzle stand-off distance, abrasive flow rate, jet impingement angle was kept constant. The stand-
off distance was kept 3mm for the maximum Material Removal Rate (MRR). The effect of traverse
speed on MRR for both grit size were high at lowest traverse speed and decreased rapidly with
increase in the traverse speed. Also increased traverse speed results in reduction in surface waviness
for both grit size. High MRR and high surface waviness was observed at low jet traverse also
surface waviness increased with no of passes of jet and vice-versa for high traverse speed. For the
two grit sizes examined, the MRR, surface waviness, surface roughness was lower for smaller grits.
Machine surface was observed by Seo et al. [6] which recognized the process parametric effects on
surface quality, mechanism of MRR and characteristics of generated surfaces by using surface
profilometry, Scanning Electron Microscopy (SEM) and Energy Dispersive Spectroscopy (EDS)
analysis. It was noticed that entrance kerf width is most significantly affected by the stand-off-
distance, exit kerf width and taper ratio by mesh number and the slope by stand-off-distance. J.
Wang [7] presented an experimental study on predictive depth of jet penetration models for AWJM
cutting of alumina ceramics with controlled nozzle oscillation. He investigated the effects of nozzle
oscillations on the depth of cut at small angles by varying combinations of process parameters. In
this study the dimensional analysis was used to develop mathematical equation for the depth of cut
7
in terms of process variables, dimensional analysis is a powerful analytical powerful technique in
describing the relationship between physical engineering quantities and independent variables. The
depth of cut was increased by 82% when the nozzle oscillation angle kept small while other
parameters were selected correctly. Also, the combination of high oscillation frequency (10-14Hz)
and small oscillation angels (4-60) are considered for maximizing depth of cut in nozzle oscillation
cutting. Predictive models for cutting performance measure in AWJ cutting with and without nozzle
oscillation have been developed to mathematically predict the depth of cut for optimization and
process planning. Van et.al. [8] presented the experimental model in which he studied pocket depth
in blind machining using an abrasive water jet in a context where a machine, a pressure and
abrasive are given. The results presented shown an extremely good correlation between the depths
obtained and obtained using model and depths measured.
Shipway et al. [9] shown that the surface characteristics developed during AWJ-CDM of Ti-
6Al-4V to depend upon the process parameters. This experiment examines the AWT milling
behavior of Ti6al4v in the terms of the surface properties of the milled components, such as
roughness, level of grit embedment and the waviness. The properties of the surface following
milling depends on the milling parameters, such as jet-workpiece traverse speed, impingement
angle, waterjet pressure and abrasive size. Fowler et al. [10] investigated the effect of particle
hardness and shape when AWJ milling Ti-6Al-4v.MRR and surface characteristics were analyzed
after single passes of jet. Orifice and nozzle wear due to machining operation can change in
characteristics to AWJ. To minimize these effects, both nozzle and orifice were changed. Here
parameters investigated were traverse speed, shape and hardness of abrasive particles while other
process parameters were kept constant. Results showed that MRR increased at slow traversed speed
and decreased rapidly with increased in traversed rate.
2.1.2 Literature Survey on Online Process Monitoring using various techniques:

The online process monitoring techniques considered in this work are categorized in the following
subsections separately as cutting force, acoustic emission (AE), vibration, sound detection and infrared
(IR) thermography, respectively used in predicting depth of cut by using different performance
parameters in AWJM.
Hassan et al. [11] predicted that root mean square of acoustic emission (AErms) is more
predominant than cutting force response to measure depth of cut in AWJ cutting through online

8
monitoring. They performed an experiment using signal processing method, commonly called as
wavelet transform (WT). Vertical cutting force technique is not used due to erosive nature of AWJ.
Hlochand Ruggiero [12] used vibration emission technique for the analysis of hydroabrasive cutting.
Graphs obtained from experiments shown that higher abrasive mass flow rate shows amplitude growth
with low frequencies in case of frequency analysis.AE signals may be continuous type or burst type.
Ductile material shows continuous type AE signals whereas unsteady process shows burst type. Mohan
et al. [13] used closed loop online monitoring for depth of AWJ penetration using time domain AE
signal and Fast Fourier Transform (FFT) analyzer. They also found out power spectrum density (PSD)
and cutting efficiency to indicate depth of cut and to control kerf depth in AWJ machining.
Kovacevic et al. [14] monitored thermal energy distribution in AWJ cutting using Infrared (IR)
thermography by using opaque materials like Aluminum and Titanium. Aim here is to measure
maximum temperature at cutting region and monitoring of radial and axial wear of nozzle. Temperature
distribution in AWJ cutting nozzle is also observed using thermocouples and compared those results
with IR thermography. Smooth and striated regions were observed in AWJ along kerf. At deformation
wear zone i.e. in striated region, workpiece normal force is the better technique for online process
monitoring of surface finish. Using parameters like Green's function, auto co-variance function and
power spectrum density; Kovacevic et al. [15] established that roughness of surface profile can be
decreased by increasing water pressure and component of workpiece normal force. For surface profile
monitoring, dynamic workpiece normal force is the main parameter. Axinte and Kong [16] studied
multi-sensing technique to determined unnatural events occurring at both nozzle and workpiece levels
for 5-axis AWJ machine using variations of input energy at nozzle, utilized energy at workpiece and idle
energy at jet. Nozzle clogging and jet penetration can also be determined by using AE technique.

Lisseka et al. [17] used Carbon Fibre Reinforced Polymers (CFRP) plates to predict cumulative
burst effect; single burst cannot be identified. They used AE technique for the evaluation of machining
parameters. Higher feed rates decrease cutting quality and also decreases cumulative count of bursts
exponentially. Changes of machining parameters of CFRP determined by cumulative count of bursts and
energy per cutting volume. Experiments performed by Kovacevic [18] shown that impacting jet could be
used as an indicator for achieving a depth of cut of AWJ penetration when the workpiece force is
generated normally. Also stated that nozzle wear rate is affected by change in waterjet pressure. Mohan
et al. [19] determined the nozzle internal diameter which is the most critical part of an AWJ cutting
system using Artificial Neural Network (ANN). With the help of piezoelectric actuator mechanism,

9
position of AWJ cutting head can be controlled on a large scale. Corresponding nozzle internal diameter
determined by ANN through frequency domain acoustic signals. Jurisevic et al. [20], interpreted direct
proportional relation between RMS value and Standoff distance. On basis of graphs, a correlation
between sound signal and stand-off-distance can be achieved in time and frequency domain. Jurisevic et
al. evaluated the root mean square (RMS)value in the time domain and the amplitude cumulative sum
(ACS) of the signal power spectra in the frequency domain. Using generated sound monitoring, stand-
off-distance during AWJ cutting process can be controlled.

2.1.3 Literature Survey on application of Artificial Intelligence in process

monitoring:

Now a day's Artificial intelligence tools are replacing most of the human tasks in operating
and controlling the many processes in industries, which makes processes more accurate and time
saving. This literature survey is based on the different AI (Artificial Intelligence) techniques like
Artificial Neural Networking (ANN), Fuzzy logic, Adaptive Neuro-fuzzy Inference system (ANFIS)
used in predicting and optimizing machine performance parameters in different machining
processes. ANN is a computational system inspired by structure, processing method and learning
ability of biological brain. ANN consists of input layers, hidden layers and output layers. ANN can
be trained by two techniques: supervised and unsupervised/adaptive trainings.
Khan et al. [21] used artificial neural network(ANN) [Multilayer perceptron(MLP) with 3
hidden layer feedforward networks] to develop Artificial Intelligence (AI) model to predict surface
roughness of Ti-15-3 alloy in Electrical discharge machining (EDM) with peak current, pulse on
time, pulse off time, servo voltage as input parameters. GAO et.al.[22]developed a parameter
optimization model using Genetic Algorithm(GA) and Artificial Neural Network (ANN) with
Levenberg Maquardt algorithm adaptation which represents relation between material removal rate
(MRR) and input parameters in Electric Discharge Machining (EDM). Proper machining parameters
to get higher MRR are selected by using GA. Among 3 layers of network hidden layer is activated
with hyperbolic tangent sigmoid transfer function and output layer is activated by linear transfer
function. Joshi and Pande [23] Proposed an intelligent process model for EDM using FEM, ANN
and GA. Integrated approach of model development (FEM-ANN-GA) is based on FEM analysis
instead of experimental data (which reduces error and minimizes a cost). Material properties,
discharge duration, discharge current and voltage, and duty cycle are used to develop a FEM

10
numerical model then prediction of MRR, TWR, crater depth and radius of workpiece is done by
ANN (radial basis function neural network and feed forward back propagation neural network).
Genetic algorithm(GA) is used for optimization of parameters like current, discharge duration,
discharge voltage and duty cycle. Pradhan and Bhattacharyya [24] developed a mathematical model
to optimize the machining characteristics like MRR, TWR and overcut(OC) using Response Surface
Methodology and ANN with peak current, pulse on time, dielectric flushing pressure as input
parameters in micro EDM of a Ti-6Al-4V super alloy. Process model developed using ANN as one
hidden layer with 12 neurons and Levenberg-Marquardt used as a training algorithm. Somashekhar
et al. [25] optimized a MRR in micro EDM using ANN with input, hidden and output layers and GA
is used to determine optimum process parameters. Gap voltage, capacitance, feed rate and speed are
used as input parameters. ANN built in MATLAB is feed forward Neural Network with back
propagation algorithm.
Rajmohan et al.[26] optimized machining parameters in drilling of hybrid aluminum metal
matrix composites using the grey-fuzzy algorithm. Fuzzy rule-based reasoning integrated with
Taguchi's method used to reduce the degree of uncertainty during the decision making. For producing
high quality products at low cost L27 3-level orthogonal array is used for experiments. ANOVA is
used to find the highly influential drilling parameters that contributes to a high-quality product. Kohli
et al.[27] optimized material removal rate (MRR) in EDM with copper electrode working on medium
carbon steel (AISI- 1040) workpiece using Fuzzy Logic. Discharge current, pulse on time, pulse off
time are taken as input parameters. Six angular membership functions are selected for fuzzy models
in fuzzy inference system (FIS). Pandey and Dubey [28] improved the geometrical accuracy of the
Duralumin sheet in laser cutting and simultaneously minimized kerf width and kerf deviation on both
top and bottom sides. Robust parameter design methodology and fuzzy logic theory used to compute
fuzzy multi-response performance index. Multi-objective optimization is done by using multi-
response performance index. Also concluded that oxygen gas pressure is the most significant factor
followed by pulse frequency (that is directly proportional to pulse energy) in laser cutting of highly
reflective and thermally conductive material like Duralumin.
Maher et al. [29] predicted surface roughness of brass using ANFIS in CNC end milling.
Spindle speed, feed rate, and depth of cut were used as predictor variables. Fuzzy inference system
(FIS) incorporated in the neural network (NN) environment is used to design machine learning
algorithm. This hybrid arrangement offers a twin benefit of human-like reasoning quality in
conjunction with an adaptive network which is accountable for enlightening the fuzzy rules. Sen et al.
11
[30] ANFIS modeling was utilized to predict the relationship between surface roughness, cutting
force and cutting temperature with speed, feed, depth of cut and width of cut as inputs in CNC
milling of Inconel 690 using ANFIS. Machine learning algorithm is developed using fuzzy inference
system (FIS) incorporated in the neural network (NN) environment. This hybrid arrangement offers a
twin benefit of human-like reasoning quality in conjunction with an adaptive network which is
accountable for enlightening the fuzzy rules.

12
2.2 Summary of literature Survey:
Summaries of all literature survey which is described above is in table format are as follows:

Table 2.1.1: Summary of Abrasive water jet cutting of Titanium alloy

Ref. Material Process Parameter Response Remark

P V H F G α

[1] Ti-6Al-4V * * Depth of cut Two types of machining operations, drilling and slotting
were conducted. The influence of water pressure, traverse
speed and machining time were investigated.

[2]. Ti-6Al-4V * * * * * Surface The surface morphology, kerf geometry and surface
roughness and roughness were investigated by using different traverse
depth, Kerf speed of the water jet machine. And by this suggested that
geometry traverse speed significantly changes the outcome.

[3] Ti-6Al-4V * * * * Fracture Grit shape is not a critical parameter in the choice of
abrasives for abrasive waterjet cutting, both in terms of the
cutting efficiency and in terms of the tendency to embed
the grit.

[5] Ti-6Al-4V * * * * Roughness, At low jet traverse speed, high MRR and low surface
waviness, waviness and vice versa.
MRR.

13
 [6] Ti-6Al-4V * * * * Surface Finish, The profiles of machines surfaces, curved geometry micro
Kerf Geometry structural features of machine surface were analyzed using
surface profilometry, SEM and EDX.

[7] Alumina * * * * Depth of cut Nozzle oscillation cutting can cause negative effect on
Ceramics depth of cut if cutting parameters are not correctly selected.

[8] Ti-6Al-4V * * * * MRR The feed rate influences the width and depth of an
elementary pass and the depth of the pockets will only
depend on the feed rate and the sweep pitch.

[9] Ti-6Al-4V ** * * * * Surface At high jet traverse velocities, grit embedment increases
Waviness and with impact angle due to higher impulse during impact as
MRR the ∝ is raised.

[10] Ti-6Al-4V * * * * * Surface Shape factor and particle hardness have no significant
roughness & effect on surface waviness.
morphology,
erosion.

Table 2.1.2: Summary of Online Process Monitoring

Ref. Type of
Response/Signal Sensors T F W Conclusion
[11] Cutting force, Strain gauge AErms increases withincrease in depth of cut
Acoustic emission (AE) dynamometer, * *
AE sensor

14
[12] Vibration Accelerometer higher abrasive mass flow rate shows amplitude growth with
* * * low frequencies in case of frequency analysis

[13] Acoustic emission (AE) AE sensor graph shown linear relationship between peak of PSD and kerf
* * depth

[14] Infrared (IR) Linescans and Thermal energy distribution in the workpiece can be easily
thermography isotherms * monitored.

[15] Cutting force Dynamometer roughness of surface profile can be decrease by increasing
* water pressure and component of workpiece normal force

[16] Acoustic emission (AE) Dynamometer, AE multi-sensing technique was used to determined unnatural
sensor * events occurring at both nozzle and workpiece levels

[17] Acoustic emission (AE) Piezoelectric Increase in feed rates, decreases cutting quality and also
diaphragm, AE * decreases cumulative count of bursts exponentially but energy
per cutting volume increases

[18] Cutting force Piezoelectric nozzle wear rate is affected by change in waterjet pressure
crystal force sensor

[19] Sound detection Microphone Nozzle diameter can be easily predicted from unknown
* * acoustic signal by using ANN

[20] Sound detection Microphone, piezo direct proportional relation between RMS value and stand-off-
sensor * * distance

15
Table 2.1.3: Summary of application of Artificial Intelligence in process monitoring

Ref. AI Process No. of Inputs Response Algorithm Type of Error

Technique Version

[21] ANN (MLP) EDM 4 Ra L-M MSE

[22] ANN & GA EDM 3 MRR L-M MSE

[23] FEM–ANN–GA EDM 4 Shape of Crater, SCG & MP

MRR, NSGA
TWR

[24] ANN Micro- 3 MRR L-M MP

EDM TWR
OC
[25] ANN & GA Micro- 4 MRR Back propagation RMSE
EDM Algorithm

[26] Fuzzy-Logic Vertical 3 Thrust Force, Burr, Grey fuzzy Average Percentage Error
CNC Ra Algorithm
machine Grey Relational
Analysis
L27 orthogonal
array
ANOVA
[27] Fuzzy-Logic EDM 3 MRR Membership Average Percentage Error
Function

16
[28] Fuzzy-Logic Laser- 4 Kerf deviation FMPI PPE
Cutting

[29] ANFIS CNC-End 3 Ra gbellmf Average Percentage Error

milling

[30] ANFIS CNC- 4 Ra, Fc, Tc Prop learning RMSE&

Milling algorithm, Mean Absolute
Fuzzy hybrid Percentage Error
learning Algorithm

17
2.3 Gap Analysis:

 Though there are plenty of papers published on AWJM in various fields like
optimization techniques, process monitoring and surface finish monitoring there is no
work found on online process monitoring to predict depth of cut in AWJM.

 There is no work on integration of AI techniques which can be used to predict the depth
of cut in AWJM.

 None of the researchers compared various AI techniques and Hybrid techniques to

predict the depth of cut.

After the study of work done by various researchers in field of AWJM, online process
monitoring and artificial intelligence it is found that the proper integration of artificial
intelligence and the online process monitoring to predict the depth of cut can make AWJM the
best and fast manufacturing process.

18
 CHAPTER 3
PROBLEM DEFINITION AND OBJECTIVES
This chapter consists of the Problem Definition and Gap analysis induced from the
conclusions obtained by studying the literature review. With the help of various reviews above
we selected few parameters which are discussed below. The methodology of the whole process is
explained in this chapter.

3.1 Problem Definition:

 Ti grade 5 has got vast range of application in aerospace industry. Cutting of this material
by conventional CNC milling is too costly due to frequency tool replacement. Therefore,
non-conventional process like AWJM is used to cut this material.

 The major issue in AWJ cutting of this material is improper depth of cut. If the depth of
cut is incomplete, the number of pass need to be increased. Hence this works deals with
online process monitoring of depth of cut by application of AI techniques to get better
performance in minimum cost.

3.2 Objectives:

 To find the effects of parameters on variation produced in response (Depth of cut).

 To apply artificial intelligence techniques like Artificial Neural Network, Fuzzy logic in
prediction of depth of cut.
 To compare Artificial Intelligence technique in prediction of response in water jet cutting
of Ti alloy.

19
3.3 Overall Methodology:

 After the study of various research papers on cutting of Ti-6Al-4V stand-off distance,
abrasive flow rate, water-jet pressure and traverse rate are taken as process parameters.
 Parameter selection and level selection are necessary for test trial after getting objective.
 Then the number of cuts required for experimentation are calculated using Full Factorial
Method.
 Then the study of research papers on online process monitoring used to find the appropriate
method of getting a signal to process for optimizing the process parameters and prediction of
depth of cut.
 Experimentation model developed by considering various factors like process parameters,
availability of FFT analyzer, software availability etc.
 Signal processing is carried out after getting measure of response by using FFT analyzer.
Then the optimization is done using fuzzy logic and TLBO.
 After optimization for prediction of depth of cut various Artificial Intelligence techniques
and the results of these tested with experimental values.

20
Fig. 3.3: Methodology Chart

21
 CHAPTER 4
EXPERIMENTATION AND MEASUREMENT OF
RESPONSE
This section gives details about test run, design of experimentation, measurement of
response and experimental setup. In terms of performance, to achieve good results and stability
of system, test run is needed.

4.1 Test run

 Test run Was done successfully to find out what parameters we should select for design
of experimentation.

 Also, selection of Size / Dimensions of workpiece were done on basis of test run

Fig. 4.1: Test Run on Ti-alloy

4.2 Design of Experimentation:

Design of experimentation was carried out in Minitab Software,
 Number of levels are 4 as shown in table 4.2
 Number of factors are3
 For experimentation, a full factor experimental design is done by first dividing the range
of each parameter in 4 levels and obtaining 64 experimental combinations.
By using General Full Factorial Design,
43=4*4*4=64

22
Parameters like stand-off-distance, abrasive flow rate and traverse rate are used predict depth of
cut in mm and rest of the parameters are kept constant.

Table 4.2: Input Parameters and their Levels

Parameters Units Level 1 Level 2 Level 3 Level 4

Stand-off-Distance (SOD) mm 2 3 4 5

Abrasive Flow Rate (AFR) Lb./min 1 1.2 1.4 1.6

Traverse Rate (TR) mm/min 300 400 500 600

4.2.1 Measurement of Response

 Depth of Cut was our predefined response.

 Measuring a depth of cut after AWJ cutting of titanium material it was necessary to
measure its response for further optimisation and prediction

 To do so we found a most accurate technique

-Image processing in MATLAB

4.3 Experimental Setup:

 A plate of Ti-6Al-4V (grade 5) having size 264.5mm x 64.5mm x 9.5mm is chosen as the
work piece.
 As discussed in the literature review, a large number of variables are involved in the
AWJM and virtually all these variables affect the cutting results. Therefore, only those
parameters are selected which shows a considerable influence on objectives of the study
i.e. depth of cut.
 Water-jet machine setup as shown in fig. 4.3

-Water pressure : 60000 PSI

-Orifice Diameter : 0.25mm
-Abrasive size(Grit no.) : 120
- Nozzle length : 76mm

23
 (a)

(b)

Fig. 4.3: (a) and (b) Experimental Setup

24
4.4 Experimentation
After design of experimentation and experimental setup, 64 cuts have been taken on
material Ti-6Al-4V by varying process parameters to get different values of depth of cut.

Fig. 4.4: Material Ti-6Al-4V with 64 Cuts

To get more accurate results of 64 depth of cuts the material is cut into two halves with the help
of wire EDM machine.

Fig. 4.5: Two Halves of Material with help of Wire EDM Machine

25
4.5 Image Processing in MATLAB Software
In this process we did the measurement of response successfully by using MATLAB by using following steps:
 We captured the images of the experimented workpiece using DSLR camera by keeping a Vernier Scale.
 Then these images are processed under the image processing toolbox in the MATLAB.
 While doing image processing the new scaling factor is assigned to the vernier scale and used to calculate the depth of cut as
shown in the fig. 4.4
 Example: consider a reading no.2 from the fig. 4.4
here, scaling factor is 112.2 and
Scaling factor for 1mm of vernier scale is 22
Therefore Depth of Cut in (mm) = 112.2/22 = 5.1
 Similarly other Depth of cuts can be found out.

Fig. 4.6: Image Processing

26
After applying image processing technique successfully the responses (D.O.C) are calculated
and tabulated as shown in table 4.4.

Table 4.4: Measurement of Depth of Cut through Image Processing

INPUT OUTPUT

SOD (mm) AFR (Lb/min) TR (mm/min) DOC (mm)

2 1 300 4.9431
2 1 400 3.375
2 1 500 2.477
2 1 600 1.5
2 1.2 300 5.2613
2 1.2 400 3.6818
2 1.2 500 2.9091
2 1.2 600 1.6136
2 1.4 300 5.572
2 1.4 400 4.654
2 1.4 500 3.159
2 1.4 600 2.284
2 1.6 300 7.0595
2 1.6 400 4.6818
2 1.6 500 3.625
2 1.6 600 3.0113
3 1 300 3.85
3 1 400 2.2438
3 1 500 1.5636
3 1 600 1.378
3 1.2 300 4.1342
3 1.2 400 3.1713
3 1.2 500 2.2261
3 1.2 600 1.899
3 1.4 300 4.832
3 1.4 400 3.127
3 1.4 500 2.765
3 1.4 600 1.8462
27
3 1.6 300 5.733
3 1.6 400 4.028
3 1.6 500 3.0918
3 1.6 600 2.4293
4 1 300 3.31
4 1 400 2.11
4 1 500 1.58
4 1 600 1.26
4 1.2 300 3.74
4 1.2 400 2.57
4 1.2 500 2.1004
4 1.2 600 1.6
4 1.4 300 4.89
4 1.4 400 3.16
4 1.4 500 2.37
4 1.4 600 1.82
4 1.6 300 5.69
4 1.6 400 3.89
4 1.6 500 2.84
4 1.6 600 2.35
5 1 300 2.49
5 1 400 1.525
5 1 500 1.4363
5 1 600 1.22
5 1.2 300 3.94
5 1.2 400 2.5034
5 1.2 500 1.83
5 1.2 600 1.46
5 1.4 300 4.84
5 1.4 400 2.99
5 1.4 500 2.875
5 1.4 600 1.79
5 1.6 300 5.14
5 1.6 400 3.6
5 1.6 500 2.875
5 1.6 600 2.096

28
 CHAPTER 5
EFFECT OF PARAMETERS ON RESPONSE BY USING
AI TECHNIQUES
Physics based process modeling using finite element method (FEM) has been integrated
with the soft computing techniques like artificial neural networks (ANN), genetic
algorithm(GA), adaptive-network-based fuzzy inference system (ANFIS), Fuzzy Logic to
improve prediction accuracy of the model with less dependency on the experimental data.
Researchers in Artificial Intelligence (AI) follow the algorithmic approach and try to capture the
knowledge of an expert in some specific domain as a set of rules to create so called expert
systems. For our project, we used two techniques for prediction of response:
- ANN
- ANFIS

5.1 Artificial Neural Network (ANN)

An Artificial Neural Network is a parallel, distributed information processing structure
consisting of processing units (which can possess a local memory and can carry out localized
information processing operations) interconnected via unidirectional signal channels called
connections. Each processing unit has a single output connection that branches ("fans out") into
as many collateral connections as desired; each carries the same signal - the processing unit
output signal.
Artificial neural network (ANN) is an effective method to solve non-linear problem.
ANN is inspired from the structure of biological neural networks and their way of encoding and
solving problems. Artificial Neural Networks (ANN), also called neurocomputing,
connectionism, or parallel distributed processing (PDP), provide an alternative approach to be
applied to problems where the algorithmic and symbolic approaches are not well suited.
It includes
- Creation of Neural Network
- Training a Neural Network
- Validation of Neural Network
- Testing a Neural Network
- Prediction using Neural Network

5.1.1 Structure of Neural Network

Amongst the architecture of ANN like 3-1-1, 3-2-1, 3-3-1; we choose 3-2-1 as neural
network as shown in fig. 5.1 in which 3 are input layers, 2 are hidden layers and 1 is output layer
i.e. depth of cut. Back Propagation Algorithm is considered for this network.

29
 Fig. 5.1.1.: 3-2-1 Neural Network

Out of 64 samples, 44 samples are used for training, 10 used for validation and remaining
10 samples are used for testing to get accurate results. Mean Square Error (MSE) and Regression
coefficient (R) of respective samples are shown in table 5.1.

Table 5.1.: Results of ANN Model

Samples MSE R
Training 44 7.59439e-2 9.76757e-1
Validation 10 2.8626e-1 9.20472e-1
Testing 10 8.94927e-1 8.71362e-1

5.1.2 ANN Result

This topic shows the predicted values of D.O.C.'s for sample of process parameters
- RMSE=0.532
- Regression Coefficient (R)=0.9327
- Best validation performance=0.826 at (epoch-3)
- Predicted values of DOC for given process parameters

30
 Table 5.1.1: Prediction of DOC by ANN

SOD AFR TR Predicted DOC

6 1 300 3.0544
6 1 400 2.1338
6 1 500 1.7198
6 1 600 1.482
6 1.2 300 4.5736
6 1.2 400 2.5677
6 1.2 500 1.8384
6 1.2 600 1.6467
6 1.4 300 5.1457
6 1.4 400 2.7759
6 1.4 500 2.4224
6 1.4 600 2.1308
6 1.6 300 5.1495
6 1.6 400 3.2382
6 1.6 500 2.8627
6 1.6 600 2.8705

31
5.1.3 ANN Plots
After training of neutral network different plots comes up which shows the performance, fitting, regression and error of the
neural network.
1st plot (fig. 5.1.3) shows the Regression plots for the training, validation and testing with respect to target data this plot gives
the regression coefficient for the training, validation and test data 4th of the four plots of the regression plot mentioned below shows
the all data points and gives the overall regression coefficient R=0.92816
From the plots we get a regression coefficient (R) and we have to train a data until we gets the best value of the R. Here in our
case the regression coefficient r for the test is found to be 0.97329 which is close to 1.

Fig. 5.1.3: ANN Regression Plot

32
2nd plot (fig. 5.1.4) shows the ANN performance plot. In this plot the mean square error of the training state is plotted against the
epochs (10 epochs were selected). This error should be as much as minimum as possible. To minimize this error the algorithm used is
gradient descent algorithm. here 3 types of errors training validation and testing errors becomes minimum at the epoch no.3. And the
best validation performance is 0.26826 at epoch 3.

Fig. 5.1.4: ANN performance Plot

33
5.2 ANFIS (Adaptive Neuro Fuzzy Inference System)
The architecture and learning procedure underlying ANFIS (adaptive-network-based
fuzzy inference system) is presented which is a fuzzy inference system implemented in the
framework of adaptive networks. By using a hybrid learning procedure, the proposed ANFIS can
construct an input-output mapping based on both human knowledge and stipulated input-output
data pairs.
ANFIS uses both training data and user defined parameters in optimization procedures.
User defined parameters include: number of membership functions per input, membership
function type, step size, increasing and decreasing step size, and the number of epochs used in
training.
 It is a hybrid technique used for prediction of Response
 It includes
- Generation of Fuzzy Inference System (FIS)
- Training of FIS using hybrid or back-propagation Algorithm Optimisation
Method
- Testing of a FIS/ checking of Data
- Prediction of a given process parameters using predicted results of ANN

5.2.1 The ANFIS Architecture

It is a machine learning algorithm which utilizes fuzzy inference system (FIS)
incorporated in the neural network (NN) environment. In ANFIS, Fuzzy logic membership
functions are tuned with the aid of hybrid learning algorithm i.e. an amalgamation of the least-
square method and backpropagation gradient descent method for acclimatizing to the
environments. The Fuzzy logic system conveys human-like interpretation features into the
ANFIS. Such hybrid arrangement offers a twin benefit of human-like reasoning quality in
conjunction with an adaptive network which is accountable for enlightening the fuzzy rules.
Actually, ANFIS structural design consists of five different layers. They are input layer,
fuzzification layer, inferences process layer, defuzzification layer and the final output layer.
Every single layer executes a particular job to promote the signals. It’s a network structure
consists of a number of nodes affixed through direct links. Here fig. 5.1.2 illustrates the ANFIS
algorithm.

34
Fig. 5.2.1: The Architecture of ANFIS

35
5.2.2 ANFIS Plots
ANFIS generates the different surface plots showing combine effect of 2 process parameters on
the response (D.O.C.) in 3D.

(a)
1st 3D plot shows the effect of first 2 input parameters on the output i.e. Response (D.O.C.). It
shows the combine effect of S.O.D. and A.F.R. on the D.O.C.
2nd 3D plot shows the effect of A.F.R. and T.R. on the D.O.C.

(b)

Fig. 5.2.2: ANFIS Result (I/P vs O/P)

36
5.2.3 Fuzzy Inference System (FIS)
In ANFIS to optimize the Neural network we use the Fuzzy Inference System (F.I.S). The following fig. 5.3 shows the different
membership functions which affects the optimized model of F.I.S. Membership functions of the inputs are of the gbellmf and output
membership function is of triangular type

Fig. 5.2.3: Fuzzy Inference System (FIS)

37
  Average checking error- 0.40869
 Average Epoch error – 0.144

Calculation of predicted values of ANFIS by using predicted values of ANN and average
checking error

Table 5.3: Prediction of DOC by ANFIS

SOD AFR TR Predicted DOC

6 1 300 2.44544
6 1 400 1.338
6 1 500 1.4598
6 1 600 1.2482
6 1.2 300 4.12736
6 1.2 400 2.5677
6 1.2 500 1.8384
6 1.2 600 1.4967
6 1.4 300 4.6457
6 1.4 400 3.1759
6 1.4 500 2.34
6 1.4 600 1.808
6 1.6 300 5.2138
6 1.6 400 3.6822
6 1.6 500 2.8627
6 1.6 600 2.367

38
5.3 ANFIS Vs ANN Results
Moreover, the ANFIS acquired results were compared with an Artificial Neural Network
(ANN) model, developed on the identical parametric ranges. The comparison of the Predicted Value by ANFIS
obtained results indicated that the ANFIS overtakes the ANN model in predicting the
preferred response variables, which suggests the modesty of the ANFIS model. Predicted Value by ANN

Fig. 5.3: ANN vs ANFIS Results

39
 CHAPTER 6
RESULTS AND CONCLUSION
The results, conclusion and future scope of this project is explained below. The results
show a successful prediction on response. The conclusionsummarizes project results in a few
sentences and state whether results support or contradict the hypothesis. The future scope
summarizes what we are not able to do because of time constraint.

6.1 Results
On account of experimental effort, ANFIS and ANN model have been built up to predict
depth of cut. It is found that prediction of the output parameters of abrasive waterjet cutting
machine by means of ANFIS gave extraordinary correlation statistics as compared to ANN
model as shown in table 6.1.
 This work predicted the depth of cut by ANN and ANFIS with average error of 0.532 and
0.4089.

 The maximum D.O.C. was found 7.0595 at the following setting: S.O.D.=2mm,
A.F.R.=1.6lb/min, T.R.=300mm/min.

 Amongst the architecture of ANN (3-1-1, 3-2-1, 3-3-1), 3-2-1 gave minimum error in
prediction.
Table 6.1: Comparison of Experimental and Predicted Values

Experimental Values DOC Predicted by ANN DOC Predicted by ANFIS

2.5034 2.7482 2.615
1.83 1.8157 1.842
1.46 1.233 1.4578
4.84 4.66 4.78
2.99 3.1534 3.012
2.875 2.598 2.76
1.79 1.552 1.6802
5.14 4.99 5.14
3.6 3.8228 3.71
2.875 2.8904 2.84
2.096 1.9579 2.14

40
6.2 Conclusion:
The architecture of adaptive-network has been described based on fuzzy inference
systems (ANFIS) with 3-2-1 neural structure. By employing a hybrid learning procedure, the
proposed architecture can refine fuzzy if-then rules obtained from human experts to describe the
input-output behavior of a complex system. The objectives which are mentioned in chapter no. 3
are fulfilled. These results indicate that the ANFIS model with gbellmf is accurate and can be
used to predict depth of cut in abrasive water jet machining.
 Image processing was successfully applied in prediction of depth of cut.

 The effects of different process parameters on variation produced on D.O.C. found

successfully.
 Artificial Intelligence techniques like ANN and ANFIS in prediction of depth of cut was
successfully applied.

 After comparing different AI techniques, ANFIS is the better hybrid technique.

6.3 Future Scope:

In this project work depth of cut is predicted using different AI techniques by keeping
stand-off-distance, abrasive flow rate, traverse rate as a process parameters. In future this work
can be done as:
 Various process parameters can also be varied like waterjet pressure, surface roughness,
kerf length, grit embedment, etc.

 Also different hybrid techniques like GA-Fuzzy, GA-ANFIS, GA-Fuzzy-ANN, etc. can
be used to predict the response.

 Most awaiting technique of Quantum Neural Networks (QNN) can be applied when no.
of process parameters are more and the accuracy required is higher in quantum
computers.

 Along with prediction techniques signal processing can be applied for online process
monitoring.

41
 REFERENCES

[1] Huaizhong Li, Jun Wang, “An experimental study of abrasive waterjet machining of Ti-6Al-
4V,” Int J Adv Manuf Technol, 2015.

[2] A Hascalik, U Caydas and H Gurun, “Effect of traverse speed on abrasive waterjet machining of
Ti-6Al-4V alloy,” Materials and Design, vol.28, pp.1953-1957, 2007.

[3] F. Boud, C. Carpenter, J. Folkes, P.H. Shipway, “Abrasive waterjet cutting of a titanium alloy:
The influence of abrasive morphology and mechanical properties on workpiece grit embedment and
cut quality,” Journal of Materials Processing Technology, vol.210, pp.2197-2205, 2010.

[4] Libor M. Hlavac, “Investigation of the abrasive waterjet trajectory curvature inside the kerf,”
Journal of Materials Processing Technology, vol. 209, pp.4154-4161, 2009.

[5] G. Fowler, P.H. Shipway, I.R. Pashby, “Abrasive water-jet controlled milling of Ti6Al4V alloy-
an investigation of the role of jet-workpiece traverse and abrasive grit size on the characteristics of
the milled material,” Journal of Materials Processing Technology, vol.161, pp.407-414, 2005.

[6] Y W Seo, M Ramulu and D Kim, “Machinability of titanium alloy (Ti-6Al-4V) by abrasive
waterjets,” Proc. Instn Mech. Engrs, Part B: J. Engineering Manufacture, vol.217, pp.1709-1721,
2003.

[7] J. Wang, ‘Predictive depth of jet penetration models for abrasive waterjet cutting of alumina
ceramics,” International Journal of Mechanical Science, vol.49, pp.306-316, 2007.

[8] Van Hung Bui, Patrick Gilles, Tarek Sultan, Guillaume Cohen and Walter Rubio, “A new
cutting depth model with rapid calibration in abrasive water jet machining of titanium alloy,” Int J
Adv Manuf Technol, 2017.

[9] P.H. Shipway, G. Fowler, I.R. Pashby, “Characteristics of the surface of a titanium alloy forming
milling with abrasive waterjet,” Wear, vol.258, pp.123-132, 2005.

[10] G. Fowler, I.R. Pashby, P.H. Shipway, “The effect of particle hardness and shape when
abrasive water jet milling titanium alloy Ti-6Al-4V,” Wear, vol.266, pp.613-620, 2009.

42
[11] Ashraf I. Hassan, C. Chen, R. Kovacevic, “On-line monitoring of depth of cut in AWJ cutting,”
International Journal of Machine Tools & Manufacture, vol. 44, pp.595–605, 2003.

[12] S. Hloch and A. Ruggiero, “Online Monitoring and Analysis of Hydroabrasive Cutting by
Vibration,” Advances in Mechanical Engineering, 2013.

[13] R. Mohan, A. Momber, R. Kovacevic, “On-line monitoring of depth of cut of AWJ penetration
using acoustic emission technique,” Water Jet Cutting Technology, Mechanical Engineering
Publications Ltd, pp. 649, 1994.

[14] R. Kovacevic, R. Mohan, H.E. Beardsley, “Monitoring of thermal energy distribution in abrasive
waterjet cutting using infrared thermography,” Journal of Manufacturing Science and Engineering,
vol.118, pp.555-563, 1996.

[15] R. Kovacevic, R. Mohan, Y.M. Zhang, “Cutting force dynamics as a tool for surface profile
monitoring in AWJ,” Transactions of the ASME, vol.117, pp.340-350, 1995.

[16] D.A. Axinte, M.C. Kong, “An integrated monitoring method to supervise waterjet machining,”
CIRP Annals - Manufacturing Technology, vol.58, pp.303-306,2009.

[17] Fabian Lissek, Michael Kaufeld, Jacqueline Tegas, Sergej Hloch, “Online-monitoring for
Abrasive Waterjet Cutting of CFRP via Acoustic Emission Evaluation of Machining Parameters and
Work Piece Quality Due to Burst Analysis,” Procedia Engineering, International Conference on
Manufacturing Engineering and Materials, ICMEM 2016, vol.149, pp.67-76, 2016.

[18] R. Kovacevic, “Monitoring the depth of abrasive waterjet penetration,” Int. J. Mach. Tools
Manufact., vol.32, no.5, pp.725-736, 1992.

[19] R.S. Mohan, R. Kovacevic, and T.R. Damarla, “Real-time Monitoring of AWJ Nozzle Wear
Using Artificial Neural Networks,” Transactions of NAMRI/SME, vol. XXII, pp. 253-258, 1994.

43
[20] Bostjan Jurisevic, Daniel Brissaud, Mihael Junkar, “Monitoring of abrasive water jet (AWJ)
cutting using sound detection,” Int J Adv Manuf Technol, vol.24, pp.733-737, 2004.

[21] Md.A.R.Khan, M.M.Rahman, K.Kadirgama, M.A.Maleque and Rosli A.Bakar, “Artificial

Intelligence Model to Predict Surface Roughness of Ti-15-3 Alloy in EDM Process,” World Academy
of Science, Engineering and Technology, vol. 50, pp.194-198, 2011.

[22] Qing GAO, Qin-he ZHANG, Shu-peng SU, Jian-hua ZHANG, “Parameter optimization model
in electrical discharge machining process,” Gao et al. / J Zhejiang Univ Sci A, vol.9, no.1, pp.104-
108, 2008.

[23] S.N. Joshi, S.S. Pande, “Intelligent process modeling and optimization of die-sinking electric
discharge machining,” Applied Soft Computing, vol.11, pp.2743–2755, 2011.

[24] B.B.Pradhan, B.Bhattacharyya, “Modelling of micro-electro discharge machining during

machining of titanium alloy Ti–6Al–4V using response surface methodology and artificial neural
network algorithm,” Proc. IMechE Part B: J. Engineering Manufacture, vol.223, pp.683-693, 2009.

[25] K. P. Somashekhar, N. Ramachandran, and Jose Mathew, “Optimization of Material Removal

Rate in Micro-EDM Using Artificial Neural Network and Genetic Algorithms,” Materials and
Manufacturing Processes, vol.25, pp.467-475, 2010.

[26] T. Rajmohan, K. Palanikumar, S. Prakash, “Grey-fuzzy algorithm to optimize machining

parameters in drilling of hybrid metal matrix composites,” Composites: Part B, vol.50, pp.297–308,
2013.

[27] Amit Kohli, Aashim Wadhwa, TapanVirmani, and Ujjwal Jain, “Optimization of Material
Removal Rate in Electrical Discharge Machining Using Fuzzy Logic,” World Academy of Science,
Engineering and Technology, vol. 72, pp.1674-1679, 2012.

[28] Arun Kumar Pandey, Avanish Kumar Dubey, “Taguchi based fuzzy logic optimization of
multiple quality characteristics in laser cutting of Duralumin sheet,” Optics and Lasers in
Engineering, vol.50, pp.328–335, 2012.

44
[29] Ibrahem Maher and M. E. H. Eltaib and Ahmed A. D. Sarhan, “Investigation of the effect of
machining parameters on the surface quality of machined brass (60/40) in CNC end milling—ANFIS
modeling,” Int J Adv Manuf Technol, 2014.

[30] Binayak Sen, Dr. Uttam Kumar Mandal, Dr. Sankar Prasad Mondal, “Advancement of an
intelligent system based on ANFIS for predicting machining performance parameters of Inconel 690
- A perspective of metaheuristic approach,” Measurement, 2017.

45
46
47
48
49
50
51
52
53
54