experimental and numerical study for the automatic gma welding process

Developed Multiple Linear Regression Model using Genetic Algorithm for Predicting Top-Bead Width in GMA Welding Process.. [1] proposed a method for determining the near-optimal settings

I would also like to express a very special thanks to the members in my research group at the Welding and Automatic Control Laboratory, especially Mr Kim Hak Hyoung, Mr Seo Joo Hwan, Miss Shim Ji Yeon and Mr Jeong Jae Won for all their help on the experiment and valuable discussions in the laboratory

Finally, on this occasion allow me to express my gratitude and appreciation to my elder sister, Doan Thi Quynh Nga, my wife Pham Thi Giang Lam, and my bosom friend Hua Huu Tho for all their support and understanding during the course of this study

Experimental and Numerical Study for the Automatic GMA Welding Process

Doan The Thao

Department of Mechanical Engineering Graduate School of Mokpo National University

(Supervised by Professor Kim Ill Soo)

On the other hand, the welding deformations in all stages of a welding process for achieving the required precision of welded structure must be taken into account and as a result, it is required to predict the welding distortions at the early stages

of welded structure design by using a simulation Therefore, a FEM (Finite

Element Method) model with a suitable moving heat source, which can accurately simulate the thermal histories of GMA welding process, needs to be developed

Furthermore, the determination of trajectory planning and continuous motion of welding torch without colliding with any obstacles for robotic welding system of complex structures should be considered It is better way to build a robotic welding system using the simulation tools in order to identify some of design and planning problems at an early stage

The sequent experiments based on full factorial design have been conducted with two levels of five process parameters to obtain bead geometry of butt and lab joint type using a GMA welding process Four empirical models have been developed: linear, curvilinear, interaction and a proposed intelligent model Regression analysis was employed for optimization of the coefficients of linear, curvilinear, and interaction model, while GA (Genetic Algorithm) was utilized to estimate the coefficients of intelligent model Not only the fitting of these models were checked and compared by using variance test, but also the prediction on bead geometry using the developed models were carried out based on the additional experiments

The thermal analysis models for automatic finding parameters of a moving heat source have also been proposed An algorithm for the combining between GA and the FEM was obtained and verified based on Goldak’s work and additional experiments for multi-pass butt and fillet welded joint with dissimilar thickness

In this algorithm, GA was effectively employed to estimate the parameters of

Furthermore, not only the initial kinematic simulation model with six degrees of freedom for Faraman AM1 welding robot has been developed using CATIA V5 software, but also the forward and inverse kinematic equations also be obtained

by using conventional numerical methods The output results computed from simulation model were employed to compare with calculated results from kinematic equations to validate the simulation one

TABLE OF CONTENTS

Contents Page

ACKNOWLEDGEMENTS i

ABSTRACT ii

TABLE OF CONTENTS v

LIST OF FIGURES ix

LIST OF TABLES xiv

NOMENCLATURE xv

ACRONYMS xix

PUBLICATIONS xxi

CHAPTER 1: INTRODUCTION 1

1.1 MOTIVATION 1

1.2 SCOPE OF THE RESEARCH WORK 5

1.3 THESIS ORGANIZATION 6

CHAPTER 2: LITERATURE SURVEY 9

2.1 INTRODUCTION 9

2.2 FULL FACTORIAL DESIGN AND EMPIRICAL MODELS 10

2.2.1 Full factorial design 10

2.2.2 Linear model 11

2.2.3 Curvilinear model 12

2.2.4 Interaction model 12

2.3 DESIGN OF EXPERIMENTS AND OPTIMIZATION PROCEDURE

2.4 GOVERNING EQUATION FOR TRANSIENT HEAT TRANSFER

ANALYSIS 18

2.5 MOVING HEAT SOURCE MODELS 19

2.5.1 Models for moving heat source 19

2.5.2 Heat source and FEM models 23

2.6 CATIA SOFTWARE AND APPLICATIONS 28

2.7 ROBOTICS WELDING SYSTEMS 31

CHAPTER 3: PREDICTION ON TOP-BEAD WIDTH FOR THE BUTT WELD USING A GENETIC ALGORITHM 35

3.1 INTRODUCTION 35

3.2 EXPERIMENTAL WORK 36

3.3 DEVELOPMENT OF EMPIRICAL MODELS 39

3.3.1 Selection of mathematical model 39

3.3.2 Developed the mathematical model 40

3.3.2.1 Genetic algorithm 40

3.3.2.2 Development of empirical models 43

3.4 RESULTS AND DISCUSSION 47

3.4.1 The verification of the developed models 47

3.4.2 The effects of process parameters 50

3.5 CONCLUSSION 61

CHAPTER 4: PREDICTING BEAD GEOMETRY FOR LAB JOINT USING AN INTERACTION MODEL 62

4.1 INTRODUCTION 62

4.2 EXPERIMENTAL WORK 63

4.3 DEVELOPMENT OF EMPIRICAL MODELS 64

4.3.1 Bead width 65

4.3.2 Bead height 68

4.4 RESULTS AND DISCUSSION 70

4.4.1 ANOVA analysis for developed models 70

4.4.2 The accurate prediction of models 71

4.4.3 Effects of main process parameters 74

4.4.4 Effects of interaction for process parameters 76

4.5 CONCLUSSION 79

CHAPTER 5: DEVELOPMENT OF NUMERICAL MODEL 80

5.1 INTRODUCTION 80

5.2 DEVELOPMENT OF FEM MODEL USING A GA 80

5.3 VERIFICATION OF THE DEVELOPED GA-FEM MODEL 85

5.3.1 Verification of batch mode model 85

5.3.2 Verification of the GA-FEM model 88

5.4 WELDING CONDITIONS AND DIMENSIONAL DETAILS OF ADDITIONAL EXPERIMENTS 94

5.5 DEVELOPMENT OF FEM MODEL 97

5.5.1 Assumptions 97

5.5.2 Governing equation 98

5.5.3 Thermal physical properties of material 98

5.5.4 Mesh generation 99

5.6 RESULTS AND DISCUSSION 102

5.7 CONCLUSSION 113

CHAPTER 6: SIMULATION MODEL FOR ROBOTIC WELDING 114

6.1 INTRODUCTION 114

6.2 CATIA SOFTWARE 115

6.3 KINEMATICS 116

6.3.1 D-H (Denavit-Hartenberg) convention 116

6.3.2 Forward kinematics 120

6.3.3 Inverse kinematics 125

6.3.3.1 Solving for θ1 125

6.3.3.2 Solving for θ3 128

6.3.3.3 Solving for θ2 129

6.3.3.4 Solving for θ4 131

6.3.3.5 Solving for θ5 132

6.3.3.6 Solving for θ6 133

6.4 DEVOLOPEMENT OF SIMULATION MODEL 135

6.5 RESULTS AND DISCUSSION 139

6.6 CONCLUSION 142

CHAPTER 7: CONCLUSIONS 143

7.1 CONCLUSIONS 143

7.2 SUGGESTION FOR FUTURE WORK 145

REFERENCES 147

LIST OF FIGURES

2.1 Coordinate system used for the FEM analysis of disc model

according to Krutz and Segerlind [36] 21

2.2 Goldak’s double ellipsoidal heat source model [37] 22

2.3 Structure and interfaces of the system according to Hackel [89] 34

3.1 Input and output parameters of the GMA welding process 36

3.2 Configuration of butt welding specimen 38

3.3 Butt welding specimen 38

3.4 Measurement of top-bead width 39

3.5 Flow chart for the GA [25] 42

3.6 Comparison on the fitting of three developed models 48

3.7 The accurate prediction of three developed models 49

3.8 The effect of welding speed on top-bead width 52

3.9 The effect of welding voltage on top-bead width 52

3.10 The effect of arc current on top-bead width 53

3.11 The effect of tip gap on top-bead width 53

3.12 The effect of gas flow rate on top-bead width 54

3.13 Interaction effect of welding voltage and welding speed on top-bead width 55

3.14 Interaction effect of gas flow rate and welding speed on top-bead width 55 3.15 Interaction effect of welding speed and tip gap on top-bead width

3.16 Interaction effect of welding speed and arc current on top-bead

width 56

3.17 Interaction effect of tip gap and gas flow rate on top-bead width 57

3.18 Interaction effect of gas flow rate and welding voltage on top-bead width 57

3.19 Interaction effect of tip gap and welding voltage on top-bead width 58

3.20 Interaction effect of gas flow rate and arc current on top-bead width 58

3.21 Interaction effect of tip gap and arc current on top-bead width

59

3.22 Interaction effect of arc current and welding voltage on top-bead width 59

4.1 Schematic diagram for measurement of bead geometry 64

4.2 Comparison of measured and calculated results for bead width 71

4.3 Comparison of measured and calculated results for bead height

71

4.4 Predictable accuracy of the developed models for bead width

73

4.5 Predictable accuracy of the developed models for bead height

74

4.6 Effect of welding voltage on bead geometry 75

4.7 Effect of arc current on bead geometry 75

4.8 Effect of welding speed on bead geometry 76

4.9 Interaction effect of welding voltage and arc current on bead

width 77

4.10 Interaction effect of CTWD and welding angle on bead width

77

4.11 Interaction effect of CTWD and welding angle on bead height

78

5.1 Heat flux distributions at plane y = 0 using (5.1), (5.2), (5.4) equations 82

5.2 Heat flux distributions at plane y = 0 using (5.1), (5.2), (5.5), (5.6) equations 82

5.3 Flow chart for the GA-FEM model 84

5.4 Comparison with Goldak’s work 5.4(a) Goldak’s work [37] 86

5.4(b) Results of developed FEM model 86

5.5 Dependence of material properties on temperature [38] 5.5(a) Thermal conductivity factor table 87

5.5(b) Specific heat factor table 87

5.6 2D symmetric FE mesh model 88

5.7 Dimension details and weld bead sequence [51] 5.7(a) Dimension details of plates 89

5.7(b) Weld bead sequence 89

5.8 Temperature distributions from the developed model 92

5.9 Temperature distributions on the left plate [51] 92

5.10 Temperature distributions on the right plate [51] 93 5.11 Dimensional details and weld bead sequence

5.11(b) For the 2 pass fillet weld 96

5.12 Thermal physical properties of SM490A material 99

5.13 2D FE mesh generation 5.13(a) For the 5 pass butt weld 100

5.13(b) For the 2 pass fillet weld 100

5.14 Temperature fields and cross section for the 5 pass butt weld 5.14(a) Actual cross section of the 5 pass butt weld bead 104

5.14(b) Temperature fields at time t = 2.999s, first pass welding 104

5.14(c) Temperature fields at time t = 531.088s, second pass welding104 5.14(d) Temperature fields at time t = 1068s, third pass welding 104

5.14(e) Temperature fields at time t = 1479s, fourth pass welding 104

5.14(f) Temperature fields at time t = 1788s, fifth pass welding 104

5.15 Temperature fields and cross section for the 2 pass fillet weld 5.15(a) Actual cross section of the 2 pass fillet weld bead 105

5.15(b) Temperature fields at time t = 3.753s, first pass welding 105

5.15(c) Temperature fields at time t = 834.562s, second pass welding105 5.16 Temperature histories in the weld pools for the 5 pass butt weld 5.16(a) First pass (Peak temperature = 1711oC) 106

5.16(b) Second pass (Peak temperature = 1896oC) 106

5.16(c) Third pass (Peak temperature = 2053oC) 106

5.16(d) Fourth pass (Peak temperature = 1990oC) 107

5.16(e) Fifth pass (Peak temperature = 2078oC) 107

5.17 Temperature histories in the weld pools for the 2 pass fillet weld 5.17(a) First pass (Peak temperature = 2126oC) 108

5.17(b) Second pass (Peak temperature = 2044oC) 108

5.18 Temperature histories on the left side plate during 5 pass butt

weld at selected points 109 5.19 Temperature histories on the right side plate during 5 pass butt

weld at selected points 109 5.20 Temperature histories during 2 pass fillet weld at selected points

for the back surface 111 5.21 Temperature histories during 2 pass fillet weld at selected points

for the front surface 111 6.1 Internal welding carriage project 115

6.2 Schematic of the adjacent axes with the appropriately assigned

reference frames for determining the Denavit-Hartenburg parameters [107] 117 6.3 Faraman AM1 welding robot 121

6.4 Work envelope dimensions of Faraman AM1 welding robot.121

6.5 D-H convention joint-coordinate diagram 123

6.6 CATIA parts, assembly model and D-H convention

joint-coordinate diagram 136 6.7 CATIA simulation model of the robot 138

6.8 Simulation with Laws icon 138 6.9 Comparisons on computed distance results of simulation model

and forward kinematic equations 139 6.10 Comparison on angle values of simulation model and inverse

kinematic results

6.10(a) For Δθ1 and Δθ2 140

6.10(b) For Δθ3 and Δθ4 140

LIST OF TABLES

Table Page

3.1 Process parameters and values 37

3.2 GA parameters and values 42

3.3 Expressions were used to check the fitting 46

3.4 Variance test for developed empirical models 47

3.5 Welding condition for the additional experiments 48

3.6 ANOVA analysis using experimental data 51

4.1 Process parameters and values 64

4.2 ANOVA analysis for the model of bead width 67

4.3 ANOVA analysis for the model of bead height 69

4.4 Analysis of variance tests for developed models 70

4.5 Welding condition and the results for the additional experiments 73

5.1 Weld parameters during welding [51] 90

5.2 Thermal properties of 304 stainless steel materials [52] 90

5.3 Parameters of double ellipsoidal heat source 91

5.4 Comparison of peak temperature value 94

5.5 Weld parameters during welding of two types of weld 95

5.6 Parameters of double ellipsoidal heat source 102

5.7 Peak temperatures (oC) for the 5 pass butt weld at selected points 110

5.8 Peak temperatures (oC) for the 2 pass fillet weld at selected points 112

6.1 Kinematic parameters of Faraman AM1 robot 120

NOMENCLATURES

As Cross-sectional area of the fusion zone

C CTWD

NTotal Total number of all i-factor interaction terms

Q Heat generation rate per unit volume

Q(0) Maximum flux at the center of the heat source Q(r) Surface flux at radius r

Qf Power density distribution for front heat source

Qr Power density distribution for rear heat source

QWB Heat input due to body flux

QWS Heat input due to surface flux

b Semi-axis of ellipsoid in the y direction

bi Estimate regression coefficients of βi

c Semi-axis of ellipsoid in the z direction

cF Characteristic radius of heat flux distribution

cf Semi-axis of front ellipsoid in the z direction

cp Specific heat capacity

cr Semi-axis of rear ellipsoid in the z direction

d Diameter of welding wire

di Link offset

ff Fractional factors of the heat deposited in the front quadrant

fr Fractional factors of the heat deposited in the rear quadrant

h Combined heat transfer coefficient

k Thermal conductivity of material

q Heat flux vector

r Radial distance from the center of heat source

t Time

ts Load time step

u Number of terms (main and interaction)

(x, ξ) Moving coordinate system attached to the surface heat source (x, y, ξ) Moving coordinate system attached to the volume heat source (x, y, z) Fixed coordinate system attached to the volume heat source

ε Emissivity of the surface of the body

εi Experimental error εi is unobservable and represents all other

factors affecting Y which are not inX1, X2,…, Xm

ANOVA Analysis of Variance

API Application Programming Interface

BPNN Back-Propagation Neural Network

CAA Component Application Architecture

CAD Computer Aided Design

CAE Computer Aided Engineering

CAM Computer Aided Manufacturing

CATIA Computer Aided Three Dimensional Interactive Application

GAWSEM Genetic Algorithm Welding Strength Estimation Model GAWVEM Genetic Algorithm Welding Velocity Estimation Model GLM General Linear Model

GMA Gas Metal Arc

GSM Generative Shape Modeling

GTA Gas Tungsten Arc

GUI Graphical User Interface

HAZ Heat Affected Zone

MOKA Methodology and Software Tools Oriented To Knowledge

Based Engineering Applications

MSE Mean Square Error

MST Mean Square Total

PLM Product Lifecycle Management

SAON Self-Adaptive Offset Network

SSE Sum of Square Error

SST Sum of Square Total

SWERS Ship Welding Robot System

PUBLICATIONS

International Journals

1 D.T Thao and I.S Kim An Evaluation Approach for Prediction of

Process Parameters with Genetic Algorithm Material Science Forum,

Vols 580-582 (2008), pp 375-378

2 D.T Thao, J.W Jeong, I.S Kim and J.W.H.J Kim Predicting Lab-Joint

Bead Geometry in GMA Welding Process Archives of Materials Science

and Engineering, Vol 32, Issue 2 (2008), pp 121-124

3 D.T Thao, I.S Kim, H.H Kim, J.W Jeong and B.Y Kang Developed

Simulation Model-Kinematics for Robotic Arc Welding Asian

International Journal of Science and Technology in Production and Manufacturing, Vol 1, No 2 (2008), pp 69-76

International Conference and Others

1 D.T Thao, I.S Kim, J.S Son and J.H Seo Developed Multiple Linear Regression Model using Genetic Algorithm for Predicting Top-Bead

Width in GMA Welding Process Proceedings of the 2006 Autumn Annual

Meeting of Korean Welding Society (KWS 2006-Autumn), 19th-20th,

October, Pusan, Korea, 2006

2 김일수, D.T Thao, 문채주, 한양주 CIM 기반 소수력발전용

추계학술대회, 12 월 8 일, 2006

3 D.T Thao and I.S Kim An Evaluation Approach for Prediction of

Process Parameters with Genetic Algorithm International Welding

Joining Conference-Korea (IWJC), 10th-12th, May, Seoul, Korea, 2007

4 D.T Thao, I.S Kim, I.J Kim, B.Y Kang, K.C Chang and D.G Lee Development of Simulation Model for Robotic Welding using Catia

Program International Symposium on Mechatronics and Automatic

Control-Vietnam (ISMA), 25th-26th, October, Ho Chi Minh city, Vietnam,

2007

5 D.T Thao, I.S Kim, H.H Kim and S.W Son A Study about Development of Welding Heat Input Model using Genetic Algorithm

Proceedings of the 2008 Spring Annual Meeting of Korean Welding and Joining Society (KWJS 2008-Spring), 8th-9th, May, Daejon, Korea, 2008

6 D.T Thao, J.W Jeong, I.S Kim and J.W.H.J Kim Predicting Lab-Joint

Bead Geometry in GMA Welding Process The 16th International

Scientific Conference on Achievements in Mechanical and Materials Engineering-Poland (AMME), 22nd-25th, June, Gliwice, Poland, 2008

7 D.T Thao, I.S Kim, H.H Kim, J.W Jeong and B.Y Kang A Study on

Simulation Model for Robotic Arc Welding The 9th Global Congress on

Manufacturing & Management-Australia (GCMM), 12th-14th, November,

Gold Coast, Australia, 2008

by melting and solidification of the joint edges together with filler metal transferred from the electrode A flow of inert gas has been shielded the weld metal from the surrounding atmosphere The sequence of heating and cooling cycle during welding process cause the nonuniform expansion and contraction of the weld and surrounding base material results in distortion in a welded structure [40] There are two ways that are used for reducing distortion of welding structures The first way is to employ a work clamp and motion devices to hold workpieces accurately in position Another method is to conduct some experiments to find out the optimal welding sequence [99] Unfortunately, two techniques are time consuming and have a high cost Effective uses of simulation tools can help to predict the distortions and optimize the welding sequence at an early stage before the real work begin In addition, bead geometry in the GMA welding process as a welding quality is an important factor in determining the

mechanical characteristics of the weld [1] which greatly effects on quality of welding structures

To achieve the high quality and welding performance, an interrelationship between bead geometry and welding parameters requires to be developed Research on finding out these relationships is not novel Many efforts have been done to develop the analytical and numerical models to study these relationships [1-13] Kim et al [1] proposed a method for determining the near-optimal settings

of welding process parameters to obtain the desired weld bead geometry in GMA welding using a CRS algorithm which is similar to the GA Raveendra et al [2] and Yang et al [3] employed multiple regression techniques to establish the empirical models for various arc welding processes Datta et al [4] developed three empirical models for predicting bead volume of submerged arc butt welding Also, Gunaraj et al [5] proposed empirical models for prediction and optimization of weld bead for the SAW process Furthermore, Gunaraj et al [6] highlighted the use of RSM by designing a central composite rotatable design matrix to develop empirical models for predicting weld bead quality in SAW for pipelines

Recently, some researches have been attempted to use AI techniques to model the relationships between process parameters and bead geometry [7-11] Kim et al [7] developed an intelligent system for GMA welding process based on factorial experimental Li et al [8] studied the non-linear relationship between the geometric variables and welding parameters for SAW process using the SAON Tarng et al [9] constructed the relationship between process parameters and the

developed a system for prediction of process parameters based on back-bead geometry in GMA welding using multiple regression analysis and artificial neural network Nagesh et al [11] employed BPNNs for prediction of bead geometry and penetration in SMA welding

GA, a global optimization algorithm was suggested to use for solving the optimization problems especially in various arc welding process [12-13] Correia

et al [12] utilized GA to decide the near-optimal setting of three process parameters in a GMA welding process Correia et al [13] also tried to compare a

GA and RSM in the optimization of a GMA welding process application

Despite the large numbers of attempts to analyze arc welding process, an empirical model which is a combination of linear and curvilinear model to study interrelationships between input and output parameters in the arc welding process are still lacking

The Goldak’s double ellipsoidal heat source is now widely used in development

of a FEM model for transient heat transfer analysis, predicting on deformation and residual stresses [38, 40-46] Gery et al [38] investigated the effects of the heat source distribution, energy input and welding speed on temperature variations using FEM transient heat transfer analysis Deng et al [40] developed a 3D FEM to simulate the welding temperature field, residual stress and welding distortion Also, Deng [41] developed a sequentially coupled thermal, metallurgical, mechanical 3D FEM to investigate the effects of solid-state phase transformation on welding residual stress and distortion In addition, Deng et al [42] developed 3D and 2D FEM to analyze the temperature fields and the residual

stress distributions Malik et al [43] presented a sequentially coupled, full 3D FEM model for the prediction of temperature distributions and the subsequent residual stress Duranton et al [44] developed the 3D FEM simulation for the multi-pass welding Sattari-Far et al [45] presented a parametric study to determine the effect of welding sequence on welding distortions in pipe-pipe joints Dong et al [46] analyzed 3D FEM using commercial FE codes MSC.Marc Element birth and death finite element technique was employed to control the process of filling metal during multi-pass welding process

The expected parameters of this heat source have been adjusted to create a desired melted zone according to the welding conditions [40] Really, it is not easy to choose accurate parameters having computed temperature field fit the experimental temperature data On the one hand this distribution function can be criticized as “fudge” factors [34]

Some efforts [81-83] have been carried out to develop various simulation models

of robotic welding systems Carvalho et al [81] described some features of graphic simulation and OLP focusing on its applications for robotic welding Cheng [82] introduced a methodology for conducting the robotic workcell simulation models by using Deneb IGRIP technology Ericsson et al [83] developed a system that combines OLP software with a FEM model to optimize the robot trajectories and welding process parameters

The preliminary stage for the development of robot simulation model is first to build the accuracy kinematic simulation and derive the correct forward and inverse kinematic equations of the robot

There are not studies on empirical models using a GA or an interaction model for predicting on bead geometry; a model for automatic finding the parameters of a moving heat source; and a kinematic simulation model for Faraman AM1 welding robot Therefore, the objectives of this thesis are: (i) Developing an empirical model using a GA and an interaction model for predicting the process parameters to achieve the desired bead geometry and to investigate the main and interaction effects of process parameters on bead geometry; (ii) Proposing a new algorithm for automatic finding the parameters of a moving heat source on multi-pass butt and fillet welded joint type with dissimilar thickness with some constraints and optimization; (iii) Integrating a kinematic simulation model for Faraman AM1 welding robot with six degrees of freedom

1.2 SCOPE OF THE RESEARCH WORK

The research work concentrates on the following aspects:

1 Developing empirical models to predict bead geometry for butt and lab joint in GMA welding process Four empirical models have been developed: linear, curvilinear, interaction, and a proposed model that was called intelligent model for studying the effects of process parameters on bead geometry Not only the fitting of these models has been checked and compared by using variance test, but also the prediction on bead geometry using the developed models has been carried out based on the additional experiments

2 Proposing a new algorithm for the automatic finding the parameters of double ellipsoidal heat source based on a GA The algorithm includes a

GA program using GA toolbox for MATLAB, and a batch mode thermal model using ANSYS software Based on this algorithm, developing FEM models to calculate the transient thermal histories in multi-pass GMA welds including the butt and fillet weld type with dissimilar thickness The developed model will be verified by comparison with Goldak’s work and with molten zone section experimental data

3 Integrating the initial kinematic simulation model with six degrees of freedom for Faraman AM1 welding robot The correction of the simulation

as well as the correction for kinematic equations of the robot must be verified

Chapter 2 pays attention to a comprehensive review of covering the fundamental

of full factorial, empirical models, and the fundamental theory of some moving heat source A brief discussion on recent published papers for the DOE,

transient thermal elastic plastic analysis has also been presented Furthermore, the developed robotic welding systems by different researchers are also included

Chapter 3 represents a proposed empirical model for the prediction of process parameters on top-bead width for butt weld type The reliable fitting and the predicting capabilities of proposed model in comparison with linear and curvilinear model has also been provided In addition, the effects of welding parameters on top-bead width are also covered

Chapter 4 focuses on investigation of the development of the interaction model for prediction of bead geometry for lab joint The GLM was used to determine the most important factors and interaction terms for achieving the interaction models The verification on the fitting and predictive capabilities given by interaction models and curvilinear models were also carried out The interaction models were used to graphically show the main and interaction effect of process parameters on bead geometry

Chapter 5 represents on development of a new algorithm for the automatic finding parameters of a moving heat source Not only the development of the algorithm that is combination of GA and FEM was paid, but also the developed models was verified based on Goldak’s work and additional experiments

Chapter 6 provides sufficient information on kinematic modelling and simulation

of Faraman AM1 arc welding robot with six degrees of freedom by using CATIA

V5 The kinematic equations with respect to various degrees of freedom

involving complex dynamics and the accuracy of simulation model have been checked

Chapter 7 covers on the concluding remarks and the suggestion for future work

at an early design stage of a welding process

In this chapter, a comprehensive review of covering the fundamental of full factorial, empirical models, and the fundamental theory of some moving welding heat source is given A brief discussion on recent published papers for the DOE, optimization procedure, moving heat source models for performing transient thermal elastic plastic analysis, and the simulation models for developing robotic welding system also are included

2.2 FULL FACTORIAL DESIGN AND EMPIRICAL MODELS

2.2.1 Full factorial design

In an experiment with emphasis on a GMA welding experiment, the factors are deliberately changed in order to observe the changes on the response under the effects of one or more factors The DOE is an efficient procedure for planning experiments to obtain experimental data that can be used for comparison, characterization, modeling and optimization

In DOE, each factor can take on a certain number of values that are called the levels of a factor Each factor and level combination is sometimes called a cell Generally, the balance and orthogonality properties are maintained in the designed experiments Balanced designs are those in which the cells have an equal number of runs Two vectors of the same length are orthogonal if the sum

of the products of their corresponding elements is equal to 0 The orthogonal property is important because it eliminates correlation between the estimates of the main effects and interactions [24] The entire design to make easy for data analysis is running more than once that is called replication

The popular experiments are known as full factorial design The most important advantages of full factorial design are that the main and interaction effects of

factors on one or more response can be known If an experiment has m factors, each factor has p levels, a full factorial design has p m runs When all factors have

been coded so that the high value is "1" and the low value is "-1", the design

orthogonal and all the columns sum to 0 Orthogonality is a very desirable

property in DOE [24]

2.2.2 Linear model

The linear model creates a relationship in the form of a straight line (linear) that

best approximates all the individual data points The multiple linear models are a

generalization of the linear model that allows for more than one independent

variable (factor)

A design matrix with n runs of experimental data is given as following;

)X, ,X,X,Y(), ,X, ,

1 1

1 1

0

i

^

Xb

X

b

b

^ n

To estimate the coefficients c0, c1,…, cm of curvilinear model, logarithms are

taken for both side of Eq (2.5) Hence;

m m 1

A full interaction model contains all the main factors and all orders of interaction

terms For example, a full interaction model with 3 factors can be written;

3 2 1 123 3 1 13 3 2 23 2 1 12 3 3 2 2 1

1

k

The advantages of full interaction model are that we can consider and uniquely

estimate interactions of all factors However, the disadvantage is that the total

number of possible i-factor interactions quickly increases as the number of factors

increases For 3 factors, the number of 2-factor interactions is 3, while for 7 factors

the number of 2-factor interactions is 21 The total number of possible i-factor

interactions is calculated by using the following equation;

2

m 1

m 0

m

The GLM is effectively employed to determine the most important main and

interaction terms in order to have interaction models with a little term but have

best fit on the experimental data

2.3 DESIGN OF EXPERIMENTS AND OPTIMIZATION PROCEDURE

To get the desired weld quality in robotic GMA welding process, it is essential to

know interrelationships between process parameters and bead geometry as a

welding quality Many efforts have been done to develop the analytical and

numerical models to study these relationships, but it was not an easy task because

there were some unknown, nonlinear process parameters [1] For this reason, it is

good for solving this problem by the experimental models One of the experimental models was a multiple regression technique that was utilized to establish the empirical models for various arc welding processes [2, 3]

Datta et al [4] developed a statistical model for predicting bead volume of submerged arc butt welds in mild steel plates Experiments based on a 33 full factorial design, without replication, were conducted with 3 levels of 3 process parameters namely welding current, welding voltage, and electrode extension The ANOVA was employed to evaluate quantitatively the significant of the main and interaction effects of 3 process parameters on bead volume Three empirical models: linear, curvilinear, and a second degree response surface model have been developed The effects of 3 process parameters were also represented graphically and it is shown that these process parameters are to represent significant effects on bead volume

Also, Gunaraj et al [5] developed empirical models using the five-level factorial design for prediction and optimization of weld bead for the SAW process of 6-mm-thick structural steel plates The second degree response surface models were developed to have relationships between the important control process parameters: welding voltage; wire feed rate; welding speed; and nozzle-to-plate distance; and 5 bead-quality parameters: penetration; reinforcement; bead width; total volume of the weld bead; and dilution ANOVA analysis was used to check the adequacy of all the empirical models The main and interaction effects of the process parameters on bead geometry were determined quantitatively and presented graphically

Furthermore, Gunaraj et al [6] employed an application of RSM for predicting weld bead quality in SAW for pipes The experiment was designed based on a four factor five level factorial central composite rotatable design The second degree response surface models, which relate the important process parameters such as the open-circuit voltage, the wire feed rate, the welding speed and the nozzle-to-plate distance, to the penetration, the reinforcement, the width and the percentage dilution of the bead geometry, were developed ANOVA analysis was effectively used to test the adequacy of the all the developed models

Recently, some researches have been concentrated on using these traditional models with AI techniques to solve the problem [7-11] Kim et al [7] developed a linear, a curvilinear model and an intelligent system for GMA welding process based on factorial experimental design Three process parameters namely arc current, welding voltage and welding speed, are considered as the main parameters influencing bead geometry and each process parameter take on three levels ANOVA was employed to determine the significance of each factor on the optimization parameter and to detect whether there were any interaction effects among the factors themselves The intelligent system was established by using a BPNN

Li et al [8] studied the non-linear relationship between the geometric variables namely height, width, penetration, fused and deposited areas of a bead and process parameters such as arc current, welding voltage and welding speed of SAW process using a SAON Tarng et al [9] investigated the relationship between process parameters and the features of the bead geometry for TIG welding process using a BPNN An SA optimization algorithmwas successfully

applied to network for obtaining the process parameters with optimal bead geometry based on an objective function

Lee et al [10] developed a system for the prediction of welding process parameters in GMA welding, where a gap exists using multiple regression analysis and artificial neural network According to Lee, the specimen of SS41 mild steel with 180mm (width) x 100mm (length) x 6mm (thickness) was used to perform 96 experiments under the welding conditions of 4 process parameters namely the grove gap, welding voltage, arc current, welding speed The width and depth of the back-bead geometry were established as dependent parameters and obtained through a laser vision sensor SPSS software was employed to make regression analysis of 2 linear models, one for the width and another for the depth

of back-bead An error back-propagation algorithm of the neural network was developed in which the four welding parameters were established as input variables, and the estimated width and depth of the back-bead geometry were established as output variables

Nagesh et al [11] employed BPNNs for prediction of bead geometry in SMA welding based on 18 experiment welding runs BPNN, in which electrode feed rate; arc power; welding voltage; arc current; arc length; and arc travel rate that were established as input variables, were used for the modeling the bead height, bead width, depth of penetration, and area of penetration

One of the stochastic search methods, GA was a powerful tool to solve the optimization problems especially in the welding area Correia et al [12] utilized

welding voltage, wire feed rate and welding speed in a GMA welding process The search was based on the minimization of an objective function, which takes into account the deposition efficiency and the geometric characteristics such as penetration, width and reinforcement of the weld bead According to Correia, the search for the near-optimal process parameters was carried out step by step, in each step, the GA generating the next experiment based on the previous without the knowledge on the modeling equations between the inputs and outputs of the GMA welding process In addition, Correia et al [13] also focused on comparison between GA and RSM in determination of the optimal process parameters in GMA welding process In the case of the RSM optimization, an experimental design, being a CCD composed of a full factorial design 23 in a total of 20 runs was conducted and tests were carried out to generate the proper models

Kim et al [1] proposed a method for determining the near-optimal settings of welding process parameters namely wire feed rate, welding voltage, and welding speed to obtain the desired bead geometry in GMA welding using a CRS algorithm The search range of each welding parameter was determined based on

a previous work and authors’ experience With each welding condition determined through CRS algorithm, the welding was carried out and the front-bead height, back-bead width, and penetration were measured The objective function was formulated based on the desired and measured bead geometry

Gunaraj [14] minimized the total volume of the bead geometry subject to minimum reinforcement, minimum bead width, minimum dilution and maximum penetration for the SAW process using optimization module in MATLAB software A quasi-Newton method was used to determine the optimum the total