submerged arc welding neural network

42
LIST OF TABLES Table 3.1 Composition of Flux Table 3.2 Responses Table 4.1 ANOVA for Surface Mean Model for Vickers Hardness Table 4.2 Final equation in terms of Coded Factors for Vickers Hardness Table 4.3 Final equation in terms of Actual Factors for Vickers Hardness Table 4.4 ANOVA for Surface Mean Model for Impact Strength Table 4.5 Final equation in terms of Coded Factors for Impact Strength Table 4.6 Final equation in terms of Actual Factors for Impact Strength

Upload: johnnydorian

Post on 30-Jan-2016

252 views

Category:

Documents


0 download

DESCRIPTION

Submerged Arc Welding and neural network

TRANSCRIPT

Page 1: Submerged Arc Welding NEURAL NETWORK

LIST OF TABLES

Table 3.1 Composition of Flux

Table 3.2 Responses

Table 4.1 ANOVA for Surface Mean Model for Vickers Hardness

Table 4.2 Final equation in terms of Coded Factors for Vickers Hardness

Table 4.3 Final equation in terms of Actual Factors for Vickers Hardness

Table 4.4 ANOVA for Surface Mean Model for Impact Strength

Table 4.5 Final equation in terms of Coded Factors for Impact Strength

Table 4.6 Final equation in terms of Actual Factors for Impact Strength

Page 2: Submerged Arc Welding NEURAL NETWORK

LIST OF FIGURES

Fig 1.1 Submerged Arc Welding Process

Fig 1.2 SAW Machine

Fig 1.3 Basic ANN Structure

Fig 1.4 Network Architecture

Fig 4.1 Regression Plot

Fig 4.2 Performance Plot

Fig 4.3 Training State

Fig 4.4 Neural network Training

Fig 4.5 Function Fitting Neural Network

Fig 4.6 Weight and Biases

Fig 4.7 Design Layout

Fig 4.8 Summary of Design Layout

Fig 4.9 Graph Columns

Fig 4.10 Fit Summary

Fig 4.11 FDS Graph

Fig 4.12 One Factor Graph

Fig 4.13 Perturbation

Fig 4.14 Interaction

Fig 4.15 Contour

Fig 4.16 3D Surface

Fig 4.17 Normal plot-Vickers Hardness

Fig 4.18 Predicted vs actual

Page 3: Submerged Arc Welding NEURAL NETWORK

ABSTRACT

Neural Network Modelling of Submerged Arc Welding of Mild Steel

In any fabrication industry, Submerged Arc Welding (SAW) is used as a heavy

metal deposition rate welding process. The process is characterized by the use

of granular flux blanket that covers the molten weld pool during operation.

Welding input parameters play a very significant role in determining the quality

of weld joint. The joint quality can be defined in terms of properties such as

weld bead geometry, mechanical properties and distortion. Generally all

welding processes are used with the aim of obtaining a weld joint with the

desired weld bead parameters, excellent mechanical properties with minimum

distortion.

Artificial neural network (ANN) is widely established in the artificial

intelligence (AI) research where a nonlinear mapping between input and output

parameters is required for a function approximation. The backpropogation

algorithm is used in layered feed forward ANN’s. The backpropagation neural

network (BPNN) has three layers: input layer, hidden layer, and output layer.

Page 4: Submerged Arc Welding NEURAL NETWORK

TABLE OF CONTENTS

CHAPTER 1 INTRODUCTION

1.1 Submerged Arc Welding

1.2 Advantages of SAW

1.3 Limitations of SAW

1.4 SAW Fluxes

1.5 Effect of flux composition on SAW

1.6 Artificial Neural Network

1.7 Taguchi

CHAPTER 2 LITERATURE SURVEY

CHAPTER 3 METHODOLOGY

CHAPTER 4 RESULTS AND ANALYSIS

CHAPTER 5 CONCLUSION

CHAPTER 6 REFERENCES

Page 5: Submerged Arc Welding NEURAL NETWORK

CHAPTER 1

INTRODUCTION

1.1 THE SUBMERGED ARC WELDING PROCESS

Submerged Arc Welding (SAW) is a method in which the heat required to fuse

the metal is generated by an arc formed by an electric current passing between

the electrode and workpiece.

A layer of granulated mineral material known as submerged arc welding flux

covers the tip of welding wire, the arc and the workpiece. There is no visible arc

and no sparks, spatter or fume. The electrode may be a solid or cored wire or a

strip.

SAW is normally a mechanised process. The welding current, arc voltage and

level speed all affect the bead shape, depth of penetration and chemical

composition of the deposited weld metal. Since the operator cannot observe

weld pool, great reliance is placed on parameter setting and positioning of the

electrode.

Page 6: Submerged Arc Welding NEURAL NETWORK

General Scope:

Current: 100-3600 A

Wires in one molten pool: from 1 to 6

Voltage: 20-50 V

Speed: 300-3500 mm/min

Deposition rate: 2-100 kg/hr

Page 7: Submerged Arc Welding NEURAL NETWORK

When the apparatus is set into operation, several things occur in the following

sequence:

The submerged arc welding flux feeds through the hopper tube and

continuously distributes itself over the seem a short distance ahead of the

welding zone

Wire feed mechanism begin to feed the welding wire into the joint data

at a controlled rate

Electric arc is established as the current flows between the electrode and

the work

The carriage is started (manually or automatically) to travel around the

seam

The tremendous heat evolved by the passage of the electric current

through the welding zone melts the end of the wire and the adjacent edges

of the workpiece, creating a pool of molten metal

The submerged arc welding flux completely shields the welding zone

from contact with the atmosphere

As the welding zone moves along the joint, the fused submerged arc

welding flux cools and hardens into a brittle, glass-like material which

protects the weld until cool, then usually detaches itself completely from

the weld.

Page 8: Submerged Arc Welding NEURAL NETWORK

1.2 ADVANTAGES

High quality

Little risk of undercut and porosity

No spatter

Very little risk of lack of fusion due to deep and safe penetration

High deposition rate

High thermal efficiency

No radiation

No need for fuel extraction

1.3 LIMITATIONS

Precise joint preparation required

No observation of arc and process during welding possible

High operational effort

The high welding speeds and deposition rates which are characteristics of

submerged arc welding require automatic control of the motor that feeds the

welding wire into the weld into the weld. No manual welder could smoothly

comparable to those of a submerged arc welding parameters. The automatic

Page 9: Submerged Arc Welding NEURAL NETWORK

control and power supply system used in submerged arc welding operates to

maintain a constant voltage and current.

1.4 SAW FLUXES

The main task of SAW fluxes is to protect the arc, the molten pool and the

solidifing weld metal from the atmosphere. Moreover fluxes have the following

tasks:

creation of ions to increase arc conductivity

Arc stabilizing

Creation of a slag which forms a cavity

Influence bead shape and surface finish

Deoxidation of molten pool

Alloying of weld metal

Influence the weld cooling rate

Page 10: Submerged Arc Welding NEURAL NETWORK

1.5 EFFECT OF FLUX COMPOSITION ON SUBMERGED

ARC WELDING

Submerged arc welding is widely used in the fabrication of pressure vessels,

pipe lines and offshore structures because of its higher metal deposition rate,

good strength of the joint and good surface appearance. The properties of the

welded joint such as strength, toughness can be improved by controlling the

microstructure of the welded joint. The element transfer from the flux has major

influence on weld metal composition and weld metal properties. To predict

weld metal properties, it is necessary to determine the weld composition, which

primarily depends upon wire, flux, parent metal, slag metal reactions, process

parameters, dilution and electrochemical reactions. Numerous investigators

have attempted to determine, which flux components are of most importance in

establishing the final weld chemistry. The weld chemistry is decided by the

metallurgical reactions in SAW but to decide the extent of metallurgical

reaction in saw is very difficult because of large variations in cycle temperature,

reaction time, high heat input. In SAW due to short reaction time during SAW

the reaction is not reached to its thermodynamic equilibrium, so the exact

prediction of weld metal chemistry is difficult. The purpose of this literature

review is to focus on an innovative approach which is needed while deciding

weld chemistry. It would be worthwhile if one could develop a frame work to

predict the Mn, Si, carbon, oxygen and other elements in the final weld

Page 11: Submerged Arc Welding NEURAL NETWORK

metal, from a given combination of electrode, flux and base metal. The work

done so far on Element transfer study is very limited. Much published

information is not available about fluxes made by Industry professionals as they

do not disclose the composition of the flux for which they claim higher

strength and better mechanical properties.

Page 12: Submerged Arc Welding NEURAL NETWORK

1.6 Artificial Neural Network (ANN)

An ANN is composed of simple elements operating in parallel. These elements

are inspired by biological nervous systems. As in nature, the connections

between elements largely determine the network function. You can train a

neural network to perform a particular function by adjusting the values of the

connections (weights) between elements.

Typically, neural networks are adjusted, or trained, so that a particular input

leads to a specific target output. The network is adjusted, based on a

comparison of the output and the target, until the network output matches the

target. Typically, many such input/target pairs are needed to train a network.

Page 13: Submerged Arc Welding NEURAL NETWORK

The Backpropagation Algorithm

The backpropogation algorithm is used in layered feed forward ANNs. This

means that the artificial neurons are organized in layers, and send their signals

forward, and then the errors are propagated backwards. The network receives

inputs by neurons in the input layer and the output of the network is given by

neurons in the output layer. There may be one or more intermediate hidden

layers. The backpropagation algorithm uses supervised learning which means

that we provide the algorithm with examples of the inputs and outputs we want

the network to compute, and then the error (difference between actual and

expected results) is calculated. The idea of backpropagation algorithm is to

reduce the error, until the ANN learns the training data. The training begins

with random weights, and the goal is to adjust them so that the error will be

minimal.

Advantages of ANN

Massive parallelism

Distributed representation and computation

Learning ability

Generalization ability

Adaptivity

Fault tolerance

Low energy consumption

Page 14: Submerged Arc Welding NEURAL NETWORK

Problems of interest

Pattern Classification

Clustering/Categorization

Function approximation

Prediction/Forecasting

Optimization

Content-addressable memory

Control

Page 15: Submerged Arc Welding NEURAL NETWORK

1.7 TAGUCHI

Taguchi experimental design methods are very complicated and difficult to use.

Additionally, these methods require a large number of experiments when the

number of process parameters increases. In order to minimize the number of test

required. Taguchi experimental design method, a powerful tool for designing

high quality system, was developed by Taguchi. This method uses a special

design of orthogonal arrays to study the entire parameter space with small

number of experiments only. Experiments were designed using Taguchi method

so that effect of all the parameters could be studied with minimum possible

number of experiments .Using Taguchi method, Appropriate Orthogonal Array

has been chosen and experiments have been performed as per the set of

experiments designed in the orthogonal array. Signal to Noise ratios are also

calculated to analyse the effect of parameters more accurately

Taguchi recommends analyzing the mean response for each run in the inner

array, and he also suggest analyzing variation using an appropriately chosen

signal to noise ratio (S/N)

Regardless of category of the performance characteristics, the lower S/N ratio

corresponds to a better performance .Therefore ,the optimal level of the process

parameters in the level with the lowest S/N value. The statistical analysis of the

data was performed by analysis of variance (ANOVA) to study the contribution

Page 16: Submerged Arc Welding NEURAL NETWORK

of the factor and interactions and to explore the effects of each process on the

observed value.

Results of the experiments were analyzed analytically as well as graphically

using ANOVA. ANOVA has determined the percentage contribution of all

factors upon each response factor individually

Page 17: Submerged Arc Welding NEURAL NETWORK

CHAPTER 2

LITERATURE SURVEY

A research article on “Effect of Flux Composition on Element Transfer during

Submerged Arc Welding” was published by Mr. Brijpal Singh, Mr. Z.A. Khan

and Mr. A.N.Siddiquee in the International Journal of Current Research.

A research article on “Optimization of submerged arc welding process

parameters in hardfacing” was published by Dr. H. L. Tsai, Y. S. Tarng, C. M.

Tseng in the International Journal of Advanced Manufacturing Technology,

1996, Volume 12, Issue 6

Davis and Bailey (1978, 1980) reveled in their research that element transfer in

SAW depends not only on its concentration in the flux but also depends upon

the other substances, which are present in the flux e.g. Si transfer from the flux

depends upon Al, Ti, Zr, which act as a network former and replace Si in a

network leaving Si free to transfer. Palm (1972) studied the effect of flux

composition in element transfer study and found that various elements like Ca,

Mg, Al,etc. do not directly affect the weld composition. Kubli and Sharav

(1961) found that less oxygen was transferred to weld while reducing the

amount of SiO2 and this again was confirmed by Tulianiet al. (1969), who

showed that oxygen content in the weld metal decreases as the BI is increased

Page 18: Submerged Arc Welding NEURAL NETWORK

CHAPTER 3

METHODOLOGY

The objective is to analyse the effect of different flux compositions on the

mechanical properties of the weld joint. The weights of NiO, MnO and MgO

were varied while the rest of the composition was kept constant. The

composition of the flux is shown in the table:

Type of flux Al2O3(70%) SiO2(70%) CaO(70%) CaCO3(70%) NiO(12.5%) MnO(12.5%) MgO(12.5%) CaF2(7.5%)

111 297 560 543 970 60 50 85 150

122 297 560 543 970 60 50 95 150

133 297 560 543 970 60 50 105 150

212 297 560 543 970 80 70 95 150

223 297 560 543 970 80 70 105 150

231 297 560 543 970 80 70 85 150

313 297 560 543 970 100 90 105 150

321 297 560 543 970 100 90 85 150

332 297 560 543 970 100 90 95 150

Response Factors

The mechanical properties considered were Vickers Hardness and Impact

strength.

Impact Strength

The impact strength describes the ability of a material to absorb shock and

impact energy without breaking. The impact strength is calculated as the ratio of

impact absorption to test specimen cross-section. Toughness is dependent upon

temperature and the shape of the test specimen.

Page 19: Submerged Arc Welding NEURAL NETWORK

Vickers Hardness

Hardness is defined as the resistance to indentation, and it is determined by

measuring the permanent depth of the indentation. The Vickers hardness test

method, also referred to as a microhardness test method, is mostly used for

small parts, thin sections, or case depth work. The Vickers method is based on

an optical measurement system.

S.NO DESIGN MATRIX A FOR

HARDNESS

SIO2 32V

HRB S/N

1 1 1 1 35 30.88

2 1 2 2 38.33 31.67

3 1 3 3 48 33.62

4 2 1 2 45 33.06

5 2 2 3 50 33.97

6 2 3 1 50 33.97

7 3 1 3 47.33 33.50

8 3 2 1 40 32.04

9 3 3 2 53.66 34.59

The inputs and responses found out in the experiments are used for NEURAL

NETWORK MODELLING using MATLAB software.

DESIGN EXPERT was used to evaluate, analyze and optimize the data using

Taguchi Orthoganl Array.

S.NO DESIGN MATRIX FOR IMPACT

STRENGTH

SIO2 32V

Impact

Strength

S/N

1 1 1 1 52.75 34.44

2 1 2 2 54 34.64

3 1 3 3 73.25 37.29

4 2 1 2 80 38.06

5 2 2 3 62.5 35.91

6 2 3 1 51 34.15

7 3 1 3 65.66 36.34

8 3 2 1 30.75 29.75

9 3 3 2 64 36.12

Page 20: Submerged Arc Welding NEURAL NETWORK

CHAPTER 4

RESULTS AND ANALYSIS

Responses found out using ANN in MATLAB:

Fig. 4.1. Regression Plot

Page 21: Submerged Arc Welding NEURAL NETWORK

Fig 4.2 Performance Plot

Fig 4.3 Training State

Page 22: Submerged Arc Welding NEURAL NETWORK

MATLAB CODE

function net = create_fit_net(inputs,targets);

numHiddenNeurons = 20;

net = newfit(inputs,targets,numHiddenNeurons);

net.divideParam.trainRatio = 70/100;

net.divideParam.valRatio = 15/100;

net.divideParam.testRatio = 15/100;

[net,tr] = train(net,inputs,targets);

outputs = sim(net,inputs);

plotperf(tr)

plotfit(net,inputs,targets)

plotregression(targets,outputs)

>>inputs = [ 60 50 85; 60 50 95; 60 50 105; 80 70 95; 80 70 105; 80 70 85; 100

90 105; 100 90 85; 100 90 95];

>>targets = [ 35 52.75; 38.33 54; 48 73.25; 45 80; 50 62.5; 50 51; 47.33 65.66;

40 30.75; 53.66 64];

>>nftool

Page 23: Submerged Arc Welding NEURAL NETWORK

>>adscript

performance = 5.8057e-23

trainPerformance = 5.8057e-23

valPerformance = NaN

testPerformance = NaN

>>adscript

Error using network/train (line 272)

Inputs and targets have different numbers of samples.

Error in adscript (line 46)

[net,tr] = train(net,inputs,targets);

>>adscript

performance = 103.4625

trainPerformance = 9.6340e-24

valPerformance = 176.9567

testPerformance = 754.2061

>>getwb(net)

>>size(getwb(net))

Page 24: Submerged Arc Welding NEURAL NETWORK

Fig 4.4 Neural Network Training

Fig 4.5 Function Fitting Neural Network

Page 25: Submerged Arc Welding NEURAL NETWORK

Fig 4.6 Weights and Biases

Page 26: Submerged Arc Welding NEURAL NETWORK
Page 27: Submerged Arc Welding NEURAL NETWORK

DESIGN OF EXPERIMENTS

In this study, three parameters were selected as control parameters and each

parameter was designed to have three levels. The experimental design was

according to an L9 array based on Taguchi method, while using the Taguchi

orthogonal array would markedly reduce the number of experiments. A set of

experiments designed using the Taguchi method was conducted to investigate

the relation between the parameters and delamination factor. DESIGN EXPERT

software was used for regression and graphical analysis of the obtained data.

DEVELOPMENT OF MODEL

Taguchi design and ANOVA consisting of 9 experiments was conducted to

develop model showing the relationship between the flux compositions and

response (impact strength, Vickers Hardness) for coded values of -1 to +1 for

each of flux constituents.

Page 28: Submerged Arc Welding NEURAL NETWORK

Fig 4.7 Design Layout

Fig 4.8 Summary

Page 29: Submerged Arc Welding NEURAL NETWORK

Fig 4.9 Graph Columns

a) NiO vs Vickers Hardness

b) MnO vs Vickers Hardness

Page 30: Submerged Arc Welding NEURAL NETWORK

c) MgO vs Vickers Hardness

d) NiO vs Impact Strength

Page 31: Submerged Arc Welding NEURAL NETWORK

e) MnO vs Impact Strength

f) MgO vs Impact Strength

Page 32: Submerged Arc Welding NEURAL NETWORK

Fig 4.10 Fit Summary

Page 33: Submerged Arc Welding NEURAL NETWORK

Table 4.1

Response 1 Vickers hardness

ANOVA for Response Surface Mean model

Analysis of variance table [Partial sum of squares - Type III]

Sum of Mean F p-value

Source Squares df Square Value Prob > F

Model 0.000 0

Residual 307.76 8 38.47

Cor Total 307.76 8

Std. Dev. 6.20 R-Squared 0.0000

Mean 45.26 Adj R-Squared 0.0000

C.V. % 13.70 Pred R-Squared -0.2656

PRESS 389.51 Adeq Precision

Coefficient Standard 95% CI 95% CI

Factor Estimate Df Error Low High

Intercept 45.26 1 2.07 40.49 50.03

Table 4.2

Final Equation in Terms of Coded Factors:

Vickers hardness =

+45.26

Table 4.3

Final Equation in Terms of Actual Factors:

Vickers hardness =

+45.26222

Page 34: Submerged Arc Welding NEURAL NETWORK

Table 4.4

Response 2 impact strength

ANOVA for Response Surface 2FI model

Analysis of variance table [Partial sum of squares - Type III]

Sum of Mean F p-value

Source Squares df Square Value Prob > F

Model 1303.46 6 217.24 1.33 0.4878 not significant

A-nio 26.36 1 26.36 0.16 0.7263

B-mno 5.66 1 5.66 0.035 0.8693

C-mgo 229.62 1 229.62 1.41 0.3569

AB 653.96 1 653.96 4.02 0.1829

AC 6.08 1 6.08 0.037 0.8646

BC 278.54 1 278.54 1.71 0.3210

Residual 325.60 2 162.80

Cor Total 1629.06 8

Std. Dev. 12.76 R-Squared 0.8001

Mean 59.10 Adj R-Squared 0.2005

C.V. % 21.59 Pred R-Squared -11.3952

PRESS 20192.57 Adeq Precision 3.984

Page 35: Submerged Arc Welding NEURAL NETWORK

Coefficient Standard 95% CI 95% CI

Factor Estimate df Error Low High VIF

Intercept 59.10 1 4.25 40.80 77.40

A-nio 3.17 1 7.88 -30.72 37.05 2.29

B-mno -1.47 1 7.88 -35.35 32.42 2.29

C-mgo -9.35 1 7.88 -43.24 24.53 2.29

AB -23.68 1 11.81 -74.50 27.15 3.43

AC 2.28 1 11.81 -48.54 53.11 3.43

BC 15.45 1 11.81 -35.38 66.28 3.43

Table 4.5

Final Equation in Terms of Coded Factors:

impact strength =

+59.10

+3.17 * A

-1.47 * B

-9.35 * C

-23.68 * AB

+2.28 * AC

+15.45 * BC

Table 4.6

Final Equation in Terms of Actual Factors:

impact strength =

+409.46563

+3.21735 * nio

-2.67771 * mno

-7.25643 * mgo

-0.059189 * nio * mno

+0.011414 * nio * mgo

+0.077257 * mno * mgo

Page 36: Submerged Arc Welding NEURAL NETWORK

Fig 4.11 FDS Graph

Fig . 4.12. One Factor Graph

Page 37: Submerged Arc Welding NEURAL NETWORK
Page 38: Submerged Arc Welding NEURAL NETWORK

Fig 4.13 Perturbation

Fig 4.14 Interaction

Page 39: Submerged Arc Welding NEURAL NETWORK

Fig 4.15 Contour

Fig 4.16 3D surface

Page 40: Submerged Arc Welding NEURAL NETWORK

Fig 4.17 Normal plot – Vickers hardness

Fig 4.18 Predicted vs Actual

Page 41: Submerged Arc Welding NEURAL NETWORK

REFERENCES

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY

RESEARCH VOLUME 3, ISSUE 1, JANUARY 2014

EFFECT OF FLUX COMPOSITION ON ELEMENT TRANSFER DURING

SUBMERGED ARC WELDING (SAW): A LITERATURE REVIEW

Optimization of Neural Networks: A Comparative Analysis

The ANN Book by R.M.Hristev

Neural Network Toolbox User’s Guide R2012a by Mark Hudson Beagle

Design and Analysis of Experiments by Douglas.C.Montgomery

Page 42: Submerged Arc Welding NEURAL NETWORK