You are on page 1of 105

DETECTION OF CRACK

IN CANTILEVER BEAM LIKE STRUCTURE


USING
ANN AND WAVELET TRANSFORM

Thesis submitted in partial fulfillment of


the requirement for the degree of

Master of Mechanical Engineering


BY

AMIT BANERJEE
(M4MEC12-11)
UNDER THE GUIDANCE OF

Dr. Goutam Pohit


Professor, Department of Mechanical Engineering

DEPARTMENT OF MECHANICAL ENGINEERING


FACULTY OF ENGINEERING &TECHNOLOGY
JADAVPUR UNIVERSITY
KOLKATA 700032
MAY 2012

FACULTY OF ENGINEERING AND TECHNOLOGY


JADAVPUR UNIVERSITY

CERTIFICATE OF APPROVAL *

This foregoing thesis is hereby approved as a credible study of an engineering


subject carried out and presented in a manner satisfactory to warrant its
acceptance as a prerequisite to the degree for which it has been submitted. It is
understood that by this approval the undersigned do not endorse or approve
any statement made, opinion expressed or conclusion drawn therein but
approve the thesis only for the purpose for which it has been submitted.

Committee

_______

On Final Examination for


Evaluation of the Thesis

_______

*Only in case the thesis is approved.

ii

FACULTY OF ENGINEERING AND TECHNOLOGY


JADAVPUR UNIVERSITY

We hereby recommend that the thesis presented under our


supervision by Mr. Amit Banerjee entitled DETECTION OF CRACK
IN CANTILEVER BEAM LIKE STRUCTURE USING ANN AND
WAVELET TRANSFORM be accepted in partial fulfillment of the

requirements for the degree of Master of Mechanical Engineering.

Countersigned

----------------------------------------Thesis Advisor

------------------------------------Head of the Department


Mechanical Engineering

iii

ACKNOWLEDGMENT

Foremost, I would like to express my sincere gratitude to my honorable guide Prof.


Goutam Pohit for the continuous support of my study and research, for his patience,
motivation, enthusiasm, and immense knowledge. His guidance helped me in all the time
of research and writing of this thesis. I could not have imagined having a better advisor and
mentor for my study.
I would like to convey my regards to the Laboratory-in-charge of Machine Elements
Laboratory and all other professors and lab technicians who helped me to complete the
thesis.
I thank my friends and lab-mates in machine design specialization for the stimulating
discussions, support and fun which were always constructive and refreshing. In this regard
I tender special thanks to Sri Kaushik Kumar who taught me how to operate the Ansys
software, Sri Anirban Mitra who taught me to operate the matlab computational simulation,
Sri Shouvik Ghosh who taught how to plot a graph in grapher software and Sri Partha
Ghosh who taught me about artificial neural network. Special mention of my friends
Supriyo, Dipendra and Milan is called for their continued support and help.

In this respect, I would also like to grab this opportunity to express my heart-felt
gratitude to my mother her constant cooperation, suggestions and helping attitudes let me
complete the thesis on time.

Date:

(AMIT BANERJEE)

iv

CONTENTS
Page No.
CERTIFICATE OF APPROVAL

ii

CERTIFICATE OF SUPERVISOR

iii

ACKNOWLEDGMENT

iv

CONTENTS

v-vii

LIST OF FIGURES

viii-xii

LIST OF TABLES

xiii-xiv

CHAPTER 1: INTRODUCTION

1-8

1.1 Literature Review

1.1.1 Crack detection by Artificial Neural Network analysis


for cracked structure

1.1.2 Crack detection by Wavelet analysis

1.2 Objective

CHAPTER 2: METHODOLOGY

9-33

2.1 Finite Element Analysis

2.2 Artificial Neural Network

14

2.2.1 Basic of neural Network

14

2.2.2 Application of Neural Network

16

2.2.3 Neuron Model

17

2.2.4 Network activation function and bias

19

2.2.5 Learning Rules

20

2.2.6 Feed-forward Networks

22

2.2.7 Backpropagation

23

[Type text]
2.2.8 Training

24

2.2.9 Simulation

25

2.2.10 Neural Network use in Present Study

25

2.3 Wavelet Transform

26

2.3.1 Introduction to wavelet

27

2.3.2 Wavelet Families

27

2.3.3 Continuous Wavelet Transform

31

2.3.4 Discrete Wavelet Transform

32

2.3.5 Wavelet Methodology for Crack Detection in Present Work

33

CHAPTER 3: RESULT AND DISCUSSION

34-86

3.1 Finite Element Simulation Result

34

3.2 Identification of Crack depth and location by Artificial Neural Network

40

3.2.1 Prediction of Depth of crack at a particular location using First


Three Relative Natural Frequencies as input

42

3.2.2 Prediction of Depth of crack at a particular location using First


Three Average Mode Shape Deviation as input

45

3.2.3 Prediction of Depth of crack at a particular location using First


Three Relative Natural Frequencies and First Three Average
Mode Shape Deviation as input

48

3.2.4 Prediction of Location of crack for particular depth using First


Three Relative Natural Frequencies as input

52

3.2.5 Prediction of Location of crack for particular depth using First


Three Average Mode Shape Deviation as input

vi

57

[Type text]

3.2.6 Prediction of Location of crack for particular depth using First


Three Relative Natural Frequencies and First Three Average
Mode Shape Deviation as input
3.2.7 Discussion

61
66

3.3 Identification of Crack Depth and Location by Wavelet Transform analysis

66

3.3.1 Beam under static loading

66

3.3.2 Dynamic analysis

75

3.4 Comparison with artificial neural network analysis and wavelet transform
analysis

85

CHAPTER 4: CONCLUTION AND SCOPE FOR FUTURE WORK

87-88

4.1 Conclusion

87

4.2 Future scope for ANN analysis of present work

87

4.3 Future scope for Wavelet Transform analysis of present work

88

REFERENCE

89-91

vii

List of Figures

Page No

Fig. 2.1

Crack cantilever beam with dimension

Fig. 2.2

Finite element modeling of the cracked cantilever beam

11

Fig.2.3(a)

First mode shape of cracked cantilever beam (crack location at 50 mm

12

from fixed end and depth 10 mm).


Fig. 2.3(b)

Second mode shape of cracked cantilever beam (crack location at 50

12

mm from fixed end and depth 10 mm).


Fig.2.3(c)

Third mode shape of cracked cantilever beam (crack location at 50 mm

13

from fixed end and depth 10 mm).


Fig.2.4

Static deflection of cracked cantilever beam (crack location at 50 mm

13

from fixed end and depth 10 mm) under 294 KN force at free end.
Fig.2.5

Layers of Neural Network

15

Fig.2.6

Simple Neuron without Bias

18

Fig.2.7

Simple Neuron with Bias

18

Fig.2.8

Neuron with Vector Input where R = number of elements in input

19

vector.
Fig.2.9

Multilayer feedforward network

22

Fig.2.10

Neuron Model for BP Algorithm

24

Fig.2.11

A typical wavelet in time domain

27

Fig.2.12

Haar wavelet

27

Fig.2.13

Morlet wavelet

28

Fig.2.14

Maxican wavelet

28

Fig.2.15

Mayer wavelet

28

Fig.2.16

Daubechies wavelet

29

Fig.2.17

Bioorthogonal Wavelet

30

Fig.2.18

Symlets Wavelet

30

Fig 3.1

Relative Natural Frequencies Vs Depth

39

viii

Fig 3.2

Relative Natural Frequencies vs. Location from fixed end for first

39

frequency
Fig.3.3

Relative Natural Frequencies vs. Location from fixed end for second

39

frequency
Fig.3.4

Relative Natural Frequencies vs. Location from fixed end for third

39

frequencies
Fig.3.5

Average Mode Shape Deviation vs Depth

40

Fig.3.6

Average Mode Shape Deviation vs. Crack Location form fixed end for

40

first natural frequencies


Fig.3.7

Average Mode Shape Deviation vs. Crack Location from fixed end for

40

second natural frequencies


Fig.3.8

Average Mode Shape Deviation vs. Crack Location from fixed end for

40

third natural frequencies


Fig.3.9

Performance curve for Relative natural frequencies and Depth network

43

Fig.3.10

Regression plot for Relative natural frequencies and Depth network

43

Fig.3.11

Performance curve for Average Mode Shape Deviation and Depth

46

network
Fig.3.12

Regression for Average Mode Shape Deviation and Depth network

47

Fig.3.13

Performance curve for Relative Natural Frequencies and Average

49

Mode Shape Deviation with Depth network


Fig.3.14

Regression for Relative Natural Frequencies and Average Mode Shape

50

Deviation with Depth network


Fig.3.15

Performance curve for Relative Natural Frequencies and Location

54

network
Fig.3.16

Regression for Relative Natural Frequencies and Location network

54

Fig.3.17

Performance curve for Mode Shape Deviation and Location network

58

Fig.3.18

Regression for Mode Shape Deviation and Location network

58

Fig.3.19

Performance curve for Relative Natural Frequencies and Average

63

Mode Shape Deviation with Location network.


Fig.3.20

Regression for Relative Natural Frequencies and Average Mode Shape

63

Deviation with Location network .


Fig.3.21

Static deflection of crack free and cracked beam with 3 mm cracked


depth at 50 mm from fixed end

ix

67

Fig.3.22

DWT plot of static deflection cracked beam with 3 mm crack depth at

68

location 50 mm from fixed end.

Fig.3.23(a)

CWT 2D plot of static deflection of cracked Beam crack location 50

68

mm from fixed end with 3 mm crack depth.


Fig.3.23(b)

CWT 3D plot of static deflection of cracked Beam crack location 50

68

mm from fixed end with 3 mm crack depth. (a) 2D plot (b) 3D plot
Fig.3.24

DWT plot of static deflection cracked beam at location 50 mm from

69

fixed end with 10 mm crack depth


Fig.3.25(a)

CWT 2D plots of static deflection of cracked beam crack location at 50

69

mm from fixed end with 10 mm crack depth.


Fig.3.25(b)

CWT 3D plots of static deflection of cracked beam crack location at

70

50 mm from fixed end with 10 mm crack depth.


Fig.3.26

Relative Static Deflection vs. Node for cracked beam with 3 mm crack

70

depth at location 50 mm from fixed end


Fig.3.27(a)

DWT plot of relative static deflection cracked beam and crack free

71

beam (crack location 50 mm from fixed end with 3 mm crack depth)


Fig.3.27(b)

CWT 2D plot of relative static deflection of cracked beam and crack

71

free beam (crack location 50 mm from fixed end with 3 mm crack


depth)
Fig.3.27(c)

CWT 3D plot of relative static deflection of cracked beam and crack

72

free beam (crack location 50 mm from fixed end with 3 mm crack


depth)
Fig.3.28(a)

DWT plot of relative static deflection cracked beam and crack free

72

beam (crack location 250 mm from fixed end with 3 mm crack depth)
Fig.3.28(b)

CWT 2D plot of relative static deflection of cracked beam and crack

72

free beam (crack location 250 mm from fixed end with 3 mm crack
depth)
Fig.3.28(c)

CWT 3D plot of relative static deflection of cracked beam and crack

73

free beam (crack location 250 mm from fixed end with 3 mm crack
depth).
Fig.3.29(a)

DWT plot of relative static deflection cracked beam and crack free
beam (crack location 450 mm from fixed end with 3 mm crack depth)

73

Fig.3.29(b)

CWT 2D plot of relative static deflection of cracked beam and crack

74

free beam (crack location 450 mm from fixed end with 3 mm crack
depth)

Fig.3.29(c)

CWT 3D plot of relative static deflection of cracked beam and crack

74

free beam (crack location 450 mm from fixed end with 3 mm crack
depth).
Fig.3.30

Maximum wavelet coefficient versus depth at location 50 mm from

75

fixed end
Fig.3.31

DWT plot of plot of original mode shape of Cracked Beam with 5 mm

76

crack depth and location 50 mm from fixed end


Fig.3.32(a)

CWT plot of original mode shape of Cracked Beam with 5 mm crack

76

depth and location 50 mm from fixed end


Fig.3.32(b)

CWT plot of original mode shape of Cracked Beam with 5 mm crack

77

depth and location 50 mm from fixed end


Fig.3.33

DWT plot of plot of original mode shape of Cracked Beam with 10

77

mm crack depth and location 50 mm from fixed end


Fig.3.34(a)

CWT 2D plot of original mode shape of Cracked Beam with 10 mm

77

crack depth and location 50 mm from fixed end


Fig.3.34(b)

CWT 3D plot of original mode shape of Cracked Beam with 10 mm

78

crack depth and location 50 mm from fixed end


Fig.3.35(a)

DWT plot of mode shape difference of Crack free and Cracked Beam

78

with 1 mm crack depth at crack location 50 mm from fixed end.


Fig.3.35(b)

CWT 2D plot of mode shape difference of Crack free and Cracked

79

Beam with 1 mm crack depth at crack location 50 mm from fixed end.


Fig.3.35(c)

CWT 3D plot of mode shape difference of Crack free and Cracked

79

Beam with 1 mm crack depth at crack location 50 mm from fixed end.


Fig.3.36(a)

DWT plot of mode shape difference of Crack free and Cracked Beam

79

with 5 mm crack depth at crack location 50 mm from fixed end.


Fig.3.36(b)

CWT 2D plot of mode shape difference of Crack free and Cracked

80

Beam with 5 mm crack depth at crack location 50 mm from fixed end.


Fig.3.36(c)

CWT 3D plot of mode shape difference of Crack free and Cracked


Beam with 5 mm crack depth at crack location 50 mm from fixed end.

xi

80

Fig.3.37(a)

DWT plot of mode shape difference of Crack free and Cracked Beam

80

with 10 mm crack depth at crack location 50 mm from fixed end.


Fig.3.37(b)

CWT 2D plot of mode shape difference of Crack free and Cracked

81

Beam with 10 mm crack depth at crack location 50 mm from fixed end.

Fig.3.37(c)

CWT 3D plot of mode shape difference of Crack free and Cracked

81

Beam with mm crack depth at crack location 50 mm from fixed end.


Fig.3.38(a)

DWT plot of mode shape difference of Crack free and Cracked Beam

82

with 3 mm crack depth at crack location 150 mm from fixed end.


Fig.3.38(b)

CWT 2D plot of mode shape difference of Crack free and Cracked

82

Beam with 3 mm crack depth at crack location 150 mm from fixed end.
Fig.3.38(c)

CWT 2D plot of mode shape difference of Crack free and Cracked

82

Beam with 3 mm crack depth at crack location 150 mm from fixed end.
Fig.3.39(a)

DWT plot of mode shape difference of Crack free and Cracked Beam

83

with 3 mm crack depth at crack location 250 mm from fixed end.


Fig.3.39(b)

CWT 2D plot of mode shape difference of Crack free and Cracked

83

Beam with 3 mm crack depth at crack location 250 mm from fixed end.
Fig.3.39(c)

CWT 3D plot of mode shape difference of Crack free and Cracked

83

Beam with 3 mm crack depth at crack location 250 mm from fixed end.
Fig.3.40

Maximum wavelet coefficient versus depth at location 50 mm from


fixed end for dynamic modal analysis of cracked beam

xii

85

List of Table

Page No
Table 2.1

Beam Characteristics

10

Table 2.2

Types of Activation Functions

20

Table 2.3

Parameters for neural network

26

Table 3.1

Different natural frequencies with relative natural frequencies

35

at different depth with crack location 50 mm from fixed end.


Table 3.2

Different natural frequencies with relative natural

36

frequencies at different location from fixed end with


fixed crack depth 3mm.
Table 3.3

Average relative mode shape deviation at different depth

37

with crack location 50 mm from fixed end.


Table 3.4

Different average mode shape deviation at different

38

location from fixed end with fixed crack depth 3mm.


Table 3.5

Different Function of Network

41

Table 3.6

Natural Frequencies with cracked depth at location 50 mm

42

from fixed end for training.


Table 3.7

Simulation data of Natural frequencies and depth at location

44

50 mm from fixed end.


Table 3.8

Comparing target result of depth with ANN result for

45

frequency depth network


Table 3.9

Average Mode Shape Deviation with cracked depth at

46

location 50 mm from fixed end for training.


Table 3.10

Simulation data of Mode Shape Deviation and Depth at

47

Crack location 50 mm from fixed end.


Table 3.11

Comparing target result of depth with ANN result for mode

48

shape deviation depth network


Table 3.12

Relative Natural Frequencies and Mode Shape Deviation


with cracked depth at location 50 mm from fixed end for
training.

xiii

49

Table 3.13

Simulation data of Relative Natural Frequencies with

51

Average Mode Shape Deviation and Depth at location of


Crack 50 mm from fixed end.
Table 3.14

Comparing target result of depth with ANN result for relative

52

natural frequency with average mode shape deviation depth


network
Table 3.15

Natural Frequencies with cracked location from fixed end

53

with fixed 3 mm depth for training.


Table 3.16

Simulation data of Natural frequencies and location from

55

fixed end for 3 mm crack depth


Table 3.17

Comparing target result of location from fixed end with ANN

56

result for relative natural frequency location network


Table 3.18

Mode Shape deviations with cracked location from fixed end

57

with fixed 3 mm depth for training.


Table 3.19

Simulation data of Mode shape deviation and location from

59

fixed end for 3 mm crack depth.


Table 3.20

Comparing target result of location from fixed end with ANN

60

result for average mode shape deviation location network


Table 3.21

Relative Natural Frequencies and Mode Shape Deviation

62

with cracked location from fixed end of 3 mm crack depth for


training.
Table 3.22

Simulation data of Relative Natural Frequencies with

64

Average Mode Shape Deviation and Location of 3 mm Crack


Depth.
Table 3.23

Comparing target result of crack location from fixed end with

65

ANN result for relative natural frequency with average mode


shape deviation location network.
Table 3.24

Maximum wavelet coefficient at different depth at location

75

50 mm from fixed end


Table 3.25

Maximum wavelet coefficient at different depth at location


50 mm from fixed end for dynamic modal analysis of cracked
beam

xiv

84

Chapter 1
INTRODUCTION
It is required that structures must safely work during its service life. However, presence of
damages may lead to breakdown of damages. Crack is one of the most common faults that if
develops, may cause catastrophic damages in structures. Therefore, it must be detected in the early
stage when it is small. Crack-like defects in mechanical and civil engineering structures are a
problem that received considerable attention by researchers. Since beam type structures are very
common used in steel construction and machinery industries, for the last few decades engineers and
scientists are working on various techniques for detection of crack in the beamlike structure.
Cracks are present in structure due to various reasons. The presences of crack could not only cause
a local variation in the stiffness but it could affect the mechanical behavior of the entire structure to a
considerable extent. Cracks may be caused by fatigue under service condition as a result of the
limited fatigue strength. Generally they are small in sizes. Such small cracks are known to propagate
due to the fluctuating stress conditions. If these propagating cracks remain undetected and reach their
critical size, then a sudden structural failure occurs.
Due to its practical importance, the crack identification problem in structures has been extensively
investigated and many methods are proposed. In practice, it is difficult to recognize most cracks by
using visual inspection techniques. Nowadays, the procedures that are often used for crack detection
are those which are called direct procedures, such as ultrasonic, X-rays, acoustic emission, wave
propagation, radiography etc. Such diagnostics can be effectively applied to damage detection in a
few known locations of the structure. However they are impractical in search of potential damage
through all engineering object and also may require minutely detailed periodic inspections, which is
very costly. In order to avoid these costs, during the last decades, people have been in the lookout of
a more efficient procedure in crack detection through vibration analysis.
A crack or local defect affects the dynamic response of structural member. It results in changes of
natural frequencies and mode shapes. A crack in a structure introduces a local flexibility that can
change the dynamics behavior of the structure and this property used to detect existence of a crack
together with its location and depth in the structural member. Most of the approaches use the modal
data of a structure before damage occurs as baseline data and all subsequent tests are compared to it.
Any deviation in the modal properties from the baseline data is used to estimate the crack size and
location.

Chapter 1
A neural network is a powerful data modeling tool that is able to capture and represent complex
input/output relationships. Recently, neural networks are expected to be a necessity for intensive
computation. Neural networks are expected to be a potential approach to detect the damage of the
structure. In this study feed-forward back-propagation networks are used to learn the input (the
frequency and depth) and output (the location and depth of a crack) relation of the structural system.
Wavelet analysis is also capable to detect the crack in structure. Wavelet transform is applied
on the static and dynamics response of a beam, The result obtained by the wavelet analysis may be
useful to detect the location of crack and its depth quite efficiently.

1.1 Literature Review


When a structure suffers from damages, its dynamics properties can change especially, crack
damage can cause a stiffness reduction, with inherent changes in natural frequencies, an increase in
modal damping and a change of the modal shapes. During the last few decades, considerable work
has been carried out for crack detection in beam like structure using vibration analysis in general and
ANN and wavelet analysis in particular. In what follows is a brief literature review of the work being
carried out relevant to the crack detection in beams.
Orhan Sadettin [1] has studied the free and forced vibration analysis of a cracked beam was
performed in order to indentify the crack in a cantilever beam. Single and two edge cracks were
evaluated. Dynamic response of the forced vibration better describes changes in crack depth and
location than the free vibration in which the difference between natural frequencies corresponding to
a change in crack depth and location only is a minor effect.
Friswell and Penny [2] have compared the different approaches to crack modeling and
demonstrate that for structural health monitoring using low frequency vibration, simple models of
crack flexibility based on beam element are adequate. The effect of the excitation for breathing
cracks, where the beam stiffness is bilinear, depending on whether the crack is open or closed has
also addressed in their studies.
Loutridis, Douka and Hadijileontiadis [3] have used a new method for crack detection in beams
based on instantaneous frequency and empirical mode decomposition is proposed. The dynamic
behaviour of a cantilever beam with a breathing crack under harmonic excitation is investigated both
theoretically and experimentally.

Introduction
Chinchalkar [4] have described a numerical method for determining the location of a crack in a
beam of varying depth when the lowest three natural frequencies of the cracked beam are known.
The crack is modeled as a rotational spring and graphs of spring stiffness versus crack location are
plotted for each natural frequency. The point of intersection of the three curves gives the location of
the crack.
Jinhee Lee [5] has presented a simple method of identifying multiple cracks in a beam using the
changes of the forced vibration amplitudes. The Newton-Raphson method and singular value
decomposition method are used for the estimation of the crack parameters.
Nandwana and Maiti [6] have used a method based measurement of natural frequencies is
presented for detection of the location and size of a crack in a stepped cantilever beam. The crack is
represented as a rotational spring, and the method involves obtaining plots of its stiffness with crack
location for any three natural modes through the characteristic equation. The point of intersection of
the three curves gives the crack location. Nandwana et al. [7] have modelling of transverse vibration
of a slender beam in the presence of an inclined edge or internal normal crack using a rotational
spring has been done to enable a possible detection of location of the crack based on the
measurement of natural frequencies. The characteristic equation obtained from the vibration analysis
of the beam is manipulated to give a relationship between the stiffness of the spring and location of
the crack.
Choubey et al. [8] have studied to analysis the effect of cracks on natural frequencies in two vessel
structure. Finite element analysis has been used to obtain the dynamic characteristics of intact and
damaged vessels for the eight modes of these structures.
Chan and Lai [9] have solved to determine the dynamics response during start-up of a turbogenerator. Investigations are performed using the advanced continuous simulation package ACSL via
a Runge Kutta numerical algorithm. Results of the simulation reveal that ultra harmonic resonances
of the system occur at one-half and one-third subcritical speeds, which can be used as a criterion for
crack growth detection.
Kim et al. [10] applied a methodology to nondestructively locate and estimate the size of damage
in structures for which a few natural frequencies or a few mode shapes are available. First, a
frequency-based damaged detection (FBDD) method is outlined. A damage-localization algorithm to
locate damage from the change of natural frequencies and a damage sizing algorithm to estimate

Chapter 1
crack-size from the natural frequency perturbation are formulated. Next, a mode-shape-based damage
detection (MBDD) method is outlined.
Peng, Lang and Chu [11] have used nonlinear output frequency response functions (NOFRFs) to
analyze a typical representation for cracked structures, a single degree of freedom system (SDOF)
bilinear model, to explain the occurrence of the nonlinear phenomena when a cracked structure is
subjected to sinusoidal excitations, including the generation of super-harmonic component and sub
resonances.
Nahvi and Jabbari [12] have presented an analytical, as well as experimental approach to the
crack detection in cantilever beams by vibration analysis. An experimental set up is designed in
which a cracked cantilever beam is excited by a hammer and response is obtained using an
accelerometer attached to the beam.
Kisa and Gurel [13] have presented a novel numerical technique applicable to analysis of uniform
and stepped cracked beams with circular cross section. In this approach the finite element and
component mode synthesis methods are used together, the beam being detached into parts from the
crack section.
Zheng et al. [14] have studied the natural frequencies and mode shapes of a cracked beam using
the finite element method. An overall additional flexibility matrix is added to the local flexibility
matrix of the corresponding intact beam element to obtain the total flexibility matrix and therefore
the stiffness matrix.
Saavedra et al. [15] have presented a theoretical and experimental dynamics behavior of different
multi-beams systems containing a transverse crack. The additional flexibility that the crack generates
in its vicinity is evaluated using the strain energy density function given by the linear fracture
mechanics theory.

1.1.1 Crack detection by Artificial Neural Network analysis for cracked


structure
Rosales et al. [16] have presented the solution of the inverse problem with a power series
technique (PST) and the use of artificial neural networks (ANNs) for detection of crack in cantilever
beam.
Bakhary et al. [17] have studied a statistical approach to take into account the effect of
uncertainties in developing an ANN model. By applying Rosenblueths point estimate method
4

Introduction
verified by Monte Carlo simulation, the statistics of the stiffness parameters are estimated. The
probability of damage existence (PDE) is then calculated based on the probability density function of
the existence of undamaged and damaged states. The developed approach is applied to detect
simulated damage in a numerical steel portal frame model and also in a laboratory tested concrete
slab.
Liu et al. [18] have studied both direct problem and inverse problem. In the direct problem, the
frequency responses of a cracked medium subjected to an impact loading are calculated by the
computational mechanics combining the finite element method with the boundary integral equation.
In the inverse problem, the back-propagation neural networks are trained by the characteristic
parameters extracted from the various surfaces responses obtained from the direct problem.
Parhi and Dash [19] have studied the dynamic behaviour of a beam structure containing multiple
transverse cracks using neural network controller. The first three natural frequencies and mode
shapes have been calculated using theoretical, finite-element, and experimental analysis for the
cracked and un-cracked beam. Comparisons of the results among theoretical, finite-element and
experimental analysis have also been presented.
Das and Parhi [20] have presented neural network technique for fault diagnosis of a cracked
cantilever beam. In the neural network system there are six input parameters and two output
parameters. The input parameters to the neural network are relative deviation of first three natural
frequencies and first three modes shapes. The output parameters of the neural network system are
relative crack depth and relative crack location. To calculate the effect of crack depths and crack
locations on natural frequencies and mode shapes, theoretical expressions have been developed.
Saeed, Galybin and Popov [21] have presented different artificial intelligence techniques for crack
identification in curvilinear beams based on changes in vibration characteristics. Vibration analysis
has been performed by applying the finite element method to compute natural frequencies and
frequency response functions for intact and damaged beams.
Mahmoud and Kiefa [22] have presented general regression neural networks (GRNN) to solve the
inverse vibration problem of cracked structures. The case study used in the investigation was a steel
cantilever beam with a single edge crack. The first six natural frequencies were used as network
inputs, and crack size and crack location were the output.
z and zkaya [23] have studied the transverse vibration of a Euler-Bernoulli type axially moving
beam. The beam is simply supported at both ends. Axial velocity is assumed as a harmonic function
5

Chapter 1
about a constant mean value. The frequency value and stability borders obtained in a previous study
are trained using ANN. For new value of flexural stiffness and mean velocities, frequencies and
stability borders are determined using ANN.
Nazari and Baghalian [24] have presented a new method for crack detection in symmetric beams.
In this research it is assumed that the structure is a rectangular beam which is fixed at both ends.
Finite Element Method was used to obtain natural frequencies of beam in different condition of
cracks. Based on data were obtained from FEM, two distinct Artificial Neural Networks were trained
for crack location and depth detection in some different conditions and then were tested. Finally
using an algorithm based on first vibration mode shape of structure, locations and depths of cracks
have been identified with good approximations.

1.1.2 Crack detection by Wavelet analysis


Zou et al.[25] have presented the modified function of torsional vibrations in cracked rotor system
based on simple hinge crack model and local flexibility and also studied the torsional stiffness
variation with crack depth. By wavelet time-frequency analysis, the time-frequency features of
torsional vibrations in cracked rotor system are investigated.
Xiang et al. [26] have studied model-based crack identification method for estimating crack location
and size in shaft by B-spline wavelet analysis. The crack is considered through local stiffness change.
Based on Rayleigh beam theory, the influences of rotary inertia on the flexural vibrations of the rotor
system are examined to construct BSWI Rayleigh beam element. The slender shaft and stiffness disc
are modeled by BSWI Rayleigh-Eular beam element and BSWI Rayleigh beam element respectively.
Then the crack identification forward and inverse problems are solved by using surface-fitting
technique and contour plotting method. The cracked shaft is modeled by wavelet-based elements to
gain frequencies. The first three measured frequencies are used in crack detection process and the
normalized crack location and depth are detected by means of genetic algorithm [27].
Quek et al. [28] have examined the sensitivity of wavelet technique in the detection of cracks in
beam structures having different crack characteristics and boundary conditions. Wavelet functions
employed are investigated by Haar and Gabor wavelets analysis.
Arcangelo Messina [29] has presented significant refinements concerning the use of wavelets and
have used in the guise of continuous wavelet transforms (CWT) for identifying damage on
transversally vibrating structural components. (E.g. beams, plates and shells).

Introduction
Ovanesova et al. [30] have presented the applications of the wavelet transform to detect cracks in
frame structures, such as beam and plane frames and also detected the localization of the crack by
using a response signal from static or dynamic loads.
Lonkar and Srivastava [31] applied wavelet transform to detect localized damage based on
simulated dynamic response data from finite element analysis of cantilever beam.
Douka, Loutridis and Trochidis [32] have presented a simple method for crack identification in
beam structures based on wavelet analysis. The fundamental vibration mode of a cracked cantilever
beam is analyzed using continuous wavelet transform and both the location and size of the crack are
estimated. Douka et al [33] have also presented crack identification in double cracked beams based
on wavelet analysis.
Wang and Wu [34] have an experimental studied for the location detection of a delamination in a
beam structure under a static displacement with a spatial wavelet transform. An invisible perturbation
in the deflection profile of the delaminated beam at the two delamination edges owing to the
curvature discontinuity is discerned or amplified through a wavelet transform.
Andrzej Katunin [35] has presented the construction of general order two-dimensional B-spline
wavelets and has applied for damage identification in polymeric composite plates.
Kim and Melhem [36] have provided the review of the research that has been conducted on damage
detection by wavelet analysis. First, the theory of wavelet analysis is presented including continuous
and discrete wavelet transform followed by its application to SHM. Then more specific application
namely crack detection of a beam and mechanical gear and roller damage are presented.
Jiang et al. [37] stated a new method for crack detection in beams by using the slop of the mode
shape to detect cracks and introduced the angle coefficients of complex continuous wavelet
transform.
Srinivasa Rao, Mallikarjuna Rao, G.V.Raju [38] have presented in their paper a method for crack
identification in beam structure by analyzing the fundamental mode of cracked cantilever beam
using continuous wavelet transform.
Okafor and Dutta [39] have used a laser-based optical system and wavelet transforms for detection
of changes in the properties of cantilevered aluminum beams as a result of damage. The beams were
modeled using the ANSYS 5.3 finite-element method and the first six mode shapes for the damaged
and the undamaged cases obtained.
7

Chapter 1
Zhong and Oyadiji [40, 41] have proposed a new approach for damage detection in beam-like
structures with small cracks, whose crack ratio [r =

] is less than 5%, without baseline modal

parameters. The approach is based on the difference of the continuous wavelet transforms and
stationary wavelet transforms of two sets of mode shape data which correspond to the left half and
the right half of the modal data of a cracked simply-supported beam.

1.2 Objective
From the literature review, it has been observed that various techniques have been already
employed for identification of crack, its location and size in case of beam like structure.

In the present study, the problem of fault recognition of a cracked cantilever beam is carried out in
two different methods using vibration signature obtained from finite element analysis of the beam
under static and dynamic loads.

In the first case, Artificial Neural Network is applied to identify crack size and its location from the
response data obtained from the numerical experiment conducted by finite element modelling of the
crack beam and subsequent analysis using ANSYS.

In the second method, wavelet transform is applied to detect the crack in the cantilever beam.
Results obtained from both the techniques are compared.

Chapter 4
METHODOLOGY
In the present work, vibration analysis is carried out on a cantilever beam with a single open
transverse cracks. The study is aimed at detecting the size and location of crack in cantilever beam.

Free vibration finite element analysis of crack free beam and a series of cracked beams are
performed. Suitable boundary conditions are used to find out natural frequencies and mode shapes.

The simulations have done with the help of ANSYS 13 software. Comparison studies are
performed on natural frequencies and mode shapes of cracked and cracked free beam. Identification
of crack depth and its location is determined by two different methods, namely, ANN analysis and
wavelet analysis.

Six distinct Artificial Neural Networks (ANNs) are trained for detection of different crack
location and depth. The inputs are six variables, first three natural frequencies and first three natural
mode shapes and output are depth and location. The MATLAB 7.8 Neural Network Toolbox is used
for this purpose. In section 4.2, a brief introduction to Artificial Neural Network system as applied to
the present study is outlined. Once the training is complete simulation is performed to predict
probable location of crack and its depth.

Wavelet transform is applied to detect the crack of cantilever beam. Relative mode shape
deviations of crack and crack free beam are used as input signals of wavelet analysis. Both discrete
and continuous analyses are performed. The wavelet analysis is performed using MATLAB 7.8
Wavelet toolbox. Both discrete and continuous wavelet transform of localized crack in a cantilever
under static and dynamic conditions are analyzed to detect crack is presented. Application of wavelet
analysis is described in section 2.3.

2.1 Finite Element Analysis

Fig. 2.1 Crack cantilever beam with dimension


9

Chapter 4
Finite element model of a cantilevers beam with single open transverse crack is developed in
ANSYS environment. The dimensions of the beam are shown in Fig 2.1. Material properties of the
beam are shown in Table 2.1. In order to perform numerical experiment, modal and structural analysis
of the beam is performed following the steps outlines below.
Define Materials

Set preferences. (Structural)

Define constant material properties.

Model the Geometry

Create the geometry

Generate Mesh

Define element type.

Mesh the area.

Apply Boundary Conditions

Apply constraints to the model.

Obtain Solution

Specify analysis types and options.

Solve.
Table 2.1 Beam Characteristics
Density Poisson Ratio
(Kg/ )
7860

0.3

Elasticity
Modulus
(GPa)
206

Length
(mm)
600

Thickness
(mm)
20

Depth
(mm)
30

Throughout transverse cracks is modeled with 1 mm width for every cracked beam.
Finite element software, ANSYS version 13 is used for free vibration analysis of the crack free and
cracked beams. Beam length, thickness and depth are along X axis, Y axis and Z axis respectively in
ANSYS coordinate system. A 20-node three dimension structural solid element under SOLID 186
was selected to model the beam because it is suitable for all structural analysis and it is mid node
element which gives the more accurate result. Fig 2.2 shows finite element model of a cracked beam.
The modal analysis of cracked and crack free beams are performed. The Block Lanczons mode
extraction method is used to calculate the natural frequencies of the beam. First five modes have been
selected as for both crack and crack free beam as first, third and fifth natural frequencies correspond
10

Methodology
to first three natural frequencies in the transverse direction (Y direction) of vibration. The
corresponding mode shapes for both cracked and cracked free beam are also captured. In order to
determine the mode shape and modal displacement, node points at the bottom surface of the beam are
considered. Distance of each sample point along the lengths is assumed to be 10 mm. First point is
taken at fixed end. Since length of the beam is 600 mm, the total number of sample data point is 61.
The mode shape of the beam is obtained by plotting transverse displacement (Y direction) of the
beam at each sample data point against its position along X direction. The displacement values are
used as a numerical data of ANN and mode shape is used as an input signal of wavelet analysis for
subsequent identification of crack length and location of the beam. For example, Figure 4.3 exhibits
first three mode shape of cracked beam with natural frequency and nodal displacements of the sample
point.
A number of simulations have been performed for various crack depth at different location. The
output values of these simulations are used as training data for the ANN analysis and wavelet
transform analysis.

Fig.2.2 Finite element modeling of the cracked cantilever beam


For static analysis, a load of 294 KN is applied at the free end of the cantilever beam. After static
analysis similar procedure is followed for taking sample data from cantilever beam as mentioned
above. The deflection curve thus obtained is used for wavelet analysis. Figure 2.4 shows static
deflection of the cracked cantilever beam under the application of load.
11

Chapter 4

Fig.2.3 (a) First mode shape of cracked cantilever beam (crack location at 50 mm from fixed end and
depth 10 mm).

Fig.2.3 (b) Second mode shape of cracked cantilever beam (crack location at 50 mm from fixed end
and depth 10 mm).
12

Methodology

Fig.2.3 (c) Third mode shape of cracked cantilever beam (crack location at 50 mm from fixed
end and depth 10 mm).

Fig 2.4 Static deflection of cracked cantilever beam (crack location at 50 mm from fixed end and
depth 10 mm) under 294 KN force at free end.

13

Chapter2
2.2 Artificial Neural Network
Studies on neural networks have been motivated to imitate the way that the brain operates .A
network is described in terms of the individual neurons, the network connectivity, the weights
associated with various interconnections between neurons, and the activation functions for each
neuron. The network maps an input vector from one space to another. The mapping is not specified,
but is learned. The network is presented with a given set of inputs and their associated outputs. The
learning process is used to determine proper interconnection weights and the network is trained to
make proper associations between the inputs and their corresponding outputs. Once trained, the
network provides rapid mapping of a given input into the desired output quantities. This, in turn, can
be used to enhance the efficiency of the design process.
The simplest definition of a neural network, more properly referred to as an artificial neural
network (ANN), is provided by the inventor of one of the first neurocomputers, Dr. Robert HechtNielsen. He defines a neural network as computing system made up of a number of simple, highly
interconnected processing elements, which process information by their dynamic state response to
external inputs. ANNs are processing devices (algorithms or actual hardware) that are loosely
modeled after the neural structure of the mammalian cerebral cortex but on much smaller scales. A
large ANN might have hundreds or thousands of processor units, whereas a mammalian brain has
billion of neurons with a corresponding increase in magnitude of their overall interaction and
emergent behavior.

2.2.1 Basic of neural Network


Neural networks are typically organized in layers. Layers are made up of a number of
interconnected nodes which contain an activation function. Patterns are presented to the network
via the input layer which communicates to one or more hidden layers where the actual processing
is done via a system of weighted connection. The hidden layers then link to an output layer where
the answer is output.

An artificial neural network is developed with a systematic step-by-step procedure which


optimizes a criterion commonly known as the learning rule. The input/output training data is
fundamental for these networks as it conveys the information which is necessary to discover the
optimal operating point. In addition, a non linear nature makes neural network processing elements a
very flexible system.

14

Methodology

Basically, an artificial neural network is a system. A system is a structure that receives an


input, process the data, and provides an output. Commonly, the input consists in a data array which
can be anything such as data from an image file, a WAVE sound or any kind of data that can be
represented in an array. Once an input is presented to the neural network, and a corresponding
desired or target response is set at the output, an error is composed from the difference of the desired
response and the real system output.

Fig 2.5 Layers of Neural Network

The error information is fed back to the system which makes all adjustments to their parameters
in a systematic fashion (commonly known as the learning rule). This process is repeated until the
desired output is acceptable. It is important to notice that the performance hinges heavily on the data.
Hence, this is why this data should pre-process with third party algorithms such as DSP algorithms.

In neural network design, the engineer or designer chooses the network topology, the trigger
function or performance function, learning rule and the criteria for stopping the training phase. So, it
is pretty difficult determining the size and parameters of the network as there is no rule or formula to
do it. The best we can do for having success with our design is playing with it. The problem with this
method is when the system does not work properly it is hard to refine the solution. Despite this issue,
neural networks based solution is very efficient in terms of development, time and resources. By

15

Chapter2
experience, I can tell that artificial neural networks provide real solutions that are difficult to match
with other technologies.
Depending on the nature of the application and the strength of the internal data patterns it can
generally expect a network to train quite well. This applies to problems where the relationships may
be quite dynamics or non linear. ANNs provide an analytical alternative to conventional techniques
which are often limited by strict assumption of normality, linearity, variable independence etc.
Because an ANN can capture many kinds of relationships it allows the user to quickly and relatively
easily model phenomena which otherwise may have been very difficult or impossible to explain
otherwise. Some of characteristics of ANN are given below.

A neural network can perform tasks that a linear program cannot.

When an element of the neural network fails, it can continue without any problem by
their parallel nature.

It can be implemented in any application.

A neural network learns and does not need to be reprogrammed.

It can be implemented without any problem.

The neural network needs training to operate.

The architecture of a neural network is different from the architecture of microprocessors


therefore needs to be emulated.

Requires high processing time for large neural networks.

Another aspect of the artificial neural networks is that there are different architectures, which
consequently requires different types of algorithms, but despite to be an apparently complex system,
a neural network is relatively simple.

2.2.2 Application of Neural Network


There are many different types of Neural Networks, each of which has different strengths
particular to their applications. The abilities of different networks can be related to their structure,
dynamics and learning methods. Neural Networks offer improved performance over conventional
technologies in areas which includes: Machine Vision, Robust Pattern Detection, Signal Filtering,
Virtual Reality, Data Segmentation, Data Compression, Data Mining, Text Mining, Artificial Life,
Adaptive Control, Optimization and Scheduling, Complex Mapping and more.
16

Methodology

ANNs have been applied successfully in a various fields of mathematics, engineering, medicine,
economics, meteorology, psychology, neurology and many others. Some of the most important ones
are:

Co Evolution of Neural Networks for Control of Pursuit & Evasion

Learning the Distribution of Object Trajectories for Event Recognition

Radiosity for Virtual Reality Systems (ROVER)

Autonomous Walker & Swimming Eel

Robocup: Robot World Cup

Using HMM's for Audio-to-Visual Conversion

Artificial Life: Galapagos

Speech reading (Lip-reading)

Detection and Tracking of Moving Targets

Real-time Target Identification for Security Applications

Facial Animation

Behavioral Animation and Evolution of Behavior

Identification of military target

Identification of explosives in passenger suitcases

Weather and market trends forecasting

Nowadays, neural network technologies are emerging as the technology choice for many
applications, such as patter recognition, prediction, system identification and control.

2.2.3 Neuron Model


A neuron with a single scalar input and no bias appears on the left is shown below. The
scalar input p is transmitted through a connection that multiplies its strength by the scalar weight w to
form the product wp, again a scalar. Here the weighted input wp is the only argument of the transfer
function f, which produces the scalar output a.

17

Chapter2

Fig. 2.6 Simple Neuron without Bias


The neuron on the right has a scalar bias, b. You can view the bias as simply being added to the
product wp as shown by the summing junction or as shifting the function f to the left by an amount b.
The bias is much like a weight, except that it has a constant input of 1.

Fig.2.7 Simple Neuron with Bias


The transfer function net input n, again a scalar, is the sum of the weighted input wp and the bias
b. This sum is the argument of the transfer function f. Here f is a transfer function, typically a step
function or a sigmoid function, that takes the argument n and produces the output a. w and b are both
adjustable scalar parameters of the neuron. The central idea of neural networks is that such
parameters can be adjusted so that the network exhibits some desired or interesting behavior. Thus,
you can train the network to do a particular job by adjusting the weight or bias parameters, or perhaps
the network itself will adjust these parameters to achieve some desired end.
A neuron with a single R-element input vector is shown below. Here the individual element
inputs

,...

are multiplied by weights

, ...

and the weighted values are fed to

the summing junction. Their sum is simply Wp, the dot product of the (single row) matrix W and the
vector p.

18

Methodology

Fig 2.8 Neuron with Vector Input where R = number of elements in input vector.

The neuron has a bias b, which is summed with the weighted inputs to form the net input n. This
sum, n, is the argument of the transfer function f.
n=

+ ... +

+b

This expression can, of course, be written in MATLAB code as:


n = W*p + b

However, you will seldom be writing code at this level, for such code is already built into
functions to define and simulate entire networks.

2.2.4 Network activation function and bias


The perception internal sum of the inputs is passed through an activation function, which can be
any monotonic function. Linear functions can be used but these will not contribute to a non-linear
transformation within a layered structure, which defeats the purpose of using a neural filter
implementation. A function that limits the amplitude range and limits the output strength of each
perception of a layered network to a defined range in a non-linear manner will contribute to a
nonlinear transformation. There are many forms of activation functions, which are selected according
to the specific problem. All the neural network architectures employ the activation function which
defines as the output of a neuron in terms of the activity level at its input (ranges from -1 to 1 or 0 to
1). The most practical activation functions are the sigmoid and the hyperbolic tangent

functions. This is because they are differentiable.


The bias gives the network an extra variable and the networks with bias are more powerful than
those of without bias. The neuron without a bias always gives a net input of zero to the activation

19

Chapter2
function when the network inputs are zero. This may not be desirable and can be avoided by the use
of a bias.

Table 2.2 Types of Activation Functions

Name

Mathematical Representation
f(x) = kx

Linear
Step

f (x) ={

Sigmoid

f(x) =

Hyperbolic Tangent

f(x) =

, >0

>0

Gaussian

f(x)=

2.2.5 Learning Rules


A learning rule defines as a procedure for modifying the weights and biases of a network.
There are three major learning paradigms, each corresponding to a particular abstract learning
task. These are supervised learning, unsupervised learning and reinforcement learning.
In supervised learning, we are given a set of example pairs (x, y), x ,
find a function f: X

and the aim is to

in the allowed class of functions that matches the examples. In other words,

we wish to infer the mapping implied by the data; the cost function is related to the mismatch
between our mapping and the data and it implicitly contains prior knowledge about the problem
domain. A commonly used cost is the mean squared error, which tries to minimize the average
squared error between the network's output, f(x), and the target value y over all the example pairs.
When one tries to minimize this cost using gradient descent for the class of neural networks called
multilayer perceptions, one obtains the common and well-known back propagation algorithm for
training neural networks.

20

Methodology
Tasks that fall within the paradigm of supervised learning are pattern recognition (also known as
classification) and regression (also known as function approximation). The supervised learning
paradigm is also applicable to sequential data (e.g., for speech and gesture recognition). This can be
thought of as learning with a "teacher," in the form of a function that provides continuous feedback
on the quality of solutions obtained thus far.
In unsupervised learning, some data x is given and the cost function to be minimized, that can be
any function of the data x and the network output, f
The cost function is dependent on the task (what we are trying to model) and our a priori
assumptions (the implicit properties of our model, its parameters and the observed variables)
As a trivial example, consider the model f(x) = a where a is a constant and the cost

[{ ( )} ] minimizing this cost will give us a value of a that is equal to the mean of the data.
The cost function can be much more complicated. Its form depends on the application: for example,
in compression it could be related to the mutual information between x and f(x) whereas in statistical
modeling, it could be related to the posterior probability of the model given the data. (Note that in
both of those examples those quantities would be maximized rather than minimized).
Tasks that fall within the paradigm of unsupervised learning are in general estimation problems;
the applications include clustering, the estimation of statistical distributions, compression and
filtering.
In reinforcement learning, data x are usually not given, but generated by an agent's interactions
with the environment. At each point in time t the agent performs an action
generates an observation

and an instantaneous cost

and the environment

according to some (usually unknown)

dynamics. The aim is to discover a policy for selecting actions that minimizes some measure of a
long-term cost; i.e., the expected cumulative cost. The environment's dynamics and the long-term
cost for each policy are usually unknown, but can be estimated.
Tasks that fall within the paradigm of reinforcement learning are control problems, games and
other sequential decision making tasks.

21

Chapter2
2.2.6 Feed-forward Networks
A feedforward neural network is a biologically inspired classification algorithm. It consists of a
(possibly large) number of simple neuron-like processing units, organized in layers. Every unit in a
layer is connected with all the units in the previous layer. These connections are not all equal; each
connection may have a different strength or weight. The weights on these connections encode the
knowledge of a network. Often the units in a neural network are also called nodes.

Data enters at the inputs and passes through the network, layer by layer, until it arrives
at the outputs. During normal operation, that is when it acts as a classifier, there is no
feedback between layers. This is why they are called feedforward neural networks.
Feed-forward networks have the following characteristics:

Perceptrons are arranged in layers, with the first layer taking in inputs and the last
layer producing outputs. The middle layers have no connection with the external
world, and hence are called hidden layers.

Each perceptron in one layer is connected to every perceptron on the next layer. Hence
information is constantly "fed forward" from one layer to the next and this explains why
these networks are called feed-forward networks.

There is no connection among perceptrons in the same layer

Fig .2.9 Multilayer feedforward network

22

Methodology
2.2.7 Backpropagation
Input vectors and the corresponding target vectors are used to train a network until it can
approximate a function, associate input vectors with specific output vectors, or classify input vectors
in an appropriate way as defined by you. Networks with biases, a sigmoid layer, and a linear output
layer are capable of approximating any function with a finite number of discontinuities.
Standard backpropagation is a gradient descent algorithm in which the network weights are
moved along the negative of the gradient of the performance function. The term backpropagation
refers to the manner in which the gradient is computed for nonlinear multilayer networks. There are a
number of variations on the basic algorithm that is based on other standard optimization techniques,
such as conjugate gradient and Newton methods. Neural Network Toolbox implements a number of
these variations. This chapter explains how to use each of these routines and discusses the advantages
and disadvantages of each.
Properly trained backpropagation networks tend to give reasonable answers when presented with
inputs that they have never seen. Typically, a new input leads to an output similar to the correct
output for input vectors used in training that are similar to the new input being presented. This
generalization property makes it possible to train a network on a representative set of input/target
pairs and get good results without training the network on all possible input/output pairs. There are
two features of Neural Network Toolbox that are designed to improve network generalization:
regularization and early stopping.
The primary objective is to explain how to use the backpropagation training functions in the
toolbox to train feedforward neural networks to solve specific problems. There are generally four
steps in the training process:
1. Assemble the training data
2. Create the network object
3. Train the network
4. Simulate the network response to new inputs

Backpropagation Algorithm
The network that is most commonly used with the backpropagation algorithm is the multilayer
feed-forward network. An elementary neuron with R inputs is shown below. Each input is weighted
23

Chapter2
with an appropriate w. The sum of the weighted inputs and the bias forms the input to the transfer
function f. Neurons can use any differentiable transfer function f to generate their output.

Fig.2.10 Neuron Model for BP Algorithm

2.2.8 Training
The first step in training a feedforward network is to create the network object. The function
newff create a feedforward network.
Before training a feedforward network, the weights and biases must be initialized. The newff
command will automatically initialize the weights but it may want to reinitialize them. This can be
done with the command init. This function takes a network object as input and returns a network
object with all weights and biases initialized. i.e. net=init (net).
After initialized, the network is ready for training. The training process requires a set of
examples of proper network behavior- network inputs and target outputs. Various training functions
can be used to train the network to reach from a particular input to a specific target output. As each
input is applied to the network the network the network output is compared with the actual target
value and the error is calculated. The goal is to minimize the average of sum of these errors which is
called as Mean Square Error (MSE) of the output.
Mean Squared Error (MSE) =

( )

( )

where t(k) is the actual value, a(k) is the network value and Q is the number of epochs.
When the MSE falls below a predetermined value or the maximum number of epochs have been
reached the training process stops.

24

Methodology
2.2.9 Simulation
The trained network can be used for simulating the system. The sim function simulates a network.
It takes the network input, and the network object and returns the network outputs. The trained
networks are supplied with new inputs and the predicted outputs of the networks can be obtained.
Method used for improving the generalization performance is called early stopping. This
technique is automatically provided for all of the supervised network creation functions, including
the backpropagation network creation functions. In this technique the available data is divided into
three subsets. The first subset is the training set, which is used for computing the gradient and
updating the network weights and biases. The second subset is the validation set. The error or the
validation set is monitored during the training process. The validation error normally decreases
during the training process. However, when the network begins to over fit the data, the error of
validation set typically begin to rise. When the validation error increase for a specified number of
iterations, the training is stopped and the weight and biases at the minimum of the validation error are
returned. The third subset is test set which is used to compare different models. The error in the test
set reaches a minimum at a significantly different iteration number than the validation set error, this
might indicate a poor division of data set.

2.2.10 Neural Network Use in Present Study


In the present work, crack detection of cantilever beam is obtained using Artificial Neural
Network. The Levenberg-Marquardt (TRINLM) algorithm is used for training the network as this
algorithm is often the fastest backpropagation algorithm in MATLAB Neural Network Toolbox and
is highly recommended as a first choice supervised algorithm. The results of finite element modal
analysis are used as an input to the neural network analysis whereas depth and location are used as
target value during the training process.
Instead of feeding the FE results directly, relative natural frequency and percentage deviation of
the corresponding mode shapes are calculated and are used as input parameters. Relative frequencies
are determined by comparing frequencies of cracked beam with cracked free beams. It should be
mentioned that first input is normalized between 0 to 1 and then it is used to train. Target parameters
are also normalized between 0 to 1. The number of neurons in the hidden layers and number of
hidden layers are varied to achieve beast results. After proper training it is simulated for other
cracked beam and the location and depth of the crack for those beams are thus obtained.

25

Chapter2
Altogether six deferent networks are used in this study to determine two output parameter as
shown in table below.
Table 2.3 Parameters for neural network
Input parameters
First three relative natural frequencies in
particular location with different depth
First three mode shapes deviation in particular
location with different depth
First three relative natural frequencies in
particular depth with different location
First three mode shapes deviation in particular
depth with different location
First three relative natural frequencies and mode
shapes deviation in particular location with
different depth
First three relative natural frequencies and mode
shapes deviation in particular depth with
different location

Output parameter
Depth
Depth
Location
Location
Depth

Location

2.3 Wavelet Transform


A wavelet is a wave-like oscillation with amplitude that starts out at zero, increases, and then
decreases back to zero. It can typically be visualized as a "brief oscillation" like one might see
recorded by a seismograph or heart monitor. Generally, wavelets are purposefully crafted to have
specific properties that make them useful for signal processing. Wavelets can be combined, using a
"reverse, shift, multiply and sum" technique called convolution, with portions of an unknown signal
to extract information from the unknown signal.
As a mathematical tool, wavelets can be used to extract information from many different kinds of
data, including -. A set of "complementary" wavelets will deconstruct data without gaps or overlap so
that the deconstruction process is mathematically reversible. Thus, sets of complementary wavelets
are useful in wavelet based but certainly not limited to - audio signals and images. Sets of wavelets
are generally needed to analyze data fully compression/decompression algorithms where it is
desirable to recover the original information with minimal loss.

26

Methodology
2.3.1 Introduction to wavelet
Wavelet transform is a tool that cuts up data, functions or operators into different frequency
components and then studies each component with a resolution matched to its scale. A wavelet is a
small wave which has its energy concentrated in time. It has an oscillating wavelike characteristic but
also has the ability to allow simultaneous time and frequency analysis and it is suitable tool for
transient, non-stationary or time-varying phenomena. Just as the Fourier transform decomposes a
signal into a family of complex sinusoidal, the wavelet transform decomposes a signal into a family
of wavelets. Unlike sinusoids, which are symmetric smooth and regular, wavelets can be symmetric
or asymmetric, sharp or smooth, regular or irregular.

Fig.2.11 A typical wavelet in time domain


Wavelet analysis is capable of revealing aspects of data that other signal analysis techniques miss
aspects like trends, breakdown points, discontinuities in higher derivatives, and self-similarity.
Further, because it affords a different view of data than those presented by traditional techniques,
wavelet analysis can often compress or de-noise a signal without appreciable degradation.

2.3.2 Wavelet Families


Haar: Haar wavelet is a sequence of rescaled "square-shaped" functions. It is discontinuous, and
resembles a step function.

Fig 2.12 Haar wavelet


27

Chapter 2

Morlet: Morlet wavelet is a wavelet composed of a complex exponential (carrier) multiplied by a


Gaussian window. This wavelet has no scaling function but it explicit.

Fig 2.13 Morlet wavelet

Mexican Hat: Mexican hat wavelet is the negative normalized second derivative of a Gaussian
function, i.e., up to scale and normalization. This wavelet has no scaling function and is derived
from a function that is proportional to the second derivative function of the Gaussian probability
density function.

ig 2.14 Maxican wavelet


Mayer: The Meyer wavelet and scaling function are defined in the frequency domain.

Fig.2.15 Mayer wavelet


28

Methodology
Daubechies

db1

db2

db5

db3

db6

db8

db4

db7

db9

db10

Fig.2.16 Daubechies wavelet


Daubechies wavelets are a family of orthogonal wavelets defining a discrete wavelet transform and
characterized by a maximal number of vanishing moments for some given support. With each
wavelet type of this class, there is a scaling function (also called father wavelet) which generates an
orthogonal multi resolution analysis.

29

Chapter 2
Bioorthogonal: A biorthogonal wavelet is a wavelet where the associated wavelet transform is
invertible but not necessarily orthogonal. Designing biorthogonal wavelets allows more degrees of
freedom than orthogonal wavelets. One additional degree of freedom is the possibility to construct
symmetric wavelet functions. This family of wavelets exhibits the property of linear phase, which is
needed for signal and image reconstruction. By using two wavelets, one for decomposition and the
other for reconstruction instead of the same single one, interesting properties are derived.

Decomposition

Reconstruction
bior 1.3

Decomposition

Reconstruction
bior 6.8

Fig 2.17 Bioorthogonal Wavelet


Symlets: The symlets are nearly symmetrical wavelets proposed by Daubechies as modifications
to the db family. The properties of the two wavelet families are similar.

sym2

sym8
Fig.2.18 Symlets Wavelet
30

Methodology

2.3.3 Continuous Wavelet Transform


The continuous wavelet transform (CWT) uses to measure the similarity between a signal and an
analyzing

function.

exponentials

In

the

Fourier

transform,

the

analyzing

functions

are

complex

.where x is the independent variable space. The resulting transform is a function of

a single variable, . In the short-time Fourier transform, the analyzing functions are windowed
complex exponentials, ( )

, and the result in a function of two variables. The STFT

coefficients, F(,) represent the match between the signal and a sinusoid with angular frequency
in an interval of a specified length centered at .
In the CWT, the analyzing function is a wavelet, . The CWT compares the signal to shifted and
compressed or stretched versions of a wavelet. Stretching or compressing a function is collectively
referred to as dilation or scaling and corresponds to the physical notion of scale. By comparing the
signal to the wavelet at various scales and positions, you obtain a function of two variables. The twodimensional representation of a one-dimensional signal is redundant. If the wavelet is complexvalued, the CWT is a complex-valued function of scale and position. If the signal is real-valued, the
CWT is a real-valued function of scale and position. For a scale parameter, a>0, and position, b, the
CWT is
( , )=

( )

By continuously varying the values of the scale parameter, a, and the position parameter, b, obtain

the cwt coefficients C(a, b).

( )=2

is the complex conjugate of . The basic function is represented as


(2

The results of the transform are wavelet coefficient that show how well a wavelet function
correlates with the signal analyzed. Hence, sharp transitions in f(x) create wavelet coefficient with
large amplitude and this precisely is the basis of the proposed identification method.
The inverse CWT permit to recover the signal from its coefficients C(a,b) and its defined as
f(x) =

where the constant

( , )

( )

depends on the wavelet type

31

Chapter 2
CWT is highly redundant, not necessary to use the full domain of C(a,b) to reconstruct f(x).
C(a,b) can also be represented as
( , ) = ( ),

( )

Therefore, CWT is a collection of inner products of a signal f(x) and the translated and dilated
wavelets , ( ).

2.3.4 Discrete Wavelet Transform

The Discrete Wavelet Transform (DWT), which is based on sub-band coding, is found to yield a
fast computation of Wavelet Transform. It is easy to implement and reduces the computation time
and resources required.
In CWT, the signals are analyzed using a set of basic functions which relate to each other by
simple scaling and translation. In the case of DWT, a time-scale representation of the digital signal is
obtained using digital filtering techniques. The signal to be analyzed is passed through filters with
different cutoff frequencies at different scales.
DWT adopts dyadic scales and translations in order to reduce the amount of computation, which
results in better efficiency of calculation. Filters of different cutoff frequencies are used for the
analysis of the signal at different scales. The signal is passed through a series of high-pass filters to
analyze the high frequencies and trough a series of low-pass filters to analyze the low frequencies.
DWT signal can be represent as
( )=
The coefficient

( )

( )

( ) are known as the level J detail coefficients.

At level j DWT function also be represent as


( )=

( )

( )

The approximation at level J is define as


( )=

( )

( )

The original function can be expressed as the sun of its approximation at level J plus all its detail
up to the same level. Finally the signal f(x) is defined as
32

Methodology
( )=

( )+

( )

If f(x) is a response signal, typically the deflection curve the signal

( ) contain the information

necessary to detect the cracks in the structure.

2.3.5 Wavelet Methodology for Crack Detection in Present Work


In the present work crack detection of cantilever beam is done by continuous and discrete wavelet
analysis. Wavelet analysis is detected small discontinuities of structural response in the presence of
cracks. The structural response in terms of mode shape difference between cracked free and cracked
beams is treated as input signal for the CWT and DWT analysis. Biorthonal6.8 wavelet is used for
the analysis. Biorthogonal wavelet is preferred for this analysis because in this wavelet analysis
scaling function and fast algorithm are available.
The Wavelet coefficients of the CWT or the DWT for different crack location and depth are
plotted. From the discontinuity of CWT and DWT plotted curve, the crack location of beam is
determined. The maximum wavelet coefficient of CWT is used for prediction of crack depth.
For static analysis relative deflection shape of cracked and cracked free beam is used as input
signal for wavelet analysis. Similar procedure is followed for detection of crack and its location.

33

Chapter 3
RESULTS AND DISCUSSION
This section is divided in three parts. In the first part, the results of the numerical experiment are
presented. In the second and third part, identification of crack depth and crack location is described
using Neural Network and Wavelet analysis respectively.

3.1. Finite Element Simulation Result


As mentioned in Section 2.1, a numerical experiment is performed on crack free and cracked
cantilever beam using ANSYS software. The relevant dimensions and other properties of the beam
are already presented in Table 2.1.
The first, second and third natural frequencies corresponding to various crack locations and
depths are obtained from the modal analysis of the beam in ANSYS environment. The first three
mode shapes under transverse vibration of cracked and crack free beams are captured. At the same
time, first three natural frequencies of cracked free beam under transverse vibration are also obtained.
They are 46.013 Hz, 286.88 Hz and 796.79 Hz respectively. The relative frequencies and average
mode shape deviation for different cracked depth and location are obtained from Equation (1) and (2),
respectively.

Relative Natural Frequency =

Average Mode Shape Deviation =

( 1)

Total Mode Shape Deviation of 60 points


60

(2)

Table 3.1 represents the values of first three relative natural frequencies of the cantilever beam
having various depths at crack location 50 mm from fixed end as obtained from the numerical
experiment. The graphical representation of the same result is also shown in Figure 3.1. Similarly,
Table 3.2 represents the results obtained from the ANSYS analysis with fixed crack depth (3 mm) but
various crack locations, corresponding graphical representations are shown in Figure 3.2 to 3.4. Table
3.3 exhibits average relative mode shape deviation at different depth, crack location being fixed at 50
mm from fixed end. Figure 3.5 represents the values of the same table in graphical from. The values

34

Result and Discussion


of average mode shape deviation at different crack location from fixed end with crack depth of 3mm
are represented in Table 3.4. Figure 3.6 to 3.8 shows them in graphical form.

Table 3.1 Different natural frequencies with relative natural frequencies at different depth with
crack location 50 mm from fixed end.

Dept
h
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
7.5
8
8.5
9
9.5
10

First
Natural
Frequencie
s
45.947
45.873
45.772
45.645
45.497
45.321
45.123
44.89
44.608
44.324
44.002
43.621
43.231
42.717
42.225
41.696
41.066
40.402
39.668

First
Relative
Natural
Frequencies
0.998565623
0.996957381
.994762349
0.99200226
0.988785778
0.984960772
0.98065764
0.975593854
0.969465151
0.963292982
0.95629496
0.948014692
0.939538826
0.92836807
0.91767544
0.906178689
0.892486906
0.878056202
0.862104188

Second
Natural
Frequencie
s
286.69
286.47
286.18
285.82
285.41
284.91
284.37
283.73
282.96
282.2
281.36
280.37
279.39
278.11
276.92
275.68
274.24
272.77
271.18

35

Second
Third
Relative
Natural
Natural
Frequencie
Frequencies s
0.999337702
796.62
0.998570831
796.42
0.997559955
796.14
0.996305075
795.81
0.994875906
795.42
0.993133017
794.96
0.991250697
794.45
0.989019799
793.85
0.98633575
793.13
0.983686559
792.43
0.980758505
791.65
0.977307585
790.73
0.973891523
789.83
0.969429727
788.65
0.965281651
787.58
0.960959286
786.46
0.955939766
785.15
0.950815672
783.83
0.945273285
782.4

Third
Relative
Natural
Frequencies
0.999786644
0.999535637
0.999184227
0.998770065
0.998280601
0.997703284
0.997063216
0.996310195
0.995406569
0.994528044
0.993549116
0.992394483
0.991264951
0.989784008
0.98844112
0.98703548
0.985391383
0.983734736
0.981940034

Chapter 3
Table 3.2 Different natural frequencies with relative natural frequencies at different location
from fixed end with fixed crack depth 3mm.

Location
from
fixed end

First
Natural
Frequencies

First Relative
Natural
Frequencies

Second
Natural
Frequencies

25
40
50
75
80
100
120
125
150
160
175
200
225
240
250
275
280
300
320
325
350
360
375
400
425
440
450
475
480
500
520
525
550
560
575

45.44
45.469
45.497
45.56
45.569
45.621
45.666
45.672
45.725
45.747
45.771
45.814
45.85
45.875
45.887
45.913
45.92
45.938
45.955
45.96
45.976
45.982
45.991
46.002
46.01
46.014
46.016
46.021
46.022
46.025
46.027
46.028
46.031
46.032
46.033

0.987546998
0.988177254
0.988785778
0.990154956
0.990350553
0.991480669
0.992458653
0.992589051
0.993740899
0.994219025
0.994740617
0.995675135
0.996457523
0.997000848
0.997261643
0.997826701
0.997978832
0.998370026
0.998739487
0.998848152
0.999195879
0.999326277
0.999521874
0.999760937
0.999934801
1.000021733
1.000065199
1.000173864
1.000195597
1.000260796
1.000304262
1.000325995
1.000391194
1.000412927
1.00043466

284.35
285
285.41
286.2
286.32
286.69
286.87
286.88
286.83
286.75
286.58
286.2
285.76
285.55
285.4
285.06
285.02
284.86
284.82
284.85
284.93
285.02
285.18
285.5
285.82
286.03
286.16
286.45
286.5
286.67
286.8
286.82
286.91
286.93
286.97
36

Second
Relative
Natural
Frequencies

0.991180982
0.993446737
0.994875906
0.997629671
0.998047964
0.999337702
0.999965142
1
0.999825711
0.999546849
0.998954267
0.997629671
0.996095929
0.995363915
0.994841049
0.993655884
0.993516453
0.992958728
0.992819297
0.992923871
0.993202733
0.993516453
0.994074177
0.995189626
0.996305075
0.997037089
0.99749024
0.998501115
0.998675404
0.999267987
0.999721138
0.999790853
1.000104573
1.000174289
1.00031372

Third
Natural
Frequencies

791.93
794.25
795.42
796.8
796.84
796.36
795.21
794.84
793.22
792.78
792.27
792.41
793.47
794.46
795.07
796.34
796.52
796.79
796.33
796.11
794.53
793.79
792.69
791.24
790.53
790.76
791.07
792.38
792.872
794.07
795.32
795.58
796.52
796.73
796.95

Third
Relative
Natural
Frequencies

0.993900526
0.996812209
0.998280601
1.00001255
1.000062752
0.999460335
0.998017043
0.99755268
0.995519522
0.994967306
0.994327238
0.994502943
0.995833281
0.997075767
0.997841338
0.999435234
0.99966114
1
0.999422684
0.999146576
0.997163619
0.996234893
0.994854353
0.993034551
0.992143476
0.992432134
0.992821195
0.994465292
0.99508277
0.996586303
0.998155097
0.998481407
0.99966114
0.999924698
1.000200806

Result and Discussion

Table 3.3 Average relative mode shape deviation at different depth with crack location 50 mm

from fixed end.

Depth
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
7.5
8
8.5
9
9.5
10

Second Average
First Average Relative
Relative Mode Shape
Third Average Relative
Mode Shape Deviation
Deviation
Mode Shape Deviation
0.00156
0.002198
0.001747
0.005358
0.007738
0.006018
0.011498
0.016552
0.01262
0.019617
0.028391
0.021645
0.029898
0.04329
0.033008
0.041879
0.060564
0.046026
0.056007
0.080872
0.061405
0.071769
0.103513
0.078484
0.090282
0.129908
0.098344
0.112653
0.161687
0.122306
0.135015
0.192954
0.145216
0.16003
0.228719
0.172146
0.189606
0.269637
0.20214
0.219432
0.310989
0.232523
0.258379
0.36471
0.271678
0.295064
0.41444
0.307388
0.333857
0.467682
0.346535
0.380022
0.528162
0.388475
0.427147
0.591732
0.434575
0.478778
0.659398
0.482486

37

Chapter 3
Table 3.4 Different average mode shape deviation at different location from fixed end with
fixed crack depth 3mm.
Location
From Fixed
End
25
40
50
75
80
100
120
125
150
160
175
200
225
240
250
275
280
300
320
325
350
360
375
400
425
440
450
475
480
500
520
525
550
560
575

Average First Mode Average Second Mode


Average Third Mode
Shape Deviation
Shape Deviation
Shape Deviation
0.060627
0.121252
0.148608
0.049444
0.08101
0.074377
0.041879
0.060564
0.046026
0.026441
0.0314
0.007746
0.024134
0.028467
0.001652
0.015975
0.01822
0.038314
0.010873
0.00744
0.081137
0.010404
0.004322
0.091617
0.010605
0.01529
0.115215
0.012449
0.023657
0.111895
0.015377
0.036247
0.102315
0.018858
0.0521
0.091285
0.020893
0.058953
0.092657
0.020636
0.055798
0.087874
0.020703
0.053642
0.083405
0.020204
0.04888
0.051732
0.019806
0.047914
0.04114
0.018597
0.048239
0.006268
0.016632
0.052183
0.054078
0.01586
0.053349
0.06356
0.013363
0.065821
0.097464
0.012225
0.072119
0.101126
0.010481
0.078754
0.099961
0.007723
0.081787
0.106798
0.005291
0.076892
0.139548
0.003927
0.068109
0.154691
0.003202
0.061869
0.166645
0.003263
0.043466
0.165261
0.003493
0.039473
0.159242
0.004192
0.025545
0.126191
0.004962
0.013121
0.078956
0.005079
0.010671
0.067342
0.005921
0.00728
0.020748
0.006273
0.008608
0.008899
0.006973
0.01109
0.011036
38

Result and Discussion

Fig 3.1 Relative Natural Frequencies Vs Depth

Fig.3.2 Relative Natural Frequencies vs. Location


from fixed end for first frequency.

Figure 3.1 shows that with increase of crack depth relative natural frequencies decrease whereas
Figure 3.5 exhibits that with increase of crack depth average mode shape deviations increases. First
relative natural frequency is monotonously increased as the crack location moves from the fixed end
to the free end when the crack depth is kept constant (Figure 3.2). On the other hand, the second and
third relative frequencies oscillate under the same situation (Figure 3.3 and 3.4). Remarkable changes
are observed in average mode deviations at different location as observed from Figure 3.6 to 3.8.

Fig.3.3 Relative Natural Frequencies vs. Location


from fixed end for second frequency

Fig. 3.4 Relative Natural Frequencies vs.


Location from fixed end for third frequencies

39

Chapter 3

Fig 3.5 Average Mode Shape Deviation vs


Depth

Fig.3.6 Average Mode Shape Deviation vs. Crack


Location form fixed end for first natural frequencies

Fig.3.7 Average Mode Shape Deviation vs. Crack

Fig 3.8Average Mode Shape Deviation vs. Crack

Location from fixed end for second natural

Location from fixed end for third natural

frequencies

frequencies

3.2 Identification of Crack depth and location by Artificial Neural Network


Six variable, three relative natural frequencies and three average mode shape deviation are used as a
input to the network .Crack depth and location are used as a target output of the network. All values
of variable are taken from finite element analysis performed by ANSYS software. The number of

40

Result and Discussion


hidden layers and number of neurons in hidden layer are varied to obtain desired result having least
deviation from the target values.
For the identification of crack location and crack depth by Artificial Neural Network, feedforward
backpropagation algorithm is used. It comprises of four functions as describe in Table 3.5.
Table 3.5 Different Function of Network
Type of function
Training Function
Adaption Learning Function
Performance Function
Transfer Function

Name of the function


TRAINLM
LEARNGDM
MSE (Mean Square Error)
TANSIG

Various terms mentioned in the right column is explained below:


TRAINLM is a network training function that updates weight and bias value according to LevenbergMarquardt optimization. In Lavenberg-Marquardent algorithm divides the entire set of input and
target vector into three groups- 60% for training, 20% for validation and the rest for testing the
results. However formation of group is achieved by random selection of vectors
LEARNGDM is the gradient decent with momentum weight and bias learning function
MSE is a network performance function -lower the value of MSE lower is the error.
TANSIG is a neural transfer function that calculates a layers output from its net output.
Before proceeding further all the input and target variables are normalized between 0 to 1 based on
the following equation.
Normalized (value) =
Then the normalized input and target variables with various combinations are entered in a MS
Excel sheet. The Excel sheet is then imported to MATLAB workspace. Finally neural network
analysis is done by means Neural Network Toolbox provided in MATLAB.
In the present study, target vectors are depth of crack and its location whereas input variable being
first three relative mode shapes values as obtained from the modal analysis performed by ANSYS. In
order to identify the crack location and depth, neural network system has to be trained first using
known sample data. For each set of target (either depth or location) it is being trained in three stages
41

Chapter 3
of input values keeping other target value unchanged. For each case of target, the network is trained
using three sets of input values, e.g. first three relative natural frequencies, first three average mode
shape deviation and both first three relative natural frequencies along with first three average mode
shape deviation. Accuracy of training is measured from the value of Mean Square Error (MSE) and
Regression coefficient (R) obtained from performance curve and regression plot produced after
training operation. Once the training is complete, output depth are obtained by simulation data of
relative natural frequency. Lastly output depth of the ANN results is compared with FE model depth
to estimate error in the ANN process. The objective of the present study is to keep the error minimum
by proper choice of layer and neuron number while ANN network is designed.
In the following sections, six different neural networks are performed for identifying crack depths
and their locations. Modal analysis using FEM are conducted to generate training data. For training
ten damage cases are used to determine depth and twenty three damage cases are used to determine
crack location.

There are two types of output target data; locations and depths are used in present

study. For location output data 23 patterns are used for training and 35 patterns are simulated. For
depth output data 10 patterns are used for training and 19 patterns are simulated.

3.2.1 Prediction of Depth of crack at a particular location using First Three


Relative Natural Frequencies as input.
Input sample data for training are presented in Table 3.6 where target is the depth of the crack having
location fixed at 50 mm from fixed end.
Table 3.6 Natural Frequencies with cracked depth at location 50 mm from fixed end for training.
Input ( Relative Natural Frequencies)
First
Second
0.998565623
0.999337702
0.99476235
0.997559955
0.988785778
0.994875906
0.98065764
0.991250697
0.969465151
0.98633575
0.95629496
0.980758505
0.939538826
0.973891523
0.91767544
0.965281651
0.892486906
0.955939766
0.862104188
0.945273285

42

Third
0.999786644
0.999184227
0.998280601
0.997063216
0.995406569
0.993549116
0.991264951
0.98844112
0.985391383
0.981940034

Target
Depth
1
2
3
4
5
6
7
8
9
10

Result and Discussion


The performance curve and regression plot for relative natural frequencies and depth network are
shown in Figure 3.9 and 3.10 respectively.

Fig 3.9 Performance curve for Relative natural frequencies and Depth network

Fig 3.10 Regression plot for Relative natural frequencies and Depth network
43

Chapter 3

It has been observed that best MSE value of validation performance is.00041412 at 20 epochs and R
value is .99996. Neural network model for this analysis is (3-1-3-1). i.e. 3 input, 1 output, 2 hidden
layers in which first layer having one neuron and second layer having three neurons. After the
network is trained, we can simulate it with all combination of input values. The simulated data is
shown in Table 3.7. Table 3.8 shows the value of depth obtained by ANN analysis and their deviation
from the assumed FE model depth values.
Table 3.7 Simulation data of Natural frequencies and depth at location 50 mm from fixed end.
Input
Normalized
Relative First
Natural
Frequencies
1
0.988211968
0.972126671
0.951900604
0.928330069
0.900300244
0.868766691
0.831659025
0.786747601
0.741517656
0.690235817
0.629557617
0.567446073
0.485586243
0.407230141
0.322981406
0.222647373
0.116898488
0

Normalized Relative
Second Natural
Frequencies
1
0.985810247
0.967112834
0.943902253
0.917467979
0.885231061
0.850415188
0.809151932
0.759507077
0.710506961
0.656348937
0.592519838
0.529335477
0.446808965
0.370085099
0.29013754
0.197295214
0.102518673
0

44

Output
Normalized Relative
Third
Natural
Frequencies
1
0.985915658
0.966225512
0.943019269
0.915593708
0.883245612
0.847381417
0.805188248
0.754556444
0.705331079
0.650479959
0.585783765
0.522494011
0.439514111
0.364269625
0.285509041
0.193387288
0.100562314
0

Depth
0.126870949
0.141410936
0.171269008
0.217542097
0.282055378
0.353808288
0.415362901
0.464963033
0.507015506
0.544012192
0.590052763
0.649720757
0.702194244
0.756723979
0.798333287
0.84459635
0.901793715
0.947403026
0.975609502

Result and Discussion


Table 3.8 Comparing target result of depth with ANN result for frequency depth network
Target (mm)
ANN (mm)
Error (mm)
1
1.268709487
-0.268709487
1.5
1.414109356
0.085890644
2
1.712690077
0.287309923
2.5
2.175420974
0.324579026
3
2.82055378
0.17944622
3.5
3.538082882
-0.038082882
4
4.153629007
-0.153629007
4.5
4.649630334
-0.149630334
5
5.070155058
-0.070155058
5.5
5.440121916
0.059878084
6
5.900527626
0.099472374
6.5
6.497207565
0.002792435
7
7.021942442
-0.021942442
7.5
7.567239788
-0.067239788
8
7.983332871
0.016667129
8.5
8.445963499
0.054036501
9
9.017937154
-0.017937154
9.5
9.474030256
0.025969744
10
9.75609502
0.24390498
Maximum Error 0.3246 mm
Average Error 0.1141 mm

3.2.2 Prediction of Depth of crack at a particular location using First Three


Average Mode Shape Deviations as input.
Input sample data for training are presented in Table 3.9 where target is the depth of the crack having
location fixed at 50 mm from fixed end.
It has been observed that best MSE value of validation performance is.00055188 at 95 epochs and R
value is .99879. Neural network model for this analysis is (3-2-1). i.e. 3 input, 1 output, 1 hidden
having two neuron. After the network is trained, we can simulate it with all combination of input
values. The simulated data is shown in Table 3.10. Table 3.11 shows the value of depth obtained by
ANN analysis and their deviation from the assumed FE model depth values

45

Chapter 3
Table 3.9 Average Mode Shape Deviation with cracked depth at location 50 mm from fixed end
for training.

Input(Average Mode Shape Deviation)


Target
First
Second
Third
Depth
0.005358
0.007738
0.006018
1
0.019617
0.028391
0.021645
2
0.041879
0.060564
0.046026
3
0.071769
0.103513
0.078484
4
0.112653
0.161687
0.122306
5
0.16003
0.228719
0.172146
6
0.219432
0.310989
0.232523
7
0.295064
0.41444
0.307388
8
0.380022
0.528162
0.388475
9
0.478778
0.659398
0.482486
10

The performance curve and regression plot for average mode shape deviation and depth network are
shown in Figure 3.11 and 3.12 respectively

Fig 3.11 Performance curve for Average Mode Shape Deviation and Depth network

46

Result and Discussion

Fig 3.12 Regression for Average Mode Shape Deviation and Depth network

Table 3.10 Simulation data of Mode Shape Deviation and Depth at Crack location 50 mm from fixed
end.

Normalized Average
First Mode Shape
Deviation
0
0.012969456
0.030119133
0.051835579
0.077142917
0.106985341
0.140279245
0.179384056
0.22663808
0.273873094
0.32671201
0.389185079
0.452186219
0.534453551
0.611942884
0.693884922
0.791398758
0.890940391
1

Input
Normalized Average
Second Mode Shape
Deviation
0
0.013525458
0.031692907
0.054556057
0.081063745
0.112227235
0.146970813
0.187475064
0.236241291
0.284221833
0.339104748
0.40189516
0.465351564
0.547788724
0.624101525
0.70580364
0.798612774
0.896163644
1
47

Output
Normalized Average
Third Mode Shape
Deviation
0
0.013856125
0.032797586
0.056645987
0.083967864
0.116244952
0.152089962
0.19377167
0.24406256
0.292145538
0.348665598
0.41161631
0.475383447
0.557561053
0.632508374
0.714669191
0.802691891
0.899445503
1

Depth
0.131103118
0.14841815
0.175580644
0.21508305
0.275025008
0.33905225
0.406008699
0.466190873
0.516321736
0.562642068
0.596402523
0.633277096
0.671062663
0.731754985
0.800693041
0.863347343
0.945569702
0.980956187
0.995953402

Chapter 3

Table 3.11 Comparing target result of depth with ANN result for mode shape deviation depth network

Target (mm)

ANN (mm)
Error (mm)
1
1.311031
-0.31103
1.5
1.484182
0.015818
2
1.755806
0.244194
2.5
2.150831
0.349169
3
2.75025
0.24975
3.5
3.390523
0.109477
4
4.060087
-0.06009
4.5
4.661909
-0.16191
5
5.163217
-0.16322
5.5
5.626421
-0.12642
6
5.964025
0.035975
6.5
6.332771
0.167229
7
6.710627
0.289373
7.5
7.31755
0.18245
8
8.00693
-0.00693
8.5
8.633473
-0.13347
9
9.455697
-0.4557
9.5
9.809562
-0.30956
10
9.959534
0.040466
Maximum Error 0.4557 mm Average Error 0.2796 mm

3.2.3 Prediction of Depth of crack at a particular location using First Three


Relative Natural Frequencies and First Three Average Mode Shape
Deviations as input
Input sample data for training are presented in Table 3.12 where target is the depth of the crack
having location fixed at 50 mm from fixed end.

48

Result and Discussion


Table 3.12 Relative Natural Frequencies and Mode Shape Deviation with cracked depth at location 50
mm from fixed end for training.
Input

Target

Average Mode Shape Deviation


First
0.998565623
0.99476235
0.988785778
0.98065764
0.969465151
0.95629496
0.939538826
0.91767544
0.892486906
0.862104188

Second
0.999338
0.99756
0.994876
0.991251
0.986336
0.980759
0.973892
0.965282
0.95594
0.945273

Third
0.999787
0.999184
0.998281
0.997063
0.995407
0.993549
0.991265
0.988441
0.985391
0.98194

Relative Natural Frequencies


First
0.005358
0.019617
0.041879
0.071769
0.112653
0.16003
0.219432
0.295064
0.380022
0.478778

Second
0.007738
0.028391
0.060564
0.103513
0.161687
0.228719
0.310989
0.41444
0.528162
0.659398

Third
0.006018
0.021645
0.046026
0.078484
0.122306
0.172146
0.232523
0.307388
0.388475
0.482486

Depth
1
2
3
4
5
6
7
8
9
10

The performance curve and regression plot for relative natural frequency with average mode shape
deviation and depth network are shown in Figure 3.13 and 3.14 respectively

Fig 3.13 Performance curve for Relative Natural Frequencies and Average Mode Shape Deviation
with Depth network
49

Chapter 3

Fig 3.14 Regression for Relative Natural Frequencies and Average Mode Shape Deviation with
Depth network

It has been observed that best MSE value of validation performance is. .000019627 at 31 epochs
and R value is .99913. Neural network model for this analysis is (6-3-1). i.e. 6 input, 1 output, 1
hidden having three neurons. After the network is trained, we can simulate it with all combination of
input values. The simulated data is shown in Table 3.10. Table 3.11 shows the value of depth
obtained by ANN analysis and their deviation from the assumed FE model depth values.

50

Result and Discussion


Table 3.13 Simulation data of Relative Natural Frequencies with Average Mode Shape Deviation and
Depth at location of Crack 50 mm from fixed end.
Input
Normalized
Relative
First
Natural
Frequencies
1
0.988211968
0.972126671
0.951900604
0.928330069
0.900300244
0.868766691
0.831659025
0.786747601
0.741517656
0.690235817
0.629557617
0.567446073
0.485586243
0.407230141
0.322981406
0.222647373
0.116898488
0

Normalized
Relative
Second
Natural
Frequencies
1
0.985810247
0.967112834
0.943902253
0.917467979
0.885231061
0.850415188
0.809151932
0.759507077
0.710506961
0.656348937
0.592519838
0.529335477
0.446808965
0.370085099
0.29013754
0.197295214
0.102518673
0

Normalized
Relative
Third
Natural
Frequencies
1
0.985915658
0.966225512
0.943019269
0.915593708
0.883245612
0.847381417
0.805188248
0.754556444
0.705331079
0.650479959
0.585783765
0.522494011
0.439514111
0.364269625
0.285509041
0.193387288
0.100562314
0

Output
Normalized
Average
First Mode
Shape
Deviation
0
0.012969456
0.030119133
0.051835579
0.077142917
0.106985341
0.140279245
0.179384056
0.22663808
0.273873094
0.32671201
0.389185079
0.452186219
0.534453551
0.611942884
0.693884922
0.791398758
0.890940391
1

51

Normalized
Average
Second
Mode
Shape
Deviation
0
0.013525458
0.031692907
0.054556057
0.081063745
0.112227235
0.146970813
0.187475064
0.236241291
0.284221833
0.339104748
0.40189516
0.465351564
0.547788724
0.624101525
0.70580364
0.798612774
0.896163644
1

Normalized
Average
Third Mode
Shape
Normalized
Deviation
Depth
0
0.114143
0.013856125
0.13962
0.032797586
0.192303
0.056645987
0.254352
0.083967864
0.304858
0.116244952
0.348735
0.152089962
0.388134
0.19377167
0.435485
0.24406256
0.496043
0.292145538
0.54674
0.348665598
0.607344
0.41161631
0.668895
0.475383447
0.716369
0.557561053
0.771355
0.632508374
0.803911
0.714669191
0.846957
0.802691891
0.901003
0.899445503
0.961898
1
0.982508

Chapter 3
Table 3.14 Comparing target result of depth with ANN result for relative natural frequency with
average mode shape deviation depth network
Target
(mm)

ANN(mm)
Error (mm)
1
1.141426
-0.14143
1.5
1.396201
0.103799
2
1.92303
0.07697
2.5
2.543521
-0.04352
3
3.048577
-0.04858
3.5
3.487354
0.012646
4
3.881336
0.118664
4.5
4.354852
0.145148
5
4.960431
0.039569
5.5
5.467401
0.032599
6
6.073442
-0.07344
6.5
6.688948
-0.18895
7
7.163691
-0.16369
7.5
7.713551
-0.21355
8
8.039112
-0.03911
8.5
8.469566
0.030434
9
9.010026
-0.01003
9.5
9.618975
-0.11898
10
9.825078
0.174922
Maximum Error 0.2136 mm Average Error 0.0935 mm

3.2.4 Prediction of Location of crack for particular depth using First Three
Relative Natural Frequencies as input.
Input sample data for training are presented in Table 3.15 where target is the location of the crack
having fixed 3 mm depth.

52

Result and Discussion


Table 3.15

Natural Frequencies with cracked location from fixed end with fixed 3 mm depth for

training.
Input ( Relative Natural Frequencies)
Target
First
Second
Third
Depth
0.987546998
0.991180982
0.993900526
0.988785778
0.994875906
0.998280601
0.990154956
0.997629671
1.00001255
0.991480669
0.999337702
0.999460335
0.992589051
1
0.99755268
0.993740899
0.999825711
0.995519522
0.994740617
0.998954267
0.994327238
0.995675135
0.997629671
0.994502943
0.996457523
0.996095929
0.995833281
0.997261643
0.994841049
0.997841338
0.997826701
0.993655884
0.999435234
0.998370026
0.992958728
1
0.998848152
0.992923871
0.999146576
0.999195879
0.993202733
0.997163619
0.999521874
0.994074177
0.994854353
0.999760937
0.995189626
0.993034551
0.999934801
0.996305075
0.992143476
1.000065199
0.99749024
0.992821195
1.000173864
0.998501115
0.994465292
1.000260796
0.999267987
0.996586303
1.000325995
0.999790853
0.998481407
1.000391194
1.000104573
0.99966114
1.00043466
1.00031372
1.000200806

25
50
75
100
125
150
175
200
225
250
275
300
325
350
375
400
425
450
475
500
525
550
575

The performance curve and regression plot for relative natural frequency and location network are
shown in Figure 3.15 and 3.16 respectively

53

Chapter 3

Fig.3.15 Performance curve for Relative Natural Frequencies and Location network

Fig 3.16 Regression for Relative Natural Frequencies and Location network

54

Result and Discussion


Table 3.16 Simulation data of Natural frequencies and location from fixed end for 3 mm crack depth.
Input
Normalized Relative
First Natural
Frequencies
0
0.04890241
0.09611871
0.202355385
0.217532053
0.305219468
0.381102808
0.391220587
0.480594298
0.51769282
0.558163934
0.630674681
0.691381353
0.733538764
0.753774321
0.797618028
0.809422103
0.839775439
0.868442479
0.876873961
0.903854704
0.913972483
0.929149151
0.947698411
0.961188783
0.967933969
0.971306562
0.979738044
0.98142434
0.98648323
0.991542119
0.989855822
0.996601008
0.998287305
1

Normalized Relative
Second Natural
Frequencies
0
0.248082482
0.40456655
0.706084632
0.751884847
0.893102176
0.961802499
0.965619183
0.94653576
0.916002284
0.851118646
0.706084632
0.53815051
0.458000134
0.400749866
0.27098259
0.255715852
0.194648898
0.17938216
0.190832214
0.22136569
0.255715852
0.316782805
0.438916711
0.561050618
0.641200994
0.690817894
0.801501746
0.820585169
0.885468807
0.942719076
0.935085707
0.977069237
0.984702606
1
55

Output
Normalized Relative
Third
Natural
Frequencies
0.218109439
0.57945011
0.761677949
0.976613348
0.98284336
0.908083221
0.728970388
0.671342781
0.419027312
0.350497185
0.271064537
0.292869578
0.457964885
0.612157671
0.707165348
0.904968215
0.933003267
0.975055845
0.903410712
0.869145648
0.623060191
0.507804977
0.336479659
0.110641739
0
0.035881601
0.08416419
0.288197069
0.364826212
0.551415058
0.786597995
0.74610292
0.933003267
0.965710828
1

Depth
0.047884999
0.065393002
0.082807912
0.122880704
0.128988608
0.168170735
0.204544428
0.208468874
0.252215691
0.271422729
0.291446647
0.332527071
0.370688518
0.402131883
0.418447744
0.457466909
0.470106134
0.5016026
0.534053729
0.54605147
0.574788236
0.588556147
0.612157563
0.651126134
0.693796524
0.729213856
0.751790022
0.809212187
0.821986114
0.854408953
0.88268662
0.877678727
0.898489669
0.902062475
0.90621364

Chapter 3
Table 3.17 Comparing target result of location from fixed end with ANN result for relative natural
frequency location network
Target (mm)

Ann (mm)
25

40
50
75
80
100
120
125
150
160
175
200
225
240
250
275
280
300
320
325
350
360
375
400
425
440
450
475
480
500
520
525
550
560
575

Maximum Error 31.27 mm

Error(mm)
28.731
-3.731
39.2358
0.764199
49.68475
0.315253
73.72842
1.271577
77.39316
2.606835
100.9024
-0.90244
122.7267
-2.72666
125.0813
-0.08132
151.3294
-1.32941
162.8536
-2.85364
174.868
0.132012
199.5162
0.483757
222.4131
2.586889
241.2791
-1.27913
251.0686
-1.06865
274.4801
0.519855
282.0637
-2.06368
300.9616
-0.96156
320.4322
-0.43224
327.6309
-2.63088
344.8729
5.127058
353.1337
6.866312
367.2945
7.705462
390.6757
9.32432
416.2779
8.722086
437.5283
2.471686
451.074
-1.07401
485.5273
-10.5273
493.1917
-13.1917
512.6454
-12.6454
529.612
-9.61197
526.6072
-1.60724
539.0938
10.9062
541.2375
18.76252
543.7282
31.27182
Average Error 5.1 mm

56

Result and Discussion


It has been observed that best MSE value of validation performance is. .000014395 at 14 epochs
and R value is .99887. Neural network model for this analysis is (3-3-1). i.e. 3 input, 1 output, 1
hidden having three neurons. After the network is trained, we can simulate it with all combination of
input values. The simulated data is shown in Table 3.16. Table 3.17 shows the value of depth
obtained by ANN analysis and their deviation from the assumed FE model depth values.

3.2.5 Prediction of Location of crack for particular depth using First Three
Average Mode Shape Deviations as input.
Input sample data for training are presented in Table 3.18 where target is the location of the crack
having fixed 3 mm depth.
Table 3.18 Mode Shape deviations with cracked location from fixed end with fixed 3 mm depth for
training
Input ( Average Mode Shape Deviation)
Target
First
Second
Third
Depth
0.060627
0.121252
0.148608
25
0.041879
0.060564
0.046026
50
0.026441
0.0314
0.007746
75
0.015975
0.01822
0.038314
100
0.010404
0.004322
0.091617
125
0.010605
0.01529
0.115215
150
0.015377
0.036247
0.102315
175
0.018858
0.0521
0.091285
200
0.020893
0.058953
0.092657
225
0.020703
0.053642
0.083405
250
0.020204
0.04888
0.051732
275
0.018597
0.048239
0.006268
300
0.01586
0.053349
0.06356
325
0.013363
0.065821
0.097464
350
0.010481
0.078754
0.099961
375
0.007723
0.081787
0.106798
400
0.005291
0.076892
0.139548
425
0.003202
0.061869
0.166645
450
0.003263
0.043466
0.165261
475
0.004192
0.025545
0.126191
500
0.005079
0.010671
0.067342
525
0.005921
0.00728
0.020748
550
0.006973
0.01109
0.011036
575
57

Chapter 3

The performance curve and regression plot for average mode shape deviation and location network
are shown in Figure 3.15 and 3.16 respectively

3.17 Performance curve for Mode Shape Deviation and Location network

3.18 Regression for Mode Shape Deviation and Location network

58

Result and Discussion


Table 3.19 Simulation data of Mode shape deviation and location from fixed end for 3 mm crack
depth.

Normalized Average
First Mode Shape
Deviation
1
0.805259034
0.673521985
0.404684371
0.364510231
0.222429256
0.133582934
0.12541576
0.128915977
0.161027427
0.212015673
0.27263387
0.308071397
0.303595995
0.304762734
0.296073139
0.28914236
0.268088811
0.233870266
0.220426643
0.17694384
0.157126687
0.126756639
0.078728777
0.036377884
0.012625163
0
0.001062255
0.005067479
0.017239878
0.032686112
0.030648672
0.047348716
0.05347845
0.065668263

Input
Normalized Second
Average Mode Shape
Deviation
1
0.655845378
0.480988626
0.231574446
0.206491063
0.118857436
0.026665526
0
0.093799709
0.165355341
0.273026597
0.408603438
0.467211152
0.440229197
0.421790815
0.381065595
0.372804242
0.375583683
0.409313264
0.419285042
0.519683
0.579808432
0.636551783
0.662490379
0.620627726
0.54551441
0.492149149
0.334764389
0.300615753
0.181501753
0.054297443
0.07525015
0.025297186
0.036654409
0.057880783
59

Output
Normalized Third
Average Mode Shape
Deviation
Location
0.916315931 0.0518652
0.440776275 0.0579487
0.27668556 0.0726661
0.037997967 0.1568779
0 0.2042596
0.228598864 0.2987911
0.481747711 0.2774727
0.560959489 0.2000622
0.708100289 0.1928135
0.66816774 0.2236895
0.627664815 0.2740666
0.558889367 0.3629567
0.56744421 0.3617726
0.522579746 0.3588123
0.509755139 0.3427681
0.312264227 0.3970693
0.239331366 0.4312363
0.028782182 0.5318792
0.317746814 0.5289886
0.386015451 0.5357272
0.597417335 0.5709642
0.602898305 0.5948169
0.6129869 0.6180335
0.655617701 0.6633266
0.85982404
0.702399
0.927548441 0.7446918
1.028782182 0.7522083
1.020152516 0.7534696
0.95513143 0.7671778
0.77653903 0.7911822
0.409597386 0.8881955
0.468528968 0.8757403
0.119069443
0.926014
0.043923076 0.9296989
0.058512131 0.9200483

Chapter 3
Table 3.20 Comparing target result of location from fixed end with ANN result for average mode
shape deviation location network
Target (mm)

Ann (mm)
Error (mm)
25
31.1191067
-6.1191067
40
34.76919888
5.230801119
50
43.59964193
6.400358068
75
94.12674136
-19.1267414
80
122.5557496
-42.5557496
100
179.2746378
-79.2746378
120
166.4836316
-46.4836316
125
120.0373382
4.962661828
150
115.6881005
34.31189955
160
134.2137235
25.78627651
175
164.4399739
10.56002614
200
217.774015
-17.774015
225
217.0635373
7.936462678
240
215.2873993
24.71260075
250
205.6608424
44.33915764
275
238.2415705
36.75842955
280
258.7417604
21.25823963
300
319.127523
-19.127523
320
317.3931605
2.606839503
325
321.43633
3.563670015
350
342.5785255
7.421474532
360
356.8901116
3.109888395
375
370.8200773
4.179922687
400
397.9959457
2.004054296
425
421.4394211
3.560578894
440
446.815058
-6.81505803
450
451.3249957
-1.32499569
475
452.0817365
22.91826349
480
460.306658
19.69334199
500
474.7093327
25.2906673
520
532.9172817
-12.9172817
525
525.444175
-0.44417498
550
555.6083954
-5.60839538
560
557.8193111
2.180688894
575
552.0289764
22.97102361
Maximum Error 79.27 mm
Average Error 17.13 mm

60

Result and Discussion


It has been observed that best MSE value of validation performance is.00056089 at 13 epochs and R
value is .98933. Neural network model for this analysis is (3-3-2-1). i.e. 3 input, 1 output, 2 hidden
layers in which first layer having three neurons and second layer having two neurons. After the
network is trained, we can simulate it with all combination of input values. The simulated data is
shown in Table 3,19. Table 3.20 shows the value of depth obtained by ANN analysis and their
deviation from the assumed FE model depth values.

3.2.6 Prediction of Location of crack for particular depth using First Three
Relative Natural Frequencies and First Three Average Mode Shape
Deviations as input.
Input sample data for training are presented in Table 3.21 where target is the location of the crack
having fixed 3 mm depth.
The performance curve and regression plot for relative natural frequency with average mode shape
deviation and location network are shown in Figure 3.15 and 3.16 respectively
It has been observed that best MSE value of validation performance is. .000010764 at 31 epochs and
R value is .99989. Neural network model for this analysis is (6-3-1). i.e. 6 input, 1 output, 1 hidden
having three neurons. After the network is trained, we can simulate it with all combination of input
values. The simulated data is shown in Table 3.19. Table 3.20 shows the value of depth obtained by
ANN analysis and their deviation from the assumed FE model depth values

61

Chapter 3
Table 3.21 Relative Natural Frequencies and Mode Shape Deviation with cracked location from fixed
end of 3 mm crack depth for training.
Input
Average Mode Shape Deviation Relative Natural Frequencies
First
0.987546998
0.988785778
0.990154956
0.991480669
0.992589051
0.993740899
0.994740617
0.995675135
0.996457523
0.997261643
0.997826701
0.998370026
0.998848152
0.999195879
0.999521874
0.999760937
0.999934801
1.000065199
1.000173864
1.000260796
1.000325995
1.000391194
1.00043466

Second
0.991181
0.994876
0.99763
0.999338
1
0.999826
0.998954
0.99763
0.996096
0.994841
0.993656
0.992959
0.992924
0.993203
0.994074
0.99519
0.996305
0.99749
0.998501
0.999268
0.999791
1.000105
1.000314

Third
0.993901
0.998281
1.000013
0.99946
0.997553
0.99552
0.994327
0.994503
0.995833
0.997841
0.999435
1
0.999147
0.997164
0.994854
0.993035
0.992143
0.992821
0.994465
0.996586
0.998481
0.999661
1.000201

First
0.060627
0.041879
0.026441
0.015975
0.010404
0.010605
0.015377
0.018858
0.020893
0.020703
0.020204
0.018597
0.01586
0.013363
0.010481
0.007723
0.005291
0.003202
0.003263
0.004192
0.005079
0.005921
0.006973

Second
0.121252
0.060564
0.0314
0.01822
0.004322
0.01529
0.036247
0.0521
0.058953
0.053642
0.04888
0.048239
0.053349
0.065821
0.078754
0.081787
0.076892
0.061869
0.043466
0.025545
0.010671
0.00728
0.01109

Target

Third
Location
0.148608
25
0.046026
50
0.007746
75
0.038314
100
0.091617
125
0.115215
150
0.102315
175
0.091285
200
0.092657
225
0.083405
250
0.051732
275
0.006268
300
0.06356
325
0.097464
350
0.099961
375
0.106798
400
0.139548
425
0.166645
450
0.165261
475
0.126191
500
0.067342
525
0.020748
550
0.011036
575

The performance curve and regression plot for relative natural frequency with average mode shape
deviation and location network are shown in Figure 3.15 and 3.16 respectively
It has been observed that best MSE value of validation performance is. .000010764 at 31 epochs and
R value is .99989. Neural network model for this analysis is (6-3-1). i.e. 6 input, 1 output, 1 hidden
having three neurons. After the network is trained, we can simulate it with all combination of input
values. The simulated data is shown in Table 3.19. Table 3.20 shows the value of depth obtained by
ANN analysis and their deviation from the assumed FE model depth values

62

Result and Discussion

Fig 3.19 Performance curve for Relative Natural Frequencies and Average Mode Shape Deviation
with Location network.

Fig 3.20 Regression for Relative Natural Frequencies and Average Mode Shape Deviation with
Location network .

63

Chapter 3
Table 3.22 Simulation data of Relative Natural Frequencies with Average Mode Shape Deviation and
Location of 3 mm Crack Depth.
Output

Input
Normalized
Relative
First
Natural
Frequencies

0.988786
0.990155
0.990351
0.991481
0.992459
0.992589
0.993741
0.994219
0.994741
0.995675
0.996458
0.997001
0.997262
0.997827
0.997979
0.99837
0.998739
0.998848
0.999196
0.999326
0.999522
0.999761
0.999935
1.000022
1.000065
1.000174
1.000196
1.000261
1.000304
1.000326
1.000391
1.000413
1.000435

Normalized
Relative
Second
Natural
Frequencies

Normalized
Relative
Third
Natural
Frequencies

Normalized
Average
First Mode
Shape
Deviation

Normalized
Average
Second Mode
Shape
Deviation

Normalized
Average
Third Mode
Shape
Deviation

0.994876
0.99763
0.998048
0.999338
0.999965
1
0.999826
0.999547
0.998954
0.99763
0.996096
0.995364
0.994841
0.993656
0.993516
0.992959
0.992819
0.992924
0.993203
0.993516
0.994074
0.99519
0.996305
0.997037
0.99749
0.998501
0.998675
0.999268
0.999721
0.999791
1.000105
1.000174
1.000314

0.998281
1.000013
1.000063
0.99946
0.998017
0.997553
0.99552
0.994967
0.994327
0.994503
0.995833
0.997076
0.997841
0.999435
0.999661
1
0.999423
0.999147
0.997164
0.996235
0.994854
0.993035
0.992143
0.992432
0.992821
0.994465
0.995083
0.996586
0.998155
0.998481
0.999661
0.999925
1.000201

0.041879
0.026441
0.024134
0.015975
0.010873
0.010404
0.010605
0.012449
0.015377
0.018858
0.020893
0.020636
0.020703
0.020204
0.019806
0.018597
0.016632
0.01586
0.013363
0.012225
0.010481
0.007723
0.005291
0.003927
0.003202
0.003263
0.003493
0.004192
0.004962
0.005079
0.005921
0.006273
0.006973

0.060564
0.0314
0.028467
0.01822
0.00744
0.004322
0.01529
0.023657
0.036247
0.0521
0.058953
0.055798
0.053642
0.04888
0.047914
0.048239
0.052183
0.053349
0.065821
0.072119
0.078754
0.081787
0.076892
0.068109
0.061869
0.043466
0.039473
0.025545
0.013121
0.010671
0.00728
0.008608
0.01109

0.046026
0.007746
0.001652
0.038314
0.081137
0.091617
0.115215
0.111895
0.102315
0.091285
0.092657
0.087874
0.083405
0.051732
0.04114
0.006268
0.054078
0.06356
0.097464
0.101126
0.099961
0.106798
0.139548
0.154691
0.166645
0.165261
0.159242
0.126191
0.078956
0.067342
0.020748
0.008899
0.011036

64

Normalized
Depth

0.082633
0.123559
0.131832
0.168108
0.197715
0.197066
0.247388
0.269442
0.294208
0.33618
0.370885
0.398858
0.415024
0.457213
0.470662
0.504992
0.538131
0.550809
0.580971
0.611235
0.643514
0.673975
0.69509
0.718823
0.737492
0.792886
0.813402
0.84767
0.882056
0.876178
0.911786
0.920095
0.928805

Result and Discussion


Table 3.23 Comparing target result of crack location from fixed end with ANN result for relative
natural frequency with average mode shape deviation location network.
Target (mm)

ANN(mm)
Error (mm)
39.42166
-14.4217
25
40
43.70619
-3.70619
49.57962
0.420384
50
74.13547
0.864535
75
79.09911
0.900891
80
100.8646
-0.86458
100
118.6291
1.370852
120
118.2397
6.760264
125
148.4329
1.567059
150
161.6652
-1.66518
160
176.5251
-1.52506
175
201.7081
-1.70812
200
222.5309
2.469117
225
239.3151
0.684943
240
249.0144
0.985629
250
274.3281
0.671944
275
282.3971
-2.39706
280
302.9953
-2.99527
300
322.8786
-2.87858
320
330.4852
-5.48519
325
348.5825
1.417452
350
366.7411
-6.74112
360
386.1085
-11.1085
375
404.3852
-4.38515
400
417.0543
7.9457
425
431.2939
8.706128
440
442.4951
7.504927
450
475.7313
-0.73133
475
488.0414
-8.04137
480
508.6022
-8.6022
500
529.2333
-9.23331
520
525.7069
-0.70692
525
547.0719
2.928132
550
552.057
7.942951
560
557.283
17.71703
575
Maximum Error 17.71mm
Average Error 4.15 mm

65

Chapter 3
3.2.7 Discussion
For all networks MSE and R value are less than 0.001 and 0.988 respectively. Maximum and
average errors of 19 patterns in relative natural frequency depth network (Table 3.8) are 0.3246 mm
and 0.1141mm respectively. Maximum and average errors in average mod shape deviation - depth
network (Table3.11) are 0.4557 mm and 0.1796 mm respectively. Maximum and average errors in
relative natural frequency and average mode shape deviation depth network (Table3.14) are 0.2136
mm and. 0.0935 respectively. Maximum and average errors of 35 patterns in relative natural
frequency location network (Table 3.17) are 31.27 mm and 5.1 mm respectively. Maximum and
average errors in average mode shape deviation - location network (Table 3.20) are 79.27 mm and
17.13 mm respectively. Maximum and average errors in relative natural frequency and average mode
shape deviation location network (Table 3.23) are 17.71 mm and 4.15 mm. Most of the predicted
results by the developed ANN and their comparison with corresponding finite element results which
show very good agreement. It is observed that neural networks can successfully identify and calculate
the depth and location of different crack cases.

3.3 Identification of Crack Depth and Location by Wavelet Transform


analysis
In order to identify the crack depth and its location application of wavelet analysis is performed on
the displacement response of the cracked beam. Two cases of the beam have been considered.
a) Beam under static load.
b) Beam under dynamic load
In both cases, identification of crack is investigated using discreet wavelet transform (DWT) and
contentious wavelet transforms (CWT).

3.3.1 Beam under static loading


In order to determine the location of crack and its depth static analysis is performed first. The beam
specification is already mentioned in Table 2.1. Beam is divided 60 finite parts with 61 nodes. First
node is being taken at the fixed end. A static load of 294 N is applied for both cracked and crack free
beam. The location of the crack of depth 3 mm is assumed to be 50 mm from fixed end. The
deflection of the beams under the said load is shown in Figure 3.21.

66

Result and Discussion

Identification of Location
In order to determine the efficacy of the wavelet technique for crack detection, two cracked beams
are considered at the same location, one crack being small (3mm depth) and the other one big (10 mm
depth). .

Fig 3.21 Static deflection of crack free and cracked beam with 3 mm cracked depth at 50 mm
from fixed end
Both DWT bior6.8 wavelet and CWT bior6.8 wavelet is applied to crack beam taking static
deflation as an input signal of wavelet analysis. Figure 3.22 and 3.23 show DWT and CWT responses
of those cracked beam. Those figures show that if the crack is big it can be clearly identified but when
crack is small like 3 mm depth it is difficult to detect.
From Figure 3.22 it has been observed that no disturbance is detected in the response signal except
at 61 node which is nothing but free end of the cantilever beam. Figure 3.23 (a) shows the CWT
response (scale and coefficient) along the length for the same cracked beam. Figure 3.23 (b) exhibits
the same result in a 3D plot. In order to eliminate the end effect, beam length is shown till 50 node.
From this analysis it is not possible to detect any damage that may exist in the beam.

67

Chapter 3

Fig 3.22 DWT plot of static deflection cracked beam with 3 mm crack depth at location 50 mm
from fixed end.

Fig 3.23(a) CWT 2D plot of static deflection of cracked Beam crack location 50 mm from fixed
end with 3 mm crack depth.

Fig. 3.23(b) CWT 3D plot of static deflection of cracked Beam crack location 50 mm from fixed end
with 3 mm crack depth. (a) 2D plot (b) 3D plot
68

Result and Discussion

A similar analysis is undertaken with a crack depth at 10 mm and the DWT and CWT response are
presented in Figure 3.24 and 3.25 respectively. A careful inspection of the plots reveals that there
could be a crack at node 6 though one may not be very sure about its existence.

Fig 3.24 DWT plot of static deflection cracked beam at location 50 mm from fixed end with 10
mm crack depth

Fig. 3.25 (a) CWT 2D plots of static deflection of cracked beam crack location at 50 mm from fixed
end with 10 mm crack depth.

69

Chapter 3

Fig. 3.25 (b) CWT 3D plots of static deflection of cracked beam crack location at 50 mm from fixed
end with 10 mm crack depth.

Fig. 3.26 Relative Static Deflection vs. Node for cracked beam with 3 mm crack depth at location 50
mm from fixed end

In order to overcome this difficulty an investigation is carried out with relative deflection of
cracked free and cracked beam as an input signal to detect the crack in a beam. Three cracked beams
70

Result and Discussion


are taken with 3 mm crack depth at a distance of 50 mm, 250 mm and 450 mm from the fixed end
respectively. Figure 3.28 shows the variation of relative static deflection along the nodes for the beam
having crack at a distance of 50 mm from fixed end. Figure 3.27, 3.28 and 3.29 shows DWT and
CWT of three cracked beam at different location, input signal being relative deflection of cracked free
and cracked beam.
.

Fig. 3.27 (a) DWT plot of relative static deflection cracked beam and crack free beam (crack location
50 mm from fixed end with 3 mm crack depth)

Fig. 3.27 (b) CWT 2D plot of relative static deflection of cracked beam and crack free beam (crack
location 50 mm from fixed end with 3 mm crack depth)

71

Chapter 3

Fig. 3.27 (c) CWT 3D plot of relative static deflection of cracked beam and crack free beam (crack
location 50 mm from fixed end with 3 mm crack depth)

Fig. 3.28 (a) DWT plot of relative static deflection cracked beam and crack free beam (crack location
250 mm from fixed end with 3 mm crack depth)

Fig. 3.28 (b) CWT 2D plot of relative static deflection of cracked beam and crack free beam (crack
location 250 mm from fixed end with 3 mm crack depth)
72

Result and Discussion

Fig. 3.28 (c) CWT 3D plot of relative static deflection of cracked beam and crack free beam (crack
location 250 mm from fixed end with 3 mm crack depth).
From Figure 3.27, it is clearly detected that there is a crack at node 6 at a distance of 50 mm from
fixed end Even DWT response shows this which is actually not possible to understand from Fig 5.28.
Similarly, for the other two cases (crack at 250 mm and 450 mm from fixed end), also existence of
crack is also recognized at nodes 26 and 46 from DWT and DWT plots shown in Figure5.28 and 5.29.
Identification of Depth
Similar analysis is performed to estimate of crack depth. If wavelet analysis is performed with
various cracked beam deferent depth at same location, the maximum wavelet coefficient is changed.
Table 3.24 shows maximum coefficients of cracked beam at location 50 mm from fixed end. Figure
5.32 shows graphical representation of Table 3.24. Therefore it is possible to determine the crack
depth of an identical beam by determining maximum wavelet coefficient of the beam (Figure 3.30) if
location of crack is at a distance of 50 mm from fixed end.

Fig. 3.29(a) DWT plot of relative static deflection cracked beam and crack free beam (crack location
450 mm from fixed end with 3 mm crack depth)
73

Chapter 3

Fig. 3.29(b) CWT 2D plot of relative static deflection of cracked beam and crack free beam (crack
location 450 mm from fixed end with 3 mm crack depth)

Fig. 3.29(c) CWT 3D plot of relative static deflection of cracked beam and crack free beam (crack
location 450 mm from fixed end with 3 mm crack depth).

74

Result and Discussion


Table 3.24 Maximum wavelet coefficient at different depth at location 50 mm from fixed end
Depth (mm)
1
3
5
7
10

Maximum wavelet coefficient


0.0011
0.0084
0.0235
0.0487
0.197

. Fig.3.30 Maximum wavelet coefficient versus depth at location 50 mm from fixed end

3.3.2 Dynamic analysis


In order to determine the location of crack and its depth now dynamic modal analysis is performed.
The beam specification is already mentioned in Table 2.1.Beam is divided 60 finite parts with 61
nodes. First node is being taken at the fixed end. The continuous wavelet transform and discrete
wavelet both are performed. For present study the wavelet transform is implemented for scales 1-16
with bior6.8 wavelet as the analyzing wavelet. Biorthogonal wavelet is preferred for this analysis
because in this wavelet analysis scaling function exists, DWT is possible and also a fast algorithm is
available.
Identification of Location
In order to determine the efficacy of the wavelet technique for crack detection, two cracked beams
are considered at the same location, one crack being small (5mm depth) and the other one big (10 mm
depth).
75

Chapter 3
Both DWT and CWT wavelet is applied to crack beam taking first mode shape of the cracked beam
as an input signal of wavelet analysis. Figure 3.31 - 3.34 shows DWT and CWT response of those
cracked beam. .
From Figure 3.31 it has been observed that no disturbance is detected in the response signal expect
at 61 node which is nothing but free end of the cantilever beam. Figure 3.32 (a) shows the CWT
response (scale and coefficient) along the length for the same cracked beam. Figure 3.32 (b) exhibits
the same result in a 3D plot. From this analysis it is not possible to detect any damage that may exist
in the beam.
A similar analysis is undertaken with a crack depth of 10 mm and the DWT and CWT response are
presented in Figure 3.33 and 3.34 respectively. Only Figure 3.33 shows little discontinuity of DWT
for crack with 10 mm depth. But CWT (Figure 5.34) plot do not show any evidence of crack.

Fig. 3.31 DWT plot of plot of original mode shape of Cracked Beam with 5 mm crack depth and
location 50 mm from fixed end

Fig. 3.32 (a) CWT plot of original mode shape of Cracked Beam with 5 mm crack depth and location
50 mm from fixed end

76

Result and Discussion

Fig. 3.32 (b) CWT plot of original mode shape of Cracked Beam with 5 mm crack depth and location
50 mm from fixed end

Fig. 3.33 DWT plot of plot of original mode shape of Cracked Beam with 10 mm crack depth and
location 50 mm from fixed end

Fig. 3.34 (a) CWT 2D plot of original mode shape of Cracked Beam with 10 mm crack depth and
location 50 mm from fixed end
77

Chapter 3

Fig. 3.34 (b) CWT 3D plot of original mode shape of Cracked Beam with 10 mm crack depth and
location 50 mm from fixed end
In order to overcome this difficulty an investigation is carried out with the mode shape difference of
cracked and cracked free beam as an input signal to wavelet analysis. Three cracked beam are taken
with 1 mm, 5mm, and 10 mm depth at a distance 50 mm from fixed end respectively. Figure 3.35,
3.36 and 3.37 shows DWT and CWT of three cracked beam of different depth input signal being
relative first mode shape difference of crack free and cracked beam. From the plots it is evident that
there is a crack at node 6 at a distance of 50 mm from fixed end. It is to be pointed out that though
CWT plots could indicate the position of the crack in all cases, DWT plot fails to identify crack
position when crack depth is small, 1 mm (Figure 3.37 (a)).

Fig 3.35 (a) DWT plot of mode shape difference of Crack free and Cracked Beam with 1 mm crack
depth at crack location 50 mm from fixed end.

78

Result and Discussion

Fig 3.35 (b) CWT 2D plot of mode shape difference of Crack free and Cracked Beam with 1 mm
crack depth at crack location 50 mm from fixed end.

Fig 3.35 (c) CWT 3D plot of mode shape difference of Crack free and Cracked Beam with 1 mm
crack depth at crack location 50 mm from fixed end.

Fig 3.36 (a) DWT plot of mode shape difference of Crack free and Cracked Beam with 5 mm crack
depth at crack location 50 mm from fixed end.
79

Chapter 3

Fig 3.36 (b) CWT 2D plot of mode shape difference of Crack free and Cracked Beam with 5 mm
crack depth at crack location 50 mm from fixed end.

Fig 3.36 (c) CWT 3D plot of mode shape difference of Crack free and Cracked Beam with 5 mm
crack depth at crack location 50 mm from fixed end.

Fig 3.37 (a) DWT plot of mode shape difference of Crack free and Cracked Beam with 10 mm crack
depth at crack location 50 mm from fixed end.
80

Result and Discussion

Fig 3.37 (b) CWT 2D plot of mode shape difference of Crack free and Cracked Beam with 10 mm
crack depth at crack location 50 mm from fixed end.

Fig 3.37 (c) CWT 3D plot of mode shape difference of Crack free and Cracked Beam with mm crack
depth at crack location 50 mm from fixed end.

81

Chapter 3

Fig 3.38 (a) DWT plot of mode shape difference of Crack free and Cracked Beam with 3 mm crack
depth at crack location 150 mm from fixed end.

Fig 3.38 (b) CWT 2D plot of mode shape difference of Crack free and Cracked Beam with 3 mm
crack depth at crack location 150 mm from fixed end.

Fig 3.38 (c) CWT 2D plot of mode shape difference of Crack free and Cracked Beam with 3 mm
crack depth at crack location 150 mm from fixed end.
82

Result and Discussion

Fig 3.39 (a) DWT plot of mode shape difference of Crack free and Cracked Beam with 3 mm crack
depth at crack location 250 mm from fixed end.

Fig 3.39 (b) CWT 2D plot of mode shape difference of Crack free and Cracked Beam with 3 mm
crack depth at crack location 250 mm from fixed end.

Fig 3.39 (c) CWT 3D plot of mode shape difference of Crack free and Cracked Beam with 3 mm
crack depth at crack location 250 mm from fixed end.
Now investigation is carried out for other location of crack. Two cracked beams are taken with 3
mm crack depth at a distance of 150 mm and 250 mm from the fixed end respectively. Figure 3.38
83

Chapter 3
and 3.39 shows DWT and CWT of three cracked beam at different location, input signal being
relative first mode shape of cracked free and cracked beam. From Figure3.38 it is clearly shown that
there is a crack at node 16 at a distance of 150 mm from fixed end. Similarly Figure 3.39 exhibits that
there is a crack at node 26 at a distance of 250 mm from fixed end.
Identification of depth
Similar analysis is carried out to estimate of crack depth. If wavelet analysis is performed with
various cracked beam having deferent depth at same location, the maximum wavelet coefficient
obtained from CWT response from Table 3.22 it is observed that maximum coefficients of cracked
beam found to occur at location 50 mm from fixed end for all cases of crack depth. Figure 3.40 shows
graphical representation of Table 3.22. Therefore it is possible to determine the crack depth of an
identical beam by knowing maximum wavelet coefficient of the beam (Figure 3.40) if location of
crack is at a distance of 50 mm from fixed end.

Table 3.25 Maximum wavelet coefficient at different depth at location 50 mm from fixed end for
dynamic modal analysis of cracked beam
Depth (mm)
1
2
3
4
5
6
7
8
9
10

Maximum coefficient
0.0066
0.0239
0.051
0.0874
0.1369
0.1944
0.266
0.3575
0.4601
0.5798

84

Result and Discussion

Fig 3.40 Maximum wavelet coefficient versus depth at location 50 mm from fixed end for
dynamic modal analysis of cracked beam

3.4 Comparison with artificial neural network analysis and wavelet


transform analysis.
Two completely different methodologies are adopted detection of crack location and its
depth in a cantilever beam one with the application of Artificial Neural Network and the
other being wavelet transform. Both the methods have some merits and demerits as
mentioned below.
It is evident from the ANN analysis result described in section that whatever may be the
process of training the network, there is existed some error when location and depth of the
crack is predicted by this method. However, in case of wavelet analysis, one may claim that
there is almost no error present while predicting the crack location. It also provides a very
good estimation in respect to prediction of depth of crack.
For crack detection by this method, one should take precaution to exclude signal arising
from geometrical discontinuity due to boundary condition at ends or otherwise. If crack
exists very near to the boundary it may be difficult to find it. On the other hand, ANN
method is free from this type of problem.

85

Chapter 3
In order to predict crack through a cracked free beam is also required. This is not essential
in case of wavelet transform.
Even with the help of static analysis also, detection of crack is possible using wavelet
transform. However it is not possible in case of ANN analysis.

86

Chapter 4
CONCLUSION AND SCOPE FOR FUTURE WORK
4.1 Conclusion
The dynamic behavior of beam structure undergoes remarkable changes in presence of crack so far
as natural frequencies and mode shapes of the structure is concerned using this characteristics of a
vibrating beam two different techniques have been employed for identification of crack of a cracked
cantilever beam. Two techniques are Artificial Neural Network and Wavelet Transform. First finite
element modeling of crack and crack free beam is done for subsequent modal analyses of the beam
using ANSYS software. Then the result obtains from finite element analysis is used for ANN and
Wavelet Transform analysis.
The neural network technique considered here is used to predict the crack location and its intensity
by using relative deviation of first three natural frequencies and first three mode shapes as inputs. Six
different Artificial Neural Networks are used for detection of crack depth and its location. Feed
forward back propagation Levenberg Marquardt algorithm is used for this analysis. After proper
training results are obtained by simulation. Lastly output targets are comparing with the FEA data. It
is observed that predicted results by ANN are reasonably acceptable and agreement with the FEA
data.
In the second method, wavelet transform is applied to detect the crack in the cantilever beam. Both
dynamic and static wavelet analysis is done in this present work. For analysis bior6.8 wavelet is used.
It has been observed that better detection of crack is possible when relative mode shape difference
between crack and crack free beam is introduced as the input signal instead of when only mode shape
of the cracked beam is fed as input for wavelet analysis. It is also shown that crack detection is also
possible from the results of static analysis also in conjunction with wavelet transform. It can be
inferred that wavelet analysis could be very efficient tool for prediction of crack in beam like
structure in most cases. However, it may fail to detection the existence of crack with certainty when
crack is situated near the boundary.

4.2 Future scope for ANN analysis of present work

ANN analysis can be used in crack identification for other different beam-type structures.

The cracked cantilever can be analyzed under the influence of external forces.

The dynamic response of the cracked beams can be analyzed for different crack
orientations.
87

Chapter 4

Use hybrid neuro genetic technique for crack detection.

4.3 Future scope for Wavelet Transform analysis of present work

To establish an accurate methodology for the selection of the optimal wavelet and to derive a
procedure to eliminate the effects free end.

To detecting cracks in non-homogenous medium

Application to bi dimensional structures such as plate and shells.

88

REFERENCES

1. Orhan Sadettin, Analysis of free and forced vibration of a cracked cantilever beam, NDT and
E International 40, (2007), 443-450.
2. Friswell, M.I. and Penny, J.E.T., Crack modeling for structural health monitoring, Structural
Health Monitoring (2002), 139-148.
3. Loutridis, S, Douka, E and Hadjileontiadis, L.J., Forced vibration behaviour and crack
detection of cracked beams using instantaneous frequency, NDT and E International 38,
(2005), 411-419.
4. Chinchalkar, S., Determination of crack location in beams using natural frequencies, Journal
of Sound and V ibration 247(3), (2001), 417-429.
5. Lee Jinhee, Identification of multiple cracks in a beam using vibration amplitudes, Journal of
Sound and Vibration 326, (2009), 205-212.
6. Nandwana, B. P., and Maiti, S. K., Detection of the location and size of a crack in stepped
cantilever beams based on measurements of natural frequencies, Journal of Sound and
Vibration 203(3), (1997), 435-446.
7. Nandwana, B. P. and Maiti, S.K., Modelling of vibration of beam in presence of inclined
edge or international crack for its possible detection based on frequency measurements,
Engineering Fracture Mechanics 58, (1997), 193-205.
8. Choubey, A., Sehgal, D.K. and Tandon, N., Finite element analysis of study changes in
natural frequencies due to cracks, International Journal of Pressure Vessels and Piping 83,
(2006), 181-187.
9. Chan, R.K.C. and Lai, T.C., Digital simulation of a rotating shaft with a transverse crack ,
Appl. Math. Modelling 19, (1995), 411-420.
10. Kim, J.T., Ryu, Y.S. and Cho Hyun-Man, S., Norris, Damage identification in beam-type
structures: frequency-based method vs mode-shape-based method, Engineering Structure 25,
(2003), 57-67.
11. Peng, Z.K., Lang, Z.Q. and Chu, F.L., Numerical analysis of cracked beams using nonlinear
output frequency response functions, Computers and Structures 86, (2008), 1809-1818.
12. Nahvi, H. and Jabbari, M., Crack detection in beams using experimental modal data and
finite element model, International Journal of Mechanical Sciences 47, (2005), 1477-1497.
13. Arif Gurel, M. and KIsa, M., Free vibration analysis of uniform and stepped cracked beams
with circular cross sections, International Journal of Engineering Science 45, (2007), 364-380.
14. Zheng, D.Y. and Kessissoglou, N.J., Free vibration analysis of a cracked beam by finite
element method, Journal of Sound and Vibration 273, (2004), 457-475.
89

15. Saavedra, P.N. and Cuitino L.A., Crack detection and vibration behavior of cracked beams,
Computers and Structures 79, (2001), 1451-1459.
16. Rosales, M. B., Filipich, C.P. and Buezas, F.S., Crack detection in beam-like structures,
Engineering Structures 31, (2009), 2257-2264.
17. Bakhary, N., Hao, H. and Deeks, A.J., Damage detection using artificial neural network with
consideration of uncertainties, Enginering Stuctures 29, (2007), 2806-2815
18. Liu, S.W., Huang Jin H., Sung J.C. and Lee C.C., Detection of cracks using neural networks
and computational mechanics, Comput. Methods Appl.Mech. Engrg. 191, (2002), 2831-2845.
19. Parhi, D.R. and Dash, A.K., Application of neural networks and finite elements for condition
monitoring of structures, Journal of Mechanical Engineering Science 225, (2011), 1329-1339.
20. Parhi, D.R. and Das, H.C., Application of Neural network for fault diagnosis of cracked
cantilever beam, World Congress on Nature and Biologically Inspired Computing, NABIC,
(2009), 1303-1308.
21. Saeed, R.A. and Galybin , A.N. and Popov, V., Crack identification in curvilinear beams by
using ANN and ANFIS based on natural frequencies and frequency response functions, Neural
Comput. and Applic.(2011).
22. Mahmoud, M.A. and Abu Kiefa M.A., Neural network solution of the inverse vibration
problem, NDT AND E International 32, (1999), 91-99.
23. zkaya, E. and z, H.R., Determination of natural frequencies and stability regions of
axially moving beams using artificial neural networks method, Journal of Sound and Vibration
252(4), (2002), 782-789.
24. Nazari, F. and Baghalian, S., A new method for damage detection in symmetric beams using
artificial neural network and finite element method, International Journal of Engineering and
Applied Sciences (IJEAS) 3, (2011), 30-36.
25. Zou, J., Chen, J. and Pu, Y.P., Wavelet time-frequency analysis of torsional vibrations in
rotor system with a transverse crack, Computers and Structures 82, (2004), 1181-1187.
26. Xiang, J., Chen, X., Mo, Q. and He, Z., Identification of crack in a rotor system based on
wavelet finite element method, Finite Element in Analysis and Design 43, (2007), 1068-1081.
27. Xiang, J., Zhong, Y., Chen, X. and He, Z., Crack detection in a shaft by combination of
wavelet based elements and genetic algorithm, International Journal of Solids and Structures
45, (2008), 4782-4795.
28. Quek, S.T., Wang, Q., Zhang, L. and Ang, K.K., Sensitivity analysis of detection in beams
by wavelet technique, International Journal of Mechanical Sciences 43, (2001), 2899-2910.

90

29. Messina, A., Refinements of damage detection methods based on wavelet analysis of
dynamical shapes, International Journal of Solids and Structures 45, (2008), 4068-4097.
30. Ovanesova, A.V., Surez, L.E., Applications of wavelet transforms to damage detection in
frame structures, Engineering Structures 26, (2004), 39-49.
31. Lokar, A.A. and Srivastava, R.K., Crack detection in structure in structure using wavelet
transform and higher order differentiated mode shapes, International Journal of Mechanical
Engineering 1, (2011), 26-34.
32. Douka, E., Loutridis, S. and Trochidis, A., Crack identification in beams using wavelet
analysis, International Journal of Solids and Structures 40, (2003), 3557-3569.
33. Loutridis, S., Douka, E. and Trochidis, A., Crack identification in double-cracked beams
using wavelet analysis, Journal of Sound and Vibration 277, (2004), 1025-1039.
34. Wang, Q. and Wu, N., Detecting the delaminating location of a beam with a wavelet
transform: an experimental study, Smart Mater. Struct. 20, (2011), 1-7.
35. Katunin, A., Damage identification in composite plates using two-dimensional B-spline
wavelets, Mechanical Systems and Signal Processing 25, (2011), 3153-3167.
36. Kim, H. and Melhem, H., Damage detection of structures by wavelet analysis, Engineering
Structures 26, (2004), 347-362.
37. Jiang, X., John Ma, Z. and Ren W.X., Crack detection from the slop of the mode shape using
complex continuous wavelet transform, Computer- Aided Civil and Infrastructure Engineering
27 (2012) 187-201.
38. Srinivasarao, D., Rao, K. M. and Raju, G.V., Crack identification on a beam by vibration
measurement and wavelet analysis, International Journal of Engineering Science and
Technology 2(5), (2010), 907-912.
39. Okafor, A. C. and Dutta, A., Structural damage detection in beams by wavelet transforms,
Smart Mater. Struct. 9, (2000), 906-917.
40. Zhong, S. and Oyadiji, S.O., Detection of cracks in simply-supported beams by continuous
wavelet transform of reconstructed modal data, Computers and Structures 89, (2011), 127-148.
41. Zhong, S and Oyadiji, S.O., Crack detection in simply supported beams without baseline
modal parameters by stationary wavelet transform, Mechanical Systems and Signal Processing
21, (2007), 1853-1884.
42. Erdogan, M and Ibrahim, G., The Finite Element Method Application in Engg. by

using ANSYS, (Springer)2007..

91