EXPERIMENTAL VALIDATION OF A NOVEL STRUCTURAL HEALTH MONITORING  
STRATEGY FOR BOLTED PIPELINE JOINTS 
 
 
 
 
by 
 
 
 
 
Julie L. Briand 
 
 
 
Submitted in partial fulfilment of the requirements 
for the degree of Master of Applied Science 
 
 
at 
 
 
Dalhousie University 
Halifax, Nova Scotia 
August 2010 
 
 
 
 
 
 
 
© Copyright by Julie L. Briand, 2010 
 
 
 
 
 
ii 
 
 
DALHOUSIE UNIVERSITY 
DEPARTMENT OF CIVIL AND RESOURCE ENGINEERING 
 
 
The undersigned hereby certify that they have read and recommend to the Faculty of 
Graduate Studies for acceptance a thesis entitled “Experimental Validation of a Novel 
Structural Health Monitoring Strategy for Bolted Pipeline Joints” by Julie L. Briand in 
partial fulfillment of the requirements for the degree of Master of Applied Science. 
 
 
Dated: August 18, 2010 
 
Supervisor: _________________________________ 
Readers: _________________________________ 
 _________________________________ 
 _________________________________ 
 
 
 
  
 
 
 
 
 
 
 
 
 
 
 
iii 
 
 
DALHOUSIE UNIVERSITY 
 
 DATE: August 18, 2010 
AUTHOR: Julie L. Briand 
TITLE: Experimental Validation of a Novel Structural Health Monitoring Strategy 
for Bolted Pipeline Joints 
DEPARTMENT OR SCHOOL: Department of Civil and Resource Engineering 
DEGREE: M.A.Sc. CONVOCATION: October YEAR: 2010 
Permission is herewith granted to Dalhousie University to circulate and to have copied 
for non-commercial purposes, at its discretion, the above title upon the request of 
individuals or institutions. 
 _______________________________ 
 Signature of Author 
 
The author reserves other publication rights, and neither the thesis nor extensive extracts 
from it may be printed or otherwise reproduced without the author’s written permission. 
 
The author attests that permission has been obtained for the use of any copyrighted 
material appearing in the thesis (other than the brief excerpts requiring only proper 
acknowledgement in scholarly writing), and that all such use is clearly acknowledged. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
iv 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
To Stephen Buchholz, who supported me through  
my uncertainty, self-doubt and discontent; 
you have my gratitude and love. 
 
 
v 
 
TABLE OF CONTENTS  
List of Tables .................................................................................................................... ix 
List of Figures .................................................................................................................... x 
List of Abbreviations and Symbols Used ..................................................................... xiii 
Acknowledgments ........................................................................................................... xv 
Abstract….. ..................................................................................................................... xvi 
Chapter 1  Introduction ................................................................................................... 1 
1.1  Structural Health Monitoring ................................................................................... 1 
1.2  Vibration-Based SHM .............................................................................................. 5 
1.3   Motivation ............................................................................................................... 7 
1.4   Objective ................................................................................................................. 8 
1.5   Thesis Layout .......................................................................................................... 8 
Chapter 2 Literature Review ......................................................................................... 10 
2.1  Bolted Joints ........................................................................................................... 10 
2.2  Bolted Joint Failure ................................................................................................ 11 
2.2.1 Initial Preload ................................................................................................. 13 
2.2.1.1 Friction ...................................................................................................... 14 
2.2.1.2 Embedment ............................................................................................... 16 
2.2.1.3 Nonparallel Joint Surfaces ........................................................................ 17 
2.2.1.4 Bolting Method ......................................................................................... 17 
2.2.2 Residual Preload ............................................................................................. 18 
2.2.2.1 Short-Term Relaxation.............................................................................. 18 
2.2.2.1.1 Sources of Short-Term Relaxation..................................................... 19 
2.2.2.1.2 Factors Affecting Short-Term Relaxation ......................................... 21 
2.2.2.2 Elastic Interactions Between Bolts ........................................................... 22 
2.2.3 Service Conditions.......................................................................................... 22 
2.2.3.1 Bolt Fatigue ............................................................................................... 23 
2.2.3.2 Corrosion................................................................................................... 23 
2.2.4 Long-Term Relaxation ................................................................................... 24 
2.3  Damage Detection Methods for Bolted Joints ....................................................... 25 
vi 
 
2.3.1 Visual Inspection ............................................................................................ 26 
2.3.2 Cylindrical Guided Wave Technique (CGWT) .............................................. 27 
2.3.3 Ultrasonic Testing (UT) ................................................................................. 27 
2.4  Structural Health Monitoring ................................................................................. 28 
2.4.1 Piezoelectric Sensors ...................................................................................... 28 
2.4.2 SHM Applied to Bolted Joints ....................................................................... 30 
2.4.3 Impedance-Based Structural Health Monitoring ............................................ 31 
2.4.4 Vibration-Based SHM of Bolted Joints .......................................................... 33 
2.5  Hilbert-Huang Transform ....................................................................................... 35 
2.5.1 Application of HHT in Structural Damage Detection .................................... 36 
2.6  EMD Energy Damage Index .................................................................................. 38 
2.6.1 Application of the EMD Energy Damage Index to SHM .............................. 39 
Chapter 3 Damage Detection of a Steel Pipeline Bolted Joint .................................... 42 
3.1  General Experimental Set-Up ................................................................................ 43 
3.2  Session 1 ................................................................................................................. 47 
3.2.1 Experimental Set-Up ...................................................................................... 47 
3.2.2 Testing ............................................................................................................ 49 
3.2.3 Data Analysis.................................................................................................. 51 
3.2.4 Results ............................................................................................................ 53 
3.2.5 Consistency..................................................................................................... 56 
3.2.6 Result Repeatability........................................................................................ 57 
3.2.7 Discussion and Conclusions ........................................................................... 58 
3.3  Session 2 ................................................................................................................. 59 
3.3.1 Experimental Set-Up ...................................................................................... 59 
3.3.2 Testing ............................................................................................................ 60 
3.3.3 Data Analysis.................................................................................................. 62 
3.3.4 Results ............................................................................................................ 62 
3.3.5 Consistency..................................................................................................... 64 
3.3.6 Result Repeatability........................................................................................ 65 
3.3.7 Discussion and Conclusions ........................................................................... 65 
3.4   Session 3 ................................................................................................................ 67 
vii 
 
3.4.1 Experimental Set-Up ...................................................................................... 67 
3.4.2 Testing ............................................................................................................ 69 
3.4.3 Data Analysis.................................................................................................. 70 
3.4.4 Results ............................................................................................................ 70 
3.4.5 Consistency..................................................................................................... 74 
3.4.6 Discussion and Conclusions ........................................................................... 74 
Chapter 4 Damage Detection of an In-Service Condensation Pipeline Joint ............ 75 
4.1  Abstract .................................................................................................................. 75 
4.2   Introduction ........................................................................................................... 76 
4.3   Methodology ......................................................................................................... 78 
4.4   Experimental Study ............................................................................................... 80 
4.5   Results and Discussion .......................................................................................... 84 
4.5.1 Result Analysis ............................................................................................... 85 
4.5.2 Discussion....................................................................................................... 90 
4.6   Conclusions ........................................................................................................... 91 
4.7   Acknowledgements ............................................................................................... 92 
4.8   References ............................................................................................................. 93 
Chapter 5 Damage Detection of a Pressurized Steel Pipeline Joint Using an 
Empirical Mode Decomposition-Based Energy Damage Index ................ 95 
5.1  Abstract .................................................................................................................. 95 
5.2   Introduction ........................................................................................................... 96 
5.3   Motivations and Objectives ................................................................................. 100 
5.4   Empirical Mode Decomposition ......................................................................... 100 
5.5   EMD Energy Damage Index ............................................................................... 103 
5.6   Experimental Set-Up ........................................................................................... 105 
5.7   Experimental Testing & Analysis ....................................................................... 111 
5.8   Results and Discussion ........................................................................................ 116 
5.9   Conclusions ......................................................................................................... 119 
5.10 Acknowledgements ............................................................................................. 120 
5.11 References ........................................................................................................... 121 
Chapter 6 Damage Detection of a PVC Pipeline Joint .............................................. 125 
viii 
 
6.1 Experimental Set-Up ............................................................................................. 126 
6.2 Testing ................................................................................................................... 131 
6.3 Data Analysis ........................................................................................................ 132 
6.4 Results ................................................................................................................... 133 
6.5 Consistency ........................................................................................................... 134 
6.6 Set-up Reliability .................................................................................................. 135 
6.7 Conclusions ........................................................................................................... 135 
Chapter 7 Conclusion ................................................................................................... 136 
7.1  Conclusions .......................................................................................................... 136 
7.1.1 Damage Detection of a Steel Pipeline Bolted Joint ...................................... 137 
7.1.2 Damage Detection of an In-Service Condensation Pipeline Joint ................ 138 
7.1.3 Damage Detection of a Pressurized Steel Pipeline Joint .............................. 139 
7.1.4 Damage Detection of a PVC Pipeline Joint ................................................. 140 
7.2   Recommendations for Future Work .................................................................... 141 
References ...................................................................................................................... 143 
Appendix A Supplementary Results from Chapter 3 ................................................ 152 
Appendix B Supplementary Results from Chapter 5 ................................................ 170 
Appendix C Supplementary Results from Chapter 6 ................................................ 178 
Appendix D Copyright Permission Letter & Contribution to Manuscript ............. 181 
  
ix 
 
LIST OF TABLES 
Table 1-1  Vibration based damage detection categories and methods. ............................. 5 
Table 2-1  Bolting method accuracy. ................................................................................ 18 
Table 2-2  Inspection techniques for joint damage ........................................................... 26 
Table 3-1  Session 1-3 ASTM pipe dimensions and material properties. ........................ 43 
Table 3-2  Session 1-3 bolt torque increments.................................................................. 44 
Table 3-3  Piezoceramic sensor dimensions. .................................................................... 45 
Table 3-4  Session 1 healthy cases and damage cases. ..................................................... 50 
Table 3-5  Natural frequencies calculated analytically. .................................................... 53 
Table 3-6  Session 2 cases. ............................................................................................... 61 
Table 3-7  Session 3 cases. ............................................................................................... 69 
Table 3-8  Session 3 frequencies ranges. .......................................................................... 70 
Table 4-1  Damaged joint scenarios.................................................................................. 84 
Table 5-1  ANSI/ASME pipe dimensions and material properties................................. 105 
Table 5-2  Torque increments. ........................................................................................ 107 
Table 5-3  Piezoceramic sensor dimensions ................................................................... 108 
Table 5-4  Damaged joint scenarios................................................................................ 114 
Table 6-1  PVC pipe dimensions and material properties. ............................................. 126 
Table 6-2  Torque increments. ........................................................................................ 128 
Table 6-3  Damaged joint scenarios................................................................................ 132 
 
 
 
  
x 
 
LIST OF FIGURES 
Figure 2-1    Relative relationship between torque, friction and bolt tension ................... 15 
Figure 2-2    Bolt hole interference ................................................................................... 17 
Figure 2-3    General trend of short-term relaxation ......................................................... 18 
Figure 2-4    Embedment .................................................................................................. 19 
Figure 2-5    Poor thread engagement ............................................................................... 20 
Figure 2-6    Poor geometry .............................................................................................. 20 
Figure 2-7    Damage detection methodology ................................................................... 39 
Figure 3-1    Overview of experimental set-up. ................................................................ 42 
Figure 3-2    Boundary conditions. ................................................................................... 44 
Figure 3-3    Session 1-3 bolt numbering. ......................................................................... 44 
Figure 3-4    Session 1 experimental set-up. ..................................................................... 47 
Figure 3-5    Example impact location. ............................................................................. 47 
Figure 3-6    Session 1 PZT sensor placement and numbering. ........................................ 48 
Figure 3-7    Session 1 sensor placement. ......................................................................... 48 
Figure 3-8    Typical Signal Responses ............................................................................ 49 
Figure 3-9    The data acquisition hardware. .................................................................... 50 
Figure 3-10  Session 1 damage overview. ........................................................................ 51 
Figure 3-11  Typical signal decomposition from EMD .................................................... 52 
Figure 3-12  Natural frequencies of pipe-joint system from FFT of PZT7 data ............... 54 
Figure 3-13  EMD-EDI, session 1, location A, IMF1. ..................................................... 55 
Figure 3-14  EMD-EDI, session 1, location B, IMF1. ...................................................... 55 
Figure 3-15  EMD-EDI, session 1, location C, IMF1. ...................................................... 55 
Figure 3-16  EMD-EDI, session 1, location D, IMF1. ..................................................... 56 
Figure 3-17  Session 2 experimental set-up. ..................................................................... 59 
Figure 3-18  Session 2 PZT sensor placement and numbering. ........................................ 60 
Figure 3-19  Session 2 sensor placement. ......................................................................... 60 
Figure 3-20  Session 2 damage overview. ........................................................................ 62 
Figure 3-21  EMD-EDI, session 2, location A, 0-2500 Hz, IMF1, aluminum tip ............ 63 
Figure 3-22  EMD-EDI, session 2, location B, 0-2500 Hz, IMF1, aluminum tip ............ 63 
Figure 3-23  EMD-EDI, session 2, location C, 0-2500 Hz, IMF1, aluminum tip ............ 63 
Figure 3-24  EMD-EDI, session 2, location D, 0-2500 Hz, IMF1, aluminum tip ............ 64 
xi 
 
Figure 3-25  EMD-EDI, session 2, location C, 0-2500 Hz, IMF1, plastic tip .................. 64 
Figure 3-26  Session 3 experimental set-up. ..................................................................... 67 
Figure 3-27  Session 3 PZT sensor placement and numbering ......................................... 68 
Figure 3-28  Session 3 sensor placement. ......................................................................... 68 
Figure 3-29  Session 3 damage overview. ........................................................................ 69 
Figure 3-30  EMD-EDI, session 3, location B, 0-2000 Hz, IMF1, aluminum tip ............ 71 
Figure 3-31  EMD-EDI, session 3, location C, 0-2000 Hz, IMF1, aluminum tip ............ 71 
Figure 3-32  EMD-EDI, session 3, location E, 0-2000 Hz, IMF1, aluminum tip ............ 71 
Figure 3-33  EMD-EDI, session 3, location F, 0-2000 Hz, IMF1, aluminum tip ............. 72 
Figure 3-34  EMD-EDI, session 3, location B, 0-2000 Hz, IMF1, plastic tip .................. 72 
Figure 3-35  EMD-EDI, session 3, location C, 0-2000 Hz, IMF1, plastic tip .................. 72 
Figure 3-36  EMD-EDI, session 3, location E, 0-2000 Hz, IMF1, plastic tip .................. 73 
Figure 3-37  EMD-EDI, session 3, location F, 0-2000 Hz, IMF1, plastic tip .................. 73 
Figure 4-1    Piezoceramic sensor locations for part 2 ...................................................... 81 
Figure 4-2    Pipeline boundary conditions ....................................................................... 82 
Figure 4-3    Experimental set-up. .................................................................................... 83 
Figure 4-4    Scenario 1, impact location B, IMF 1. ......................................................... 85 
Figure 4-5    Scenario 1, impact location B, IMF 2. ......................................................... 86 
Figure 4-6    Scenario 2, impact location B, IMF 1. ......................................................... 86 
Figure 4-7    Scenario 2, impact location D, IMF 1. ......................................................... 87 
Figure 4-8    Scenario 2, impact location B, IMF 2. ......................................................... 87 
Figure 4-9    Scenario 2, impact location D, IMF 2. ......................................................... 88 
Figure 4-10  Scenario 3, impact location B, IMF 1. ......................................................... 89 
Figure 4-11  Scenario 3, impact location D, IMF 1. ......................................................... 89 
Figure 4-12  Scenario 3, impact location B, IMF 2. ......................................................... 89 
Figure 4-13  Scenario 3, impact location D, IMF 2. ......................................................... 90 
Figure 5-1    A schematic representation of one sifting cycle ........................................ 102 
Figure 5-2    Damage detection methodology ................................................................. 104 
Figure 5-3    Experimental set-up overview.................................................................... 106 
Figure 5-4    a) Boundary condition b) Pressure connection. ......................................... 106 
Figure 5-5    Bolt sequence. ............................................................................................ 107 
Figure 5-6    a) Typical bonded piezoceramic sensor b) Impact locations. .................... 109 
Figure 5-7    Experimental set-up. .................................................................................. 110 
xii 
 
Figure 5-8    PZT arrangement. ....................................................................................... 110 
Figure 5-9    Piezoceramic arrangement. ........................................................................ 111 
Figure 5-10  Typical signal responses. ........................................................................... 112 
Figure 5-11  The data acquisition hardware. .................................................................. 113 
Figure 5-12  Damage scenarios. ...................................................................................... 115 
Figure 5-13  Damage scenario 1, 0-2000 Hz, impact location A, added IMFs. ............. 117 
Figure 5-14  Damage scenario 1, 0-2000 Hz, impact location S, added IMFs. .............. 117 
Figure 5-15  Damage scenario 2, 0-2000 Hz, impact location A, added IMFs. ............. 117 
Figure 5-16  Damage scenario 2, 0-2000 Hz, impact location S, added IMFs. .............. 118 
Figure 5-17  Damage scenario 3, 0-2000 Hz, impact location A, added IMFs. ............. 118 
Figure 5-18  Damage scenario 3, 0-2000 Hz, impact location S, added IMFs. .............. 118 
Figure 5-19  Damage scenario 4, 0-2000 Hz, impact location A, added IMFs. ............. 119 
Figure 5-20  Damage scenario 4, 0-2000 Hz, impact location S, added IMFs. .............. 119 
Figure 6-1    Overview of experimental set-up ............................................................... 125 
Figure 6-2    Free-free boundary condition. .................................................................... 127 
Figure 6-3    Cantilevered boundary condition. .............................................................. 127 
Figure 6-4    Bolt numbering........................................................................................... 128 
Figure 6-5    Free-free experimental set-up. ................................................................... 129 
Figure 6-6    Cantilevered experimental set-up. .............................................................. 129 
Figure 6-7    Impact location. .......................................................................................... 129 
Figure 6-8    PZT sensor placement and numbering. ...................................................... 130 
Figure 6-9    Sensor placement. ...................................................................................... 130 
Figure 6-10  Typical signal responses ............................................................................ 131 
Figure 6-11  Joint damage scenarios ............................................................................... 132 
Figure 6-12  Damage scenario 1, 0-2000 Hz, IMF1. ...................................................... 133 
Figure 6-13  Damage scenario 2, 0-2000 Hz, IMF1. ...................................................... 134 
Figure 6-14  Damage scenario 1, 0-1500 Hz, IMF1. ...................................................... 134 
Figure 6-15  Damage scenario 2, 0-1500 Hz, IMF1. ...................................................... 134 
 
 
 
 
  
xiii 
 
LIST OF ABBREVIATIONS AND SYMBOLS USED 
Latin Alphabet 
1c  The first IMF 
ic  IMFs 
nc  n
th
 IMF derived from sifting process 
CGWT Cylindrical Guided Wave Technique 
DAQ Data Acquisition 
  
 
 
  Matrix of piezoelectric strain coefficients 
  EMD energy of an IMF 
  Modulus of elasticity 
     Average EMD energy of multiple impacts, for one sensor 
         EMD energy of an IMF calculated for the damaged state of the system 
         EMD energy of an IMF calculated for the healthy state of the system 
     Maximum EMD energy of multiple impacts, for one sensor 
     Minimum EMD energy of multiple impacts, for one sensor 
EMD Empirical Mode Decomposition 
EMD-EDI EMD Energy Damage Index 
FFT Fast Fourier Transform 
FRF Frequency Response Functions 
1h  Difference between the signal )(tx  and mean 1m in the first sifting 
kh1  Difference between the signal kh1  and the mean km1  in the k
th 
sifting 
HHT
 
Hilbert-Huang Transform 
I Second area moment of Inertia 
IMF Intrinsic Mode Function 
L Length 
1m  Mean of the upper and lower envelopes in the first sifting 
km1  Mean of the upper and lower envelopes in the k
th 
sifting process 
m Mass 
NDE Non-Destructive Evaluation 
PVC Polyvinyl Chloride 
PZT Piezoceramic, Lead Zirconate Titanate 
1r  Residual after extraction of first IMF 
nr  n
th
 residual derived after extraction of n
th
 IMF 
SEM Spectral Element Model 
xiv 
 
SD Standard Deviation 
SHM Structural Health Monitoring 
SMA  Shape Memory Alloy 
   
  
 
  Matrix of mechanical compliance coefficients 
t  Time 
UT Ultrasonic Testing 
VBDD Vibration Based Damage Detection 
VBDT Vibration Based Damage Testing 
)(tx  Continuous time signal 
 
 
Greek Symbols 
  Angular instantaneous frequency 
 Poisson’s ratio  
   
 
 
  Matrix of dielectric permittivity 
  
xv 
 
ACKNOWLEDGMENTS  
My heartfelt appreciation and gratitude goes out to Davood Rezaei, whose endless supply 
of patience, generosity, and continual willingness to help enabled me to complete this 
thesis. Providing me guidance, encouragement and understanding, he not only became 
my de facto supervisor but also a close friend who I greatly admire.  
I am sincerely grateful to my supervisor, Professor Farid Taheri, whose invaluable 
support and guidance allowed me to complete this thesis.  
I wish to extend my thanks to my supervisory committee members Dr. T.S. KoKo and 
Dr. Serguei Iakovlev, for reading, evaluating and commenting on my thesis. 
I also offer my sincere thanks to Mr. Brian Kennedy and Mr. Blair Nickerson who were 
tremendously helpful throughout my experimental testing. My experiments would have 
taken much, much longer to set-up and dismantle without their assistance.  
I would like to gratefully acknowledge the financial support of the Natural Sciences and 
Engineering Council of Canada (NSERC) and of Petroleum Research Atlantic Canada 
(PRAC).  
Last but not least, I would like to extend my appreciation and love to my fellow graduate 
students, past and present, whose support and friendship made the journey all the better.  
 
 
 
 
 
 
 
 
 
 
 
 
xvi 
 
ABSTRACT 
The early detection of damage of in-service structural or mechanical systems is of vital 
importance. With early detection, the damage may be repaired before the integrity of the 
system is jeopardized, avoiding possible monetary losses, environmental impacts, injury 
and death. With this goal in mind, many structural health monitoring techniques have 
been developed which use a combination of sensors and algorithms to collect, process 
and interpret data to detect damage in a structure. This thesis presents work completed in 
support of the experimental validation of a novel structural health monitoring technique 
developed with the aim of providing improved qualitative results compared to those 
methods currently available. 
This novel method is based on the detection of changes within the dynamic 
characteristics of a structure due to damage. In the developed methodology, the free 
vibration response of the structure is monitored and collected through piezoceramic 
sensors before and after the occurrence of damage. Subsequently, this novel method uses 
the decomposition part of the Hilbert-Huang Transform (HHT), known as Empirical 
Mode Decomposition (EMD), to decompose the free vibration signatures of a structure 
into a collection of oscillatory modes, called Intrinsic Mode Functions (IMFs). Next, an 
index, called the EMD Energy Damage Index, is created based on the energy change of 
the IMFs measured from the healthy-state baseline. The index therefore reflects any 
deviations in the structural integrity of the structure thereby providing a means to detect 
the presence of damage.  
As part of an effort to establish the integrity and limitation of this novel damage 
detection method, a series of experimental testing on mechanically bolted joints was 
performed. The damage detection of bolted joints is an important issue due to their wide 
use in aerospace, mechanical and structural systems. Wherever bolted systems exist in 
environments which impose vibration, shocks or thermal cycling on the threaded 
fasteners, they are at risk of self-loosening caused by such dynamic loads.  Four different 
bolted joint systems were tested under different conditions including a steel pipe with an 
8-bolt flanged joint, a pressurized steel pipe with a 12-bolt flanged steel joint, an in-
service steel pipe with a 12-bolt flanged joint and a PVC pipe with an 8-bolt flanged 
joint.  
The proposed methodology was generally successful in detecting the inflicted damage 
in all cases. Numerous parameters were investigated such as impact location, sensor 
location, hammer tip type, frequency bandwidth, intrinsic mode functions, and boundary 
conditions. Two significant conclusions to this work include the necessity of establishing 
a method to determine which frequency band has the most sensitivity to damage and the 
requirement of excitation consistency. 
 
 
 
            
 
1 
 
CHAPTER 1 
INTRODUCTION  
 
1.1  STRUCTURAL HEALTH MONITORING 
With passing time, the majority of systems undergo some form of deterioration which 
may advance to a point of jeopardizing the integrity of the system. However, with early 
detection and maintenance, the health condition of the system may be improved, thereby 
prolonging its life span and avoiding possible monetary losses, environmental impacts, 
injury and death. Therefore, an ideal objective is to have continuous, real-time knowledge 
of the integrity of in-service structures. With such knowledge, users are given confidence 
in the optimal usage of the structure and are able to reduce costs, minimize system 
downtime and avoid catastrophic failures while manufacturers are able to improve their 
products (Chang, 1999). Consequently, the development of diagnostic techniques to 
provide early detection and monitoring of new and existing systems is of vital 
importance.  
Recent sensing technology improvements coupled with current developments in 
computations and communications have resulted in considerable interest in the creation 
of reliable and efficient health monitoring strategies (Chang, 1999). The term created to 
describe this relatively new, interdisciplinary field of engineering is Structural Health 
Monitoring (SHM).  
According to Chang (1999), the objective of structural health monitoring technology is 
to develop autonomous systems which continually monitor and inspect structures for the 
purpose of detecting damage while minimizing labour involvement. For the purposes of 
SHM of mechanical and structural systems, damage may be defined as changes to the 
material and/or geometric properties of infrastructure systems, including changes to 
boundary conditions and system connectivity which may adversely affect the current or 
future performance of the system (Worden et al., 2007). 
2 
 
Conventional understanding of SHM entails a five step process to determine the 
damage state in a system (Farrar and Worden, 2007; Rytter, 1993): 
I) Existence: Is there damage in the structure?  
II) Location: Where is the damage in the structure?  
III) Type: What kind of damage is present?  
IV) Extent: How severe is the damage?  
V) Prognosis: What is the remaining service life of the structure? 
An SHM methodology must include a damage detection technique to fulfill the first 
four steps. Accurate performance of the damage detection method depends upon the 
sensitivity of the sensors and the interpretation algorithm of the damage sensitive features 
(Chang, 1999). The extent and efficiency with which the above questions are fulfilled is 
an excellent method of evaluating the superiority of the damage detection method and the 
SHM procedure overall.  
The process of SHM involves observation of a system and periodic collection of 
response measurements, the subsequent extraction of damage sensitive features and 
finally an analysis to determine the health state of the structure (Farrar and Worden, 
2007). It is important to note that SHM analysis uses changes in the damage sensitive 
features occurring between the previously assessed healthy state and the newly monitored 
state to detect damage. Not only is SHM useful for aging infrastructure, but it is an 
essential tool to establish system integrity after an extreme event such as an earthquake.  
Structural Health Monitoring requires knowledge of many different disciplines 
including structures, materials, computation, signal processing, statistics and sensor 
technology. With such an interdisciplinary field, the body of work performed to date is 
very large and widespread. In an effort to provide a starting point for new SHM 
researchers, Worden et al. (2007) have summarized the fundamental truths of SHM by 
defining the following axioms based on the decades of previous work performed: 
 Axiom I: All materials have inherent flaws or defects; 
 Axiom II: The assessment of damage requires a comparison between two system 
states. 
3 
 
 Axiom III: Identifying the existence and location of damage may be done 
without performing a comparison between the healthy and damaged states, but 
identifying the type of damage present and the damage severity can generally 
only be done when data is available from both the damaged and undamaged 
structures. 
 Axiom IVa: Sensors cannot measure damage. Feature extraction through signal 
processing and statistical classification is necessary to convert sensor data into 
damage information. 
 Axiom IVb: Without intelligent feature extraction, the more sensitive a 
measurement is to damage, the more sensitive it is to changing operational and 
environmental conditions; 
 Axiom V: The length and time-scales associated with damage initiation and 
evolution dictate the required properties of the SHM sensing system. 
 Axiom VI: There is a trade-off between the sensitivity to damage of an algorithm 
and its noise rejection capability. 
 Axiom VII: The size of damage that can be detected from changes in system 
dynamics is inversely proportional to the frequency range of excitation. 
Novel SHM techniques are constantly being developed and evolved in a response to 
demands to increase the serviceable life of our deteriorating infrastructure. With the large 
investments in developing new infrastructure there is a considerable market for SHM 
strategies that provide detailed integrity information capable of providing early warnings 
of structural deficiencies. Furthermore, reliable SHM strategies that replace scheduled 
maintenance and periodic inspection routines will decrease the extensive labour and 
system downtime costs (Chang, 1999). An effective and efficient SHM strategy is one 
that includes a damage detection method capable of detecting and quantifying global 
damage as well as local damage. 
Many conventional damage detection methods are no longer considered efficient 
because they require prior knowledge of the damage location and onsite access to each 
damaged location. These localized methods include: visual inspection, Liquid Penetrant 
4 
 
Testing (PT), Magnetic Particle Testing (MT), Ultrasonic Testing (UT), Magnetic Flux 
Leakage (MFL), Radiography Testing (RT), Eddy-current testing and thermal field 
testing. Furthermore, these methods entail undesirable expenses due to their heavy 
reliance on equipment and/or requirement of highly skilled operators (Rezaei, 2010). 
In response to the limitations of the localized damage detection methods, Vibration-
Based Damage Testing (VBDT) methods have emerged as an alternative option over the 
past three decades. Damage detection using VBDT methods is based on the fact that 
modal properties such as frequencies, mode shapes, and modal damping of a structure are 
functions of physical properties like mass, stiffness, and damping. When defects or 
damages occur in a structure, all or some of these physical properties change and lead to 
variations in the modal characteristics.  
In contrast to conventional methods, VBDT techniques examine the general behavior 
of the structure and provide information on the overall, global, health state of the system. 
This feature makes this type of method perfect for integration with an SHM 
methodology. The observation of the system could then be accomplished continuously or 
periodically based on a collection of response measurements from an array of sensors 
attached to or embedded within the structure.  
This global monitoring ability has made VBDT techniques an incredibly attractive 
research topic for various industries. Doebling et al. (1998) noted that the response to 
VBDT methods has been in part due to media coverage of spectacular structural failures, 
economic concerns and recent technical advancements. The history of quantifiable 
structural VBDT methods dates back to the mid 1970’s when the oil industry spent 
considerable effort to develop the technology for their offshore platforms.  
Since then, the aerospace, infrastructure and composite fields have joined the oil 
industry in developing VBDT techniques. Doebling et al. (1996) and Sohn et al. (2003) 
have performed extensive literature reviews which detail the first two and a half decades 
of work in the field of VBDT for structural and mechanical systems. However, despite 
the dedicated body of research, it is Brownjohn’s belief (2007) that the effectiveness of 
VBDT methods still remains to be proven in practical applications of civil infrastructure 
such as bridges, dams, buildings, towers, offshore installations, nuclear installations and 
tunnels.  
5 
 
1.2  VIBRATION-BASED SHM 
Table 1 lists the different categories of VBDT methods that have been developed to 
date. Each of these methods has its own advantages and disadvantages. 
Table 1-1  Vibration based damage detection categories and methods (adapted from Lee et al., 
2004). 
Category Methodology 
Modal 
Parameters  
Natural Frequencies  
 Frequency change  
 Residual force optimization 
Mode Shapes  
 Mode shape changes  
 Modal strain energy 
 Mode shape derivatives 
Structural 
Matrix 
Methods  
Stiffness Based   Optimization techniques  
 Model updating 
 Dynamically measured flexibility Flexibility Based  
Machine 
Learning  
Genetic Algorithm  
 Stiffness parameter optimization  
 Minimization of the objective function 
Artificial Neural 
Network  
 Back propagation network training 
 Time delay neural network 
 Neural network systems identification with 
neural network damage detection 
Other techniques  
 Time history analysis  
 Evaluation of frequency response functions 
(FRF) 
Natural frequency analysis, used for damage sensitive feature extraction, has been the 
most common method in the field of VBDT. Natural frequency based methods rely on 
the premise that structural damage causes a reduction in the structural stiffness, which in 
turn changes the resonant frequencies. However, extensive research has proven that 
natural frequencies are a poor damage indicator, because they generally have low damage 
sensitivity (Lee et al., 2004) and they do not convey any spatial information that could be 
used for damage localization (Farrar et al., 2001).  
Methods based on the evaluation of mode shapes have been more successful in 
detection and localizing damage. In general, mode shapes are able to provide spatial 
information about the structure which can be used to localize damage. However, these 
methods have two major practical drawbacks which make mode shape methods an 
6 
 
unappealing choice. According to Doebling et al. (1996), there is a lot of difficulty in 
taking satisfactory mode shape measurements; as well as obtaining the number of 
measurements required to provide sufficient resolution for damage localization.  
Structural matrix methods, which are mainly based on the investigation of structures` 
stiffness and flexibility matrices, have also been investigated for damage detection 
applications. In these approaches, a change in the stiffness or flexibility matrix before and 
after damage is considered as an indication of damage. Structural matrix methods also 
suffer from practical issues in terms of an inability to take precise and complete 
measurements of all modes of a structure (Lee et al., 2004). 
Machine learning refers to computational intelligence methods which have learning 
capability (Adeli and Hung, 1995). Among these methods, neural networks and genetic 
algorithms have drawn attention of many researchers for structural damage identification 
due to their ability to cope with uncertainty, insufficient information and noise (Adeli and 
Hung, 1995). However, these techniques have four major challenges in practical 
applications: brittleness, lack of Meta knowledge, knowledge acquisition, and validation 
(Melhem and Nagaraja, 1996). 
The last category listed in Table 1 refers to VBDT methods that incorporate time 
history analysis and the frequency response of structures. Within this category, statistical 
processes are used to analyze the time history signals in order to identify damage 
assuming that changes in the mean and/or variance of the structure’s vibration pattern 
result from damage. The statistical methods are limited to identification of change in the 
statistical parameters and do not convey any information about location nor damage 
severity (Lee et al., 2004).  
Methods based on the frequency response functions (FRF) are an extension of mode 
shape curvature methods (Lee et al., 2004). The FRF of a structure in all frequencies in 
the measurement range is used to detect changes in stiffness assumed to be caused by 
damage (Lee et al., 2004). In practice, FRFs have the advantage of measurement 
simplicity as opposed to mode shapes; however, experimental studies on the FRF based 
damage detection method have demonstrated unreliability (Lee et al., 2004).   
  
7 
 
1.3   MOTIVATION 
The limitations of the majority of the previous VBDT methods has prompted the 
emergence of a new category of time history analyses which includes damage detection 
methods using signal processing techniques such as the Hilbert-Huang Transform (HHT). 
The objective of these methods is the extraction of damage metrics based on parameters 
obtained from the chosen signal processing technique.  
Introduced by Huang et al. (Huang et al., 1998), the HHT is in part based on the 
Empirical Mode Decomposition (EMD). Empirical Mode Decomposition decomposes a 
real signal into a collection of simpler modes or Intrinsic Mode Functions (IMFs). Each 
IMF has unique characteristics and represents a major component of the original signal. 
Damage sensitive features can subsequently be extracted from the IMFs and used for 
damage identification.  
The advantages of HHT are its simplicity, its ability to deal with non-stationary and 
nonlinear signals, and its capability of extracting buried details within a signal (Rezaei, 
2010). HHT can provide a lot of information about the structural dynamic behavior of the 
system which allows damage sensitive features to be discovered and utilized. Recently, 
SHM methods incorporating the HHT have been growing in popularity and have been 
successfully applied in various damage detection applications. 
One SHM methodology based on HHT that has recently been developed is a novel 
damage index, referred to as the Empirical Mode Decomposition-Energy Damage Index 
(EMD-EDI), created by Cheraghi et al. (2005). This index uses the energy of the first 
IMF of the response signals before and after damage. The index, therefore, reflects any 
deviations in the structural integrity of the structure, thereby providing a means to detect 
the presence of damage.   
Cheraghi et al. (2005) investigated the validity of this method by conducting an 
experiment on adhesively bonded joints in simply supported plastic pipes. Further 
experimental studies using the same method were completed by Cheraghi and Taheri 
(2007) to detect corrosion defects in a cantilevered aluminum pipe, Rezaei and Taheri 
(2009) to detect corrosion damage in a cantilevered steel pipe, Rezaei and Taheri (2010b) 
to detect notches in beams, and Rezaei and Taheri (2010a) to detect mid-span girth weld 
8 
 
cracks in a steel pipe. The EMD-EDI has not yet been experimentally validated for use 
with bolted pipeline joints.  
1.4   OBJECTIVE 
The object of this thesis was to explore the application of the Empirical Mode 
Decomposition Energy Damage Index, coupled with the use of piezoelectric sensors, and 
its integrity for detecting damage in bolted pipeline joints. This study serves to further 
expand the previous work accomplished on this method by focusing on a different type of 
structural damage than has been studied to date. For the purpose of this study, 
experimental testing on four different bolted pipeline joints was completed using two 
different materials (namely steel and PVC). The type of damage considered was loss of 
clamping force, the most common cause of bolted joint failure. This was achieved by bolt 
loosening and/or removal.  
1.5   THESIS LAYOUT 
This thesis is divided into seven chapters; of the seven, three chapters include 
publications generated in the course of this thesis.  
Chapter 2 guides the reader through a detailed literature review which begins with an 
introduction to bolted joints, which is then followed by a summary of the joint failure 
problems facing owners or uses of bolted joint systems. Next, the traditional damage 
detection methods used for bolted joints are outlined. Subsequently, the topic of structural 
health monitoring, as applied to bolted joints, is addressed including the use of 
piezoelectric transducers in SHM. Afterward, an overview is presented of the Hilbert-
Huang transform and its application to structural health monitoring. Concluding the 
chapter is a presentation of the EMD-EDI, followed by its application to pipelines for the 
purposes of structural health monitoring.   
Chapter 3 details the experimental investigation performed on an 8-bolt steel pipe 
joint, using three different combinations of piezoceramic sensors. The objective of this 
study was to detect varying degrees of loosened bolts using the EMD-EDI. A description 
of the experimental setup, vibration testing procedure, data analysis and damage 
9 
 
detection process is presented. The experimental set-up involved a simply supported steel 
pipe jointed at the mid-section using an 8-bolt flanged joint.  
Chapter 4 presents a conference paper entitled “Detection of an in-Service 
Condensation Pipeline Joint” which was written for the SPIE 2010 conference in San 
Diego, California. This paper details the experimental testing performed on an in-service 
joint located underground at the Central Services, Dalhousie University. The objective of 
this study was to detect varying degrees of loosened bolts using the EMD-EDI. A 
description of the experimental setup, vibration testing procedure, data analysis and 
damage detection process is presented. 
Chapter 5 is a prepared journal article entitled “Damage Detection of a Pressurized 
Steel Pipeline Joint using an Empirical Mode Decomposition-Based Energy Damage 
Index”. The objective of this study was to detect loosened bolts using the EMD-EDI, and 
it presents a detailed description of the experimental setup, vibration testing procedure, 
data analysis and damage detection process. The experimental set-up involved a free 
hanging pressurized steel pipe jointed at the mid-section using a 12-bolt flanged joint.  
Chapter 6 provides an in depth report on the experimental tests performed on an 8-bolt 
PVC bolted pipeline joint. Once again, the objective of this study was to detect loosened 
bolts using the EMD-EDI. A detailed description of the experimental setup, vibration 
testing procedure, data analysis and damage detection process is presented. The 
experimental set-up involved a free hanging PVC pipe jointed at the mid-section using an 
8-bolt flanged joint. The same tests were also complete for a cantilevered boundary 
condition.  
As a conclusion to this thesis, Chapter 7 summarizes the experimental work 
completed, and gives a review of the conclusions arrived at throughout this work. 
Furthermore, recommendations for future work are suggested.  
 
 
 
 
10 
 
CHAPTER 2 
LITERATURE REVIEW  
 
2.1 BOLTED JOINTS  
Bolted joints are the most commonly used components employed in the construction 
of civil structures and mechanical systems (Jiang et al., 2003). Often, bolts are the 
fastener of choice because they are a simple, low-cost, method of joining two 
components together, particularly in situations where disassembly and reassembly is 
expected to be necessary. Moreover, if a specific clamping force is required, there is little 
other choice beyond that of a bolted joint (Bickford, 1995). However, even after centuries 
of widespread use, bolted joints are still not well understood.  
The bolted joint is still widely viewed as a low-tech, low priority item and its 
perceived importance is directly reflected in the awareness and attitudes of many of those 
responsible for the integrity of bolted joint connections (Metzger, 2004). There is little or 
no difference between the function of a welded joint and a bolted joint, nor between the 
consequences of potential failure of either joint. Nevertheless, for many years industry 
has set extensive controls over material choice, weld specifications, technician training 
and integrity testing of the welding operation, whilst in many cases effectively ignoring 
bolted joints (Metzger, 2004; Noble, 2006). 
Due to inadequate awareness of the problems and issues related to bolted joints in the 
past, there have been many failures, including some catastrophic events. Oil drilling 
platforms have tipped over, airplane engines have failed, roofs have collapsed and 
astronauts have died because of bolted joint failures (Bickford, 1995). The Energy 
Institute of Great Britain reported that leaking joints were the main cause of hydrocarbon 
releases on UK Continental Shelf offshore installations and that similar concerns existed 
for many onshore facilities handling petroleum and other dangerous substances (Energy 
Institute (Great Britain), 2007).  
11 
 
Some of these failures have led to extensive news coverage. All of these failures have 
resulted in monetary losses for the involved parties. These impacts have led to industries 
becoming increasingly aware of the potential hazards posed by bolted joints and has 
resulted in a great deal of research into the mechanisms of bolted joint failures. It is now 
known that the failure of a bolted joint involves a large number of variables that are 
difficult if not impossible to predict and/or control (Bickford, 1995).  
2.2 BOLTED JOINT FAILURE  
There are three basic, overall factors which determine the reliability of a bolted 
connection and the probability of a joint failing in-service (Corbett, 1998):  
1. Joint design  
2. Bolt/component quality 
3. Achieving design bolt tension during installation 
Assuming a joint is correctly designed for the in-service loadings, and the components 
provided are initially satisfactory, achieving the correct bolt tension during installation 
may be reasonably difficult. Corbett (1998) states that “90% of all bolted joint failures 
can be attributed to incorrect bolt tightening” because “the vast majority of bolted joints 
in engineering are tightened in an uncontrolled manner”. This may explain why the 
durability of a system is often proportional to the number of bolted joints used (Jiang et 
al., 2003). 
The first few factors that may affect in-service bolt tension have one common link; 
they do not have anything to do with the joint itself. The first significant variable that 
affects the life of the bolted joint is the technicians that assemble the joint. The answers 
to whether the technicians are well trained, if they are adequately supervised and if they 
actually care about what they are doing would have a considerable impact on the joint 
integrity. Furthermore, the conditions that the technician has to work under to perform 
the assembly also have a major impact. Can the technician easily see and reach all the 
bolts in the joint; and, do the environmental conditions favor good concentration?  
 
 
 
12 
 
Common assembly errors made by bolting mechanics are as follows (Noble, 2006):    
 Fitting the wrong gasket 
 Not inspecting gasket for handling/storage damage 
 Application of grease to gasket 
 Not lubricating bolt threads 
 Not lubricating nut spot face 
 Using an incorrect lubricant 
 Leaving dirt or grit on gasket face 
 Greasing the groove on a ring type joint 
 Re-using gaskets or rings 
 Fitting nuts the wrong way so that stampings gall the flange face 
 Not setting bolts correctly for tightening 
 Leaving tape on gaskets 
 Not inspecting and cleaning flange faces 
When the joint is assembled or reassembled, time should be set aside for the joint 
components to be individually inspected for the correct specification and damage so that 
future joint problems may be prevented at the source. A prior history of the joint will also 
aid in preventing future problems.  
A bolted joint ready for service needs to have attained the designed clamping force, 
which is reached by achieving the correct bolt tension in all the bolts of the joint. The 
following list presents additional factors which have a considerable effect on the final in-
service bolt tension and thus the clamping force and the integrity of the joint. Four of 
these factors will be expanded on further in the following sections.  
1. Initial preloads: the load developed in the bolts from torquing or tensioning 
during the initial tightening process. 
2. Sequence/procedure: the procedure that is used to tighten the bolts of a joint can 
significantly affect the final in-service bolt tension such as if the bolts are 
tightened all in one pass at the final torque or in several passes in increments.  
13 
 
3. Residual preloads: the preload left in the bolts at the end of the assembly process 
after embedment, elastic interactions and short-term relaxation.  
4. External loads: there may be external loading acting on the joint which may add 
or subtract from the bolt tension. Loadings such as pipeline pressure, weight of 
additional structural components, etc. If these loadings are not taken into 
consideration in the design, the clamping force may not be sufficient to maintain 
joint integrity.  
5. Service conditions: the local environment surrounding the joint may cause 
substantial problems in the operating conditions. Conditions such as severe 
temperatures may cause differential expansion or contraction, or a wet 
environment to cause corrosion. 
6. Long-term relaxation: relaxation may be caused by corrosion, stress relaxation or 
creep, and vibration.  
7. Component quality: proper size, material, and satisfactory condition.  
When a joint is assembled and put into service, there are hundreds of factors that can 
modify the initial and residual preload in the individual fasteners (Bickford and Nassar, 
1998). Without trained technicians familiar with these variables a joint may be put into 
service without a sufficient clamping force. Consequently, the bolted joint will be unable 
to withstand the necessary loadings and will have, at the minimum, a shortened life span 
before leakage and or failure occurs.  
2.2.1 Initial Preload   
An incorrect preload on the bolts may cause a variety of joint problems affecting the 
behavior and life of the bolted joint. According to Bickford (1995), the following list 
summarizes the main issues resulting from incorrect preload:  
1. Static failure of the fastener: the application of too much preload may cause the 
bolt to break or the threads to strip.  
 
14 
 
2. Static failure of the joint member: excessive preload may crush, gall, warp or 
fracture joint members (such as castings and flanges).  
3. Vibration loosening of the nut: in most applications, proper preload can eliminate 
vibration induced loosening of the nut. 
4. Fatigue failure of the bolt: a higher preload on the bolt will usually improve the 
fatigue life of the fastener.  
5. Stress corrosion cracking: may cause bolt breakage and is encouraged by a high 
preload if it is above a specific threshold.  
6. Joint separation: a low preload may allow the joint to separate, resulting in 
leakage and/or joint failure. 
7. Joint slip: inadequate preload may cause the joint to slip under shear loads 
causing misalignment, cramping, fretting corrosion, fatigue, or bolt shear. 
Too much preload may be just as troublesome as too little preload. Furthermore, a 
uniform preload for all the fasteners in a joint is also important to prevent joint failure. 
Factors contributing to an incorrect initial preload include friction, embedment, 
nonparallel joint surfaces and the method used for bolting the joint.  
2.2.1.1 FRICTION  
Even though a bolted joint would not work without friction, it is also a significant 
factor that contributes to a smaller preload than desired if torque is used to assemble the 
joint. Figure 2-1 illustrates the relationship between torque, friction and the resulting bolt 
tension.  
Ninety percent of the torque applied to tighten a bolt is typically absorbed by friction. 
Approximately 50% is lost to interface friction between the nut and joint, and another 40% is lost 
to friction between the nut and the bolt thread surfaces. Figure 2-1 suggests that only 10% of the 
applied torque is actually converted into bolt preload and that any increase in friction has a 
significant impact on the amount of torque converted to bolt tension.  
 
 
15 
 
 
 
 
 
 
 
 
 
Figure 2-1  Relative relationship between torque, friction and bolt tension (adapted from 
Bickford, 1995). 
Listed below are factors that affect the torque-preload relationship by possibly 
affecting the available friction (Bickford, 1998b): 
 Lubrication: amount, type, temperature and location  
 The hardness of the bolt, nut, joint members, washers 
 Surface finishes of each part 
 The type of materials used for each part 
 The thickness, condition, consistency and type of plating 
 The manufacturing process used to form the threads 
 The thread form 
 Hole finish, alignment, concentricity 
 Burrs around edges of holes 
 Whether or not the holes have been countersunk 
 Type, number, size, location, and hardness of washers used 
 Fit of wrench on nut or bolt 
 The order in which the torque is applied to the nut or bolt 
 
 
 
Torque converted to 
Tension: 10% 
Nut & Joint Friction 
50% 
Thread Friction 
40% 
16 
 
Additional geometric factors that may affect the friction and therefore the final bolt 
preload are (Bickford, 1998b): 
 Fit between threads: tighter-fitting threads result in larger contact areas 
than loose fitting threads. Higher contact pressure result in greater friction 
forces. 
 Hole clearance: larger holes with greater clearance will reduce the contact 
area between the joint face and the bolt head/nut face increasing the contact 
pressure. 
 Hole perpendicularity: the angle of the bolt hole to the joint face will affect 
how well the bolt head and nut sit against the joint and if any increase in 
contact pressure would occur.  
 Angle of threads: if the thread angle of the nut does not perfectly match the 
thread angle of the bolt, the contact area would be reduced.  
 Hole interference: if the holes in the mating surfaces are misaligned, then 
when the bolt is tightened, it will scrape against the hole wall, contributing 
additional friction to the tightening process. 
Further factors that also have an effect on friction will be mentioned throughout the 
following sections.  
2.2.1.2 EMBEDMENT  
In some situations, as shown in Figure 2-2, there may be hole interference due to an 
undersized bolt hole caused by manufacturing tolerances in the flange and/or bolt. 
Additionally, the interference may be caused by flange misalignment. These frictional 
and/or embedment constraints cause some of the tension in the bolt to be lost as the bolt 
is forced past the walls of the hole, thereby reducing the final joint clamping force. 
 
 
 
 
 
17 
 
R R 
Tight 
Fit 
Normal 
Fit 
F  
 
 
  
  
 
 
 
 
Figure 2-2  Bolt hole interference (adapted from Bickford, 1995). 
2.2.1.3 NONPARALLEL JOINT SURFACES 
Another factor that can reduce the clamping load is misalignment of the mating 
flanges of the joint. If the flanges are initially nonparallel or simply too far apart, and the 
bolt tightening process is relied upon to bring flanges into contact with each other, it 
would require torque to advance the flanges towards each other. If this additional 
required torque was not taken into account by the designer, then there would not be 
sufficient torque left to provide the desired clamping force.  
2.2.1.4 BOLTING METHOD  
The type of tool used to tighten the joint will have a significant impact on the final 
preload of each bolt, if the bolts are not being inspected for the final tension. Table 2-1  
lists the accuracy of a variety of bolting methods. If a method such as direct tension 
indicators and ultrasonic extensometers are being used then even a crude bolting method 
can yield accuracy within ±2% (Sturdevant, 1998). It should also be kept in mind that the 
torque accuracies listed above are for a midscale reading, the variation in torque accuracy 
at the lower and high end of the gauge scale will be even greater.  
  
18 
 
Table 2-1  Bolting method accuracy (adapted from Sturdevant, 1998). 
Bolting Tool Torque Accuracy (%) Preload Accuracy (%) 
Hydraulic torque wrench ±3-5 ±23-28 
Hand wrench with torque 
multiplier 
Unknown ±70-150 
Pneumatic impact wrench Unknown      -300 to +150 
Manual slugging wrench Unknown -48 to +50 
Hydraulic stud tensioners N/A ±20 
Dial or click torque wrench ±5 ±60-80 
2.2.2 Residual Preload   
The common assumption made when bolting a joint is that there is an equal and 
opposite relationship between the applied preload, or tension in the bolts, and the 
clamping force between joint members (Bickford, 1998a). However, there are multiple 
factors that may contribute to an incorrect residual preload in the bolts at the end of the 
assembly process. 
2.2.2.1 SHORT-TERM RELAXATION  
During a relatively short period of time after bolts are initially tightened, individual 
bolts will invariably lose some of their tension due to relaxation. Bickford (1998a) 
explains that this short-term relaxation generally occurs in a component, such as a bolt or 
gasket, or a portion of a component (such as the threads in a nut) that is loaded past its 
yield point. Once past the yield point, creep will begin to occur resulting in joint 
relaxation. Although there is no specifically known duration during which this occurs, the 
general pattern of relaxation is illustrated in Figure 2-3.  
 
 
 
 
 
Figure 2-3  General trend of short-term relaxation (adapted from Bickford, 1995). 
B
o
lt
 T
en
si
o
n
 
Time 
19 
 
There are many causes and contributing factors to the amount of relaxation a joint will 
experience. Therefore, it is incredibly difficult to predict how much relaxation will occur 
in any specific joint. 
2.2.2.1.1 SOURCES OF SHORT-TERM RELAXATION  
One of the most common causes of short-term relaxation is embedment (Bickford, 
1998a). The surfaces of any component are never perfectly flat; under magnification, the 
surfaces of bolt threads, bolts, washers, nuts and flanges are uneven with a series of high 
points and low points (Figure 2-4 (a) and (b)). If the surface unevenness is sufficiently 
large, when loading is initially applied the high points of a surface will undergo 
extremely large pressures since they only constitute a small portion of the surface.  
Consequently, plastic deformation will occur and the high points will flatten until a 
sufficient amount of the surface is in contact thus decreasing the concentrated pressure 
and stabilizing the system. Embedment is of particular concern with new parts since the 
surfaces have not suffered any wear; previously used parts do not suffer from as much 
embedment. If embedment occurs with a conical joint surface (Figure 2-4 (c)), the 
diminished bolt tension is even greater than that for a flat joint surface (Bickford, 1998a).  
              
                 a)                                              b)                                                   c) 
Figure 2-4  Embedment a) Thread b) Joint surface c) Conical joint surface (adapted from 
Bickford, 1995). 
An additional cause of short-term relaxation is the geometry of the nuts and bolts. An 
undersized bolt or oversized nut will result in little contact between threads and therefore 
may results in substantial plastic deformation. Moreover, if the thread engagement length 
20 
 
is too short, the thread contact area will be insufficient resulting in excessive relaxation. 
Furthermore, Bickford states (1998a) that the contact faces of nuts and bolt heads are 
never exactly perpendicular to the axis of the threads or to the axis of the bolt hole. 
Consequently, there will be little contact between threads, illustrated in Figure 2-5, 
causing plastic deformation. 
 
Figure 2-5  Poor thread engagement (adapted from Bickford, 1995). 
Loss of preload may also come from the geometry of the bolt hole. If the fillet joining 
the body and head of the bolt contacts the edge of the bolt hole as shown in Figure 2-6 
(a), the edge of the bolt hole will break down when the bolt is tightened. Once it does, the 
amount of stretch in the bolt will lessen, and bolt tension will decrease. As well, if a bolt 
is paired with an oversized hole without the benefit of a washer (illustrated in Figure 2-6 
(b)), there will only be a small contact surface available for the bolt/nut head.  
                                  
                                        a)                                                     b)                                    
Figure 2-6  Poor geometry a) Contact of edge with fillet b) Oversized bolt hole (adapted from 
Bickford, 1998a). 
21 
 
Depending on the surface material, the bolt/nut head may embed in the joint surface 
causing relaxation. 
If a part is soft, due to a manufacturing error, such as improper heat treatment or using 
incorrect material, creep and relaxation will again occur. If a fastener is bent as it is 
tightened, as in the case of flange misalignment, one side of the fastener will undergo 
higher stresses. If these stresses exceed the material’s yield limit, the bend in the fastener 
will cause greater than normal embedment (Bickford, 1998a). 
2.2.2.1.2 FACTORS AFFECTING SHORT-TERM RELAXATION  
The degree of relaxation that will occur in a particular bolt or joint as a whole depends 
on some secondary factors in addition to the primary causes listed above. The effects of 
all the relaxation sources listed above will be greater in a more complicated joint. A joint 
that has more members and more contact surfaces will suffer from more relaxation. It is 
also important to remember that relaxation takes time. Joints that are tightened very 
quickly will not have any time to relax during the process. If the bolts of a joint are 
tightened in passes, the components are given time to stabilize which allows the 
necessary bolt tension to be reached. 
An additional factor may be the process of tightening one bolt at a time (Prodan, 1974, 
as cited in Bickford, 1998a). The initially tightened bolt may undergo higher concentrated 
stress levels resulting in a greater level of relaxation for that fastener. Furthermore, if 
joint members are bent out of shape, tightening one bolt may cause additional stress in 
other fasteners, possibly resulting in further relaxation.  
Another of these factors is bolt length. Because of a smaller thread to length ratio, 
long, thin bolts will relax to a lesser degree than shorter bolts with a larger diameter. If 
both bolts have the same amount of threads, the embedment will be the same for both; 
however, preload loss is proportional to the change in length, so shorter bolts will cause 
greater preload losses. Bushings are often used to take advantage of this fact by 
effectively increasing the length of the joint which allows longer bolts to be used, 
resulting in less joint relaxation (Bickford, 1998a).  
22 
 
2.2.2.2 ELASTIC INTERACTIONS BETWEEN BOLTS  
During the assembly of a multi-bolt joint, tightening one bolt will cause changes in the 
preload of the other bolts in the joint. The term used by Bibel (1998) to describe this 
phenomenon is “elastic interactions”. When one bolt is tightened, the joint is compressed, 
allowing the previously tightened bolts to relax a little. With each subsequently tightened 
bolt, each previously tightened bolt will react the same way. A technician unaware of this 
fact may assume a joint has the correct preload if the correct tightening procedure is not 
followed.  
2.2.3 Service Conditions  
Depending upon the design of the joint and the chosen components, the environmental 
conditions may have a significant impact on the joint once it is put in-service. For 
example, gaskets may be fatigued by fluctuating internal pressures, or they may be 
chemically altered or suffer changes in their elasticity and strength due to heat. Elevated 
temperatures will reduce component strength and thermal fluctuations may cause 
differential expansions and contractions in various joint components changing the loading 
conditions and possibly contributing to eventual joint failure. Anything that causes a 
gasket to shrink may cause a loss of clamping force, leading to joint instability and gasket 
blow out.   
Components are also susceptible to hydrogen embrittlement. Hydrogen embrittlement 
can occur during the manufacturing process or during in-service operation by processes 
such as electroplating and cathodic protection. Hydrogen embrittlement is caused by the 
diffusion of hydrogen atoms into a material and their subsequent assembly in cavities 
which exerts pressure on the material. The buildup in pressure causes the material to 
become brittle and susceptible to cracking.  
In addition to temperature fluctuations, pressure fluctuations, chemical alterations and 
hydrogen embrittlement, two more environmental causes of component damage are 
fatigue and corrosion. 
23 
 
2.2.3.1 BOLT FATIGUE   
If a joint is subject to cyclic loading, the affect that bolt fatigue could have on the life 
of the joint must be considered. Proper maintenance of the clamping load on the joint is 
extremely important to bolt fatigue life. Kull (1998) states that in well designed joints, 
where the clamping load has been maintained, the bolts may only see 5-10% of the total 
load on the joint. Therefore, all of the factors that reduce the required clamping force 
would have an effect on bolt fatigue life.  
Joint design, the surrounding environment, and the type of loading on the joint will all 
have an effect on the bolt fatigue. The joint stiffness ratio describes the relationship 
between the elasticity and geometry of the bolted members to those of the bolt. The 
stiffer the joint and the more elastic the bolt, the less variable load the bolt would see 
when a working load is imposed on the joint (Kull, 1998). Accordingly, soft gasketed 
joints may have shorter fatigue lives than metal joints having a greater stiffness. 
Environment conditions such as high, low or fluctuating temperatures and/or a 
corrosive situation may also reduce bolt fatigue life. If the in-service loading results in 
the transverse movement of the joint (evidence of insufficient clamping force), the bolts 
may loosen significantly, thus affecting bolt fatigue life (Junker, 1972).  
If plated bolts need to be used for a corrosive environment or if the bolt threads have 
been damaged in any way, the fatigue life of the bolt is also reduced (Kull, 1998). Any 
manufacturing defect in the bolt will also have an effect, including any anomalies in the 
geometric configuration of the bolt.  
2.2.3.2 CORROSION  
Corrosion is the deterioration of a material due to electrochemical reactions with its 
surrounding environment. There are many different types of corrosion that may occur in a 
joint: 
 General (uniform corrosion): occurs over the entire exposed surface of the 
metal due to microscopic anodes and cathodes. 
 Galvanic corrosion: occurs when two dissimilar metals are in electoral 
contact and one metal corrodes preferentially. 
24 
 
 Pitting corrosion: localized corrosion due to high localized energy at 
inhomogeneities in the structure.   
 Crevice corrosion: occurs in crevices where liquid may gather and become 
stagnant. Crevice corrosion is commonly found under rivets, bolts, gaskets 
and between valve disks and seats.  
 Intergranular corrosion: occurs at grain boundaries and weakens the alloy.  
 Stress corrosion cracking (SCC): may occur unexpectedly when a metal 
experiences tensile stress in a corrosive environment. The stress may be 
from assembly, in-service loadings or manufacturing residuals. This type of 
corrosion is very chemically specific; only certain combinations of metal 
and environmental conditions would result in SCC.  
 Fretting corrosion: occurs at the contact surfaces of components under 
load. When components are not designed to move relative to one another as 
may happen with vibration and/or joint slippage, metal fragments or 
surface oxides may break loose and become an abrasive, wearing away the 
contact surface due to the load.  
Corrosion may be controlled by using corrosion resistance materials, protective 
coatings, inhibitors, and cathodic protection. Furthermore, the design of the joint itself 
would have a large impact; whether dissimilar metals are used side by side, or there are 
any drainage options for stagnant liquid, if corrosion is taken into account when 
designing the thickness of the components, and if the production of areas of high stress 
concentration is avoided.  
2.2.4 Long-Term Relaxation   
There are several factors affecting long-term in-service bolt tension relaxation. Among 
these are creep, stress relaxation and dynamic loadings. Creep, the tendency of a solid 
material to slowly deform under constant loading, may occur in materials after long term 
exposure to levels of stress that are below the yield strength of the material. Stress 
relaxation, which is temperature dependant, occurs when the elastic deformation becomes 
25 
 
plastic or permanent and the loss of elasticity lowers the clamping force. Another factor 
affecting long-term relaxation is the self-loosening of bolts from dynamic loadings.  
Joints must often exist in environments where they are subjected to dynamic loadings. 
Fasteners in a joint may loosen (self-loosen) if they are exposed to vibration and/or 
shocks, thermal cycles, and flexing of joint members. The self-loosening phenomenon is 
also observed if the joint undergoes external transverse or shear loading, or any type of 
cyclical loading that causes relative slip at the joint. The two most common modes of 
failure of threaded fasteners subjected to dynamic loads are fatigue and vibration induced 
loosening (Pai and Hess, 2002) and vibration is the most common cause of self-loosening 
(Bickford, 1995). 
Since 1945, there has been a great deal of experimental research into the causes of 
dynamic bolt self-loosening. It has been found that loosening begins with a gradual 
relaxation of the initial preload until a critical clamping force is reached. During this 
early stage, preload loss is due to material deformation (Jiang et al., 2003). Once the 
critical point is reached, the nut will begin to rotate and back off the bolt; this would 
occur very rapidly. For further detailed explanation of the research performed on the 
mechanisms of self-loosening as well as methods of arresting this failure mode, the 
reader is referred to the Handbook of Bolts and Bolted Joints, (Bickford and Nassar, 
1998). 
2.3 DAMAGE DETECTION METHODS FOR BOLTED JOINTS 
Until recently, the damage detection methods used to detect joint problems have all 
required individual effort, time and training on the part of a technician. Table 2-2 lists the 
types of damages previously mentioned and the typical inspection techniques used for 
damage detection. The majority of these techniques fall under the category of non-
destructive evaluations (NDE). Non-destructive testing are methods which allow the 
examination, and evaluation, of structures, materials, and components without causing 
any damage or destruction to the object being tested while providing a means to ensure 
product reliability and quality (CINDE, 2010).  
  
26 
 
Table 2-2  Inspection techniques for joint damage (adapted from Energy Institute (Great 
Britain), 2007). 
Damage Type  Typical Conditions Inspection Technique 
General corrosion Exposed Areas 
Visual, sample removal, Cylindrical 
Guided Wave technique 
Galvanic 
Corrosion 
Dissimilar metals  
(flanges, bolts, gaskets) 
Visual 
Localised bolt 
corrosion 
Dissimilar metals, exposed areas 
Sample removal, visual, Phased Array 
UT*, CGWT** 
Crevice corrosion Exposed areas Visual, sample removal 
Fatigue 
Joints subject to vibration, cyclic 
loading 
Visual, Phased Array UT 
Creep High temperature applications Time-of-flight UT 
Stress corrosion 
cracking 
A combination of a chloride-
containing environment, 
susceptible material and tensile 
stress 
Phased Array UT 
Hydrogen 
embrittlement 
Hydrogen can form on surface 
during manufacturing or be caused 
by Cathodic Protection 
Visual, highlighting any corroded 
High Strength Fasteners for 
replacement 
Flange face 
Pipework containing a corrosive 
medium, dissimilar materials 
Intrusive visual, UT 
*UT stands for Ultrasonic Testing; **CGWT stands for Cylindrically Guided Wave Technique 
If damage evaluation is not feasible or possible through non-destructive techniques, 
then a sample of bolts may be removed for destructive testing. For bolts that have been 
removed for one reason or another, black light testing can be used to detect stress 
cracking of threads and body of bolts to be reused.   
2.3.1 Visual Inspection 
Visual inspection is the most common method of performing non-destructive damage 
detection. A straightforward method, it relies on the inspector’s experience and judgment 
for accurate damage assessment. The only equipment necessary for a visual inspection is 
a magnifying glass or a microscope if desired by the inspector. This method has several 
limitations, in that only the external parts of the joint are visible to the inspector. 
Therefore, visual assessment could only detect obvious damage such as extremely loose 
27 
 
bolts and surface corrosion. Any damage extending beyond the surface or damage 
without any signs on the joint surface will remain unnoticed.  
2.3.2 Cylindrical Guided Wave Technique (CGWT) 
The cylindrical guided wave technique can be used to detect corrosion wastage and 
cracks in cylindrical objects. The CGWT employs transducers to emit a continuous or 
pulse-wave signal. Low-frequency ultrasound waves are guided down the length of the 
cylindrical component being tested by its boundaries. The reflected signals that the 
transducers receive are then used to detect and locate damage based on arrival times, 
velocities and frequencies. However, because this method is based on the constant cross-
section of the object being tested, it cannot differentiate between damages. The CGWT 
requires skilled operators due to subjective results. The advantages to CGWT are that 
only a small section of a pipeline needs to be accessed in order to test a relatively large 
area. This permits testing of buried or insulated components.  
2.3.3 Ultrasonic Testing (UT) 
Ultrasonic testing uses pulses of ultrasonic waves to detect defects or flaws due to 
their different acoustic nature (CINDE, 2010); it can be used to test any material that 
transmits mechanical vibrations for damage. Ultrasonic testing requires expensive 
equipment and well trained operators; operators should have sufficient experience since 
the interpretation of results is often difficult. Concerning bolted joints, ultrasonic testing 
may be used to test flange faces for corrosion and erosion using shear wave transducers.  
Phased Array Ultrasonic Testing may also be used to detect thread wear and cracking 
through the use of a transducer array which allows the creation of a visual image of the 
inspected material. Time of flight Ultrasonic Testing (TOF) is used to measure bolt 
preload by determining bolt elongation. Likewise, it can also be used to detect creep. 
Time of flight UT works by introducing a sonic pulse at one end of a bolt and measuring 
the time until the echo of the signal returns from the opposite end.  
28 
 
2.4 STRUCTURAL HEALTH MONITORING  
A vast amount of civil infrastructure has been built over the last century such as 
buildings, bridges, dams, water supply lines, tunnels, and off-shore platforms. Over time, 
these structures withstand damage caused by fatigue, earthquakes, strong winds, traffic 
and other environmental impacts, causing deterioration. Early detection of any structural 
damage or deterioration prior to local failure may prevent a catastrophic failure of the 
entire structure (Park et al., 2006) thereby keeping the general public from harm.  
With the current increased interest in ascertaining the safety and the serviceability of 
aged structures combined with the economics of repair and maintenance, efficient 
structural health monitoring methods have become essential. Because bolts are among the 
most prevalent form of fastening employed in structures, they make an attractive subject 
for the field of structural health monitoring (Mascarenas et al., 2005). Furthermore, with 
industries focused on the most efficient and economical design of products, redundancy 
is no longer such an attractive option which puts more emphasis on the need for a well 
developed SHM system.   
The traditional method of inspecting bolted joints has been by visual assessment or 
other assessment that require onsite personnel. For a system with a limited number of 
bolted joints, these methods work well and are relatively efficient. However, when one 
considers that some systems, such as an offshore oil platform, can contain up to 15,000 
bolted joints on critical services alone, the need for an online system becomes clear 
(Noble, 2006). In comparable situations, individual inspections would be extremely 
labor-intensive and expensive; they may also possibly expose the inspectors to hazardous 
conditions and uncomfortable working environments (Ritdumrongkul et al., 2004). 
Additionally, there may be a need for a system shutdown in order for the inspection to 
take place at the cost of production.  
2.4.1 Piezoelectric Sensors 
A very important consideration for an online structural health monitoring system is the 
choice of sensors. Over the past several decades, there has been significant advancement 
in sensing technology. There are currently a variety of sensors that may be used with 
29 
 
structural health monitoring systems. Commonly used sensors are optical fiber sensors, 
accelerometers, displacement transducers, velocity transducers, and piezoelectric sensors.  
When selecting a sensor, it is important to keep in mind that the sensor should be 
robust enough to remain in good condition under various loading and environmental 
conditions (Noman, 2008). Furthermore, according to Park et al. (2006), sensors used to 
monitor structures should be inexpensive as many are required, should preferably be 
wireless with some data processing capabilities, and should have minimal power 
requirements. These capabilities will aid in the development of an efficient and low-cost 
SHM systems. It is also important to consider the capability of the sensor to measure the 
desired component of a particular quantity (acceleration, velocity and displacement), 
noise factor and the effect the sensor’s mass may have on dynamics of the structure (Lee 
et al., 2004).  
Piezoelectric transducers are widely used in the field of structural health monitoring. 
Their appeal comes from numerous advantageous properties such as their small size and 
light weight. In addition, piezoelectric sensors are attractive because they do not require 
excitation, they have a high signal to noise ratio, and they can be easily bonded to a 
structure. Furthermore, they have a fast response, high sensitivity, linear behavior, high 
operational bandwidth (Riley, 2005), and can operate over a large temperature range. 
Conversely, piezoelectric transducers have a drawback in that they are brittle due to their 
crystalline structure and are relatively weak in tension and shear (Rezaei, 2010).   
Piezoelectric transducers operate on the basis of the electromechanical coupling 
exhibited by piezoelectric materials. Piezoelectric materials exhibit electromechanical 
coupling: they produce an electrical displacement with an application of mechanical 
stress, and can produce mechanical strain from an application of an electrical field (Leo, 
2007). The mechanical-to-electrical coupling is known as the direct effect and the 
electrical-to-mechanical coupling is known as the converse piezoelectric effect. 
Piezoelectric sensors and actuators take respective advantage of the direct and converse 
piezoelectric effects. Piezoelectric self-sensing actuators have the capability of acting as 
both a sensor and an actuator, thereby providing a significant reduction in the number of 
system components for certain applications (Ritdumrongkul and Fujino, 2006).  
30 
 
There are three categories of piezoelectric materials: naturally occurring materials 
such as quartz and Rochelle salt, and two types of synthetic materials. The synthetic 
piezoelectric materials have greater electromechanical abilities due to their 
manufacturing process. The synthetics are piezoelectric polymers such as polyvinylidene 
fluoride, and piezoceramics such as barium titanate and lead zirconate titanate (PZT), 
which is the most widely used piezoelectric material (Leo, 2007).  
Accordingly, piezoelectric transducers are available in two forms: piezoceramics and 
piezoelectric polymers. Piezoceramic sensors are available in different shapes such as 
plates and disks; they have a high stiffness but low cost. In contract, piezoelectric 
polymer patches are available in a variety of sizes as flexible, thin films. However, the 
polymers have greater electromechanical ability but are also more expensive than their 
counterparts.        
Due to the benefits of piezoelectric sensors, they have been widely used in structural 
health monitoring methods by surface bonding or embedment. The following sections 
will highlight many applications of piezoelectric transducers in the field of structural 
health monitoring.  
2.4.2 SHM Applied to Bolted Joints 
To surmount the issues surrounding traditional non-destructive evaluation methods 
described earlier, there has been a great deal of research performed in the development of 
real-time, in-service, structural health monitoring systems over the past decade. Many 
non-destructive evaluation (NDE) techniques for bolted joints have been and are 
currently being developed for structural health monitoring. 
For example, Lovell and Pines (1998) conducted a preliminary experimental and 
analytical investigation into using continuum wave models for the damage assessment of 
a single bolt lap joint. The system consisted of two aluminium beams bolted together and 
fixed at either end. Damage was created by altering the torque used to tighten the bolt. 
Piezoceramic actuators were employed to excite the structure in vibration, and 
piezoelectric strain sensors were used to monitor vibration around joint. Lovell et al. 
determined that their method was capable, but would be costly due to the number of 
sensors, actuators and other diagnostic tools required.  
31 
 
The objective of Stinemates et al. (2002) was to demonstrate that finite element 
analysis can be used to define the parameters of a structural health monitoring system. 
They used a finite element model to analyze a three story building with loosened bolts 
and validated their FEM model by comparing mode shapes and weights to the actual set-
up.  
Park et al. (2005) proposed the use of a PZT-induced Lamb waves-based method for 
the structural health monitoring of jointed steel plate-type structures. Their experimental 
set-up consisted of two steel plates on simple supports connected by eight bolts. Using 
two PZT patches for transmitter/sensor of the lamb waves, they used the Time-of-flight 
and wavelet coefficient as damage features to detect bolt removal.  
With the goal in mind of developing a method to detect damage in a structure from 
ambient excitation, Nichols et al. (2007) performed experiments on a composite beam 
bolted at either end to two steel plates. Excitation was provided by means of an 
electrodynamic shaker located at mid-span and damage was bolt loosening. Damage 
assessment was made using the Pierson-Moskowitz distribution and in the presence of 
strongly varying temperatures.  
Further work using auto-regressive modeling was completed by Fasel et al. (2008). 
Experiments and numerical analyses were performed to detect bolt preload reduction 
using chaotic structural excitation with a damage detection algorithm based on auto-
regressive modeling. Signals were applied to the structure using macro fiber composite 
(MFC) patches.  
2.4.3 Impedance-Based Structural Health Monitoring 
Impedance-based structural health monitoring takes advantage of the direction relation 
between the electrical impedance of a bonded PZT patch to the mechanical impedance of 
the host structure. By monitoring the host structure’s mechanical properties using the 
measured electrical impedance, changes in the structural integrity may be determined 
through changes in the electrical impedance. Impedance-based methods rely on high 
frequency structural excitations, providing a high sensitivity to minor defects.  
However, due to the complexities of the damaged structures and the difficulties in 
high-frequency analysis, only limited information about the nature of damage can be 
32 
 
obtained (Yan et al., 2007). Damage location is ascertained due to the known localized 
detection area surrounding each sensor and the damage severity is qualitatively assessed 
using a damage index which measures the change in the impedance before and after 
damage (Ritdumrongkul and Fujino, 2006).  
As a qualitative health monitoring method, the electromechanical impedance method 
has been widely studied (Yan et al., 2007) and bolted joints make an attractive subject 
due to their localized damage. For instance, Park et al. (2001) performed preliminary 
experimental testing on a single bolt lap-joint in an effort to monitor bolt preload loss in 
real time. The system consisted of two joined beams suspended from one end using 
fishing wire. To supplement the preload monitoring, they used a shape memory alloy 
(SMA) washer to restore preload loss to the joint. They successfully illustrated that the 
impedance method in conjunction with an SMA actuator may be used for condition 
monitoring and repair until a more permanent repair is possible.  
To improve upon this method and make it more economically feasible, Peairs et al. 
(2002) experimentally investigated the use of a low cost impedance method which uses a 
digital signal analyzer with a Fast Fourier transform (FFT) function instead of the very 
expensive impedance analyzer. The low cost method was tested on the same set-up used 
by Park et al. (2001) as well as on a model pipeline with bolted joints. Peairs et al. were 
as able to detect preload loss as successfully as they could using the traditional method.  
Mascarenas et al. (2005) conducted experimental research using a piezoelectric (PZT) 
enhanced washer in conjunction with the impedance method to detect the self-loosening 
mode in a moment-resisting portal frame. They used mechanical impedance matching 
between the PZT enhanced washers and the joint connections to detect changes in the 
bolt preloads. Their results demonstrated that the proposed device is only useful after 
large preload losses.  
Park et al. (2006) explored the use of an active sensing device which consisted of a 
miniaturized impedance measuring chip and a self-sensing macro-fibre composite patch 
together with the use of the electro-mechanical impedance method to detect and quantify 
damage. Experimental validation was performed using a system consisted of four bolts 
joining together two aluminium plates. Park et al. illustrated that their method has good 
damage detection capability by providing consistent results according to repeated 
33 
 
loosening and tightening. However, it was determined that this method had difficulty 
determining the damage location.  
Also using the impedance-based method coupled with another method, Rutherford et 
al. (2007) employed a non-linear feature identification technique for structural damage 
detection. The non-linear features associated with the electro-mechanical impedance 
measurements are extracted using the frequency domain autoregressive model. The 
experimental set-up was very similar to that of Mascarenas et al. (2005).  
To improve on the conventional impedance method and create a quantitative analysis, 
two different approaches have been proposed, one based on neural networks and the other 
on wave-propagation modeling (Park et al., 2003). Lopes et al. (2000) combined the 
impedance based SHM technique with an artificial neural network. The artificial neural 
networks were integrated with the aim of further characterizing the structural damage. 
Lopes et al. (2000) experimentally investigated these combined methods on a quarter 
scale bridge section with bolted joints, and were successfully able to locate and estimate 
damage severity. The damage was inflicted by loosening bolts.  
Ritdumrongkul and Fajino (2006) presented the use of PZT actuator-sensors in 
conjunction with a spectral element method (SEM) model for damage characterization. 
They performed laboratory experiments on a suspended two-joint and four-joint 
aluminum beam where damage was simulated by loosening bolts. Once the impedance 
was measured, numerical models of the structures were formulated in order to 
quantitatively locate and assess the level of damage.  
2.4.4 Vibration-Based SHM of Bolted Joints 
There has been substantial work on vibration-based structural health monitoring over 
the last several decades. For a comprehensive review of the advances in vibration based 
SHM, the reader is directed to Doebling et al. (1996; 1998) and Carden and Fanning 
(2004). Research is still ongoing, because no single method has yet been found that could 
identify every type of damage in every type of system (Carden and Fanning, 2004). 
However, considering all the research accomplished in vibration based SHM, the work on 
bolted joints has been limited.  
34 
 
Yu et al. (1999) used modal frequencies and mode shapes to identify damage in a 
frame structure with bolted joints. They used a combination of an analytical model 
coupled with experimental data to successfully detect damage. In another study, one 
approach that Caccese et al. (2004) used to detect bolt load loss was a low frequency 
modal analysis. They also investigated high frequency transfer functions between the 
actuator and sensors and high frequency transmittance functions between pairs of sensors.  
Todd et al. (2004) used a novel technique adapted from nonlinear time series 
prediction. Their approach was applied to the detection of preload loss in a bolted joint in 
an aluminum frame structure excited by an applied chaotic waveform. Todd et al. 
demonstrated that their method was able to distinguish among preload losses in the joint. 
Also employing a time series technique, Moniz et al. (2005) used a multivariate, attractor 
based approach for feature extraction from the time series information resulting from a 
system’s vibration. This approach was used to detect loosened bolts at either end of a 
composite beam bolted to a plate. 
In a change from previous experimental work, Coelho et al. (2007) focused on the 
fatigue crack growth of bolted joints of aging aircraft. To investigate the use of an 
unsupervised machine learning method, they conducted damage detection experiments on 
a bolted single-lap joint instrumented with piezoelectric sensors and actuators. The 
damage consisted of a crack growth from a notched bolt hole at various torque levels; 
their method was most successful with a completely loose bolt.  
  Continuing the work of Coelho et al. (2007), Chattopadhyay et al. (2007) used a 
kernel-based machine learning technique to detect and classify various forms of damage 
states in both metallic and composite structures. A bolted single lap joint representative 
of those commonly used in aerospace structures was chosen for the experimental testing. 
The type of damage addressed was as combination of bolt loosening and fatigue crack 
damage. The damage detection algorithm employed was able to distinguish between 
different torque states and associated changes in the crack length of the joint.  
Using vibration modal analysis, Okugawa and Tanaka (2007) conducted experimental 
testing on a one-bolt system; sub-space state identification to detect changes in the 
natural frequencies which identified a loosening bolt. Okugawa and Tanaka used a smart 
washer, a cantilevered plate type washer with a bonded PZT patch for self-sensing and 
35 
 
actuation. It was determined that this method had a very low sensitivity for detection 
preload loss.  
In recent years, there has been an increase in interest in methods that rely on the 
ambient structural vibrations. With this in mind, Milanese et al. (2008) explored several 
frequency and time domain signal processing techniques to detect progressive bolt 
loosening in a bolted composite structure. Milanese et al. only used the output vibration 
response data from the structure using fiber optic strain sensors to simulate a scenario 
where the ambient vibrations could be used.   
2.5 HILBERT-HUANG TRANSFORM 
As mentioned in the introduction, the limitations of the majority of the vibration based 
damage detection methods has prompted the emergence of a new category of time-history 
analyses; this includes damage detection methods using signal processing techniques 
such as the Hilbert-Huang Transform (HHT). Huang et al. (1998) developed the Hilbert-
Huang transform (HHT) in an effort to create an improved analysis method of nonlinear 
and non-stationary signals and, according to Loutridis (2004), to create a method for 
accurate time and frequency localization.  
The key component of the HHT is a method called the Empirical Mode 
Decomposition (EMD), which is used in conjunction with the Hilbert transform. The first 
step of the HHT process is the decomposition of a signal using EMD into components, 
called Intrinsic Mode Functions (IMF), which represent the intrinsic oscillating modes of 
the signal. The IMFs are associated with energy at different time scales and contain 
important characteristics of the data (Rezaei, 2010). The second step is to apply the 
Hilbert transform to the decomposed IMFs and construct an energy-frequency-time 
distribution, known as the Hilbert Spectrum, in order to ascertain the local energy and 
instantaneous frequency. A more detailed description of the decomposition of the IMFs 
from the original signal can be found in Chapter 4: section 4.3, and Chapter 5: section 
5.4.    
36 
 
2.5.1 Application of HHT in Structural Damage Detection 
Rezaei (2010) states that the capability of the HHT to reveal physical meaning of the 
data has only been empirically proven (Huang et al., 1998), and although the HHT has 
been successfully tested and validated empirically, it lacks a strong theoretical 
background and analytical formulation. In spite of this, since its introduction in 1998, the 
Hilbert-Huang transform has been used by researchers in many fields including ocean 
and seismic engineering, nuclear physics, biomedical diagnostics and image processing 
(Loutridis, 2004). Furthermore, it has also been applied to structural damage detection in 
the field of structural health monitoring.   
The Hilbert-Huang transform has been determined to be a powerful tool in the 
Vibration Based Damage Testing (VBDT) field (Rezaei, 2010), due to its simplicity, 
ability to deal with non-stationary and nonlinear signals, and its capability of extracting 
details buried in signals. Features of the HHT used in damage identification include the 
intrinsic mode functions, instantaneous frequency, instantaneous phase, and the Hilbert 
spectrum.  
An early example of the use of HHT for structural health monitoring was presented by 
Quek et al. (2003), who illustrated the use of HHT for anomaly detection in beams and 
plates. The suitability of HHT was explored through four different types of anomalies: an 
aluminum beam with a crack, a sandwiched aluminum beam with an internal 
delamination, a reinforced concrete slab with different degrees of damage and a plate 
with distorted input signal. Quek et al. determined that the HHT is able to represent a 
localized event well and is sensitive to even slight distortions in the signal caused by 
damage. 
The applicability of EMD for structural damage identification was experimentally 
investigated by Xu and Chen (2004). Their experimental set-up consisted of a three-story 
shear building model installed on a shaking table. Free vibration, random vibration and 
earthquake simulation tests were performed which all included damage in the form of a 
sudden stiffness change. Varying degrees of damage were induced using springs 
horizontally connected to the first floor; releasing the connecting springs decreased the 
structural stiffness. The EMD was applied to the dynamic response measurements of the 
37 
 
structure to identify the damage time instant and damage location for various test cases. 
Xu and Chen were successfully able to determine the damage time instant using the first 
IMF of the acceleration response, and concluded that the EMD was a promising tool for 
damage detection in real structures.  
Similarly, Lin et al. (2005) used the HHT technique to analyze the data of an IASC-
ASCE benchmark problem (Johnson et al., 2004) for the purposes of structural health 
monitoring. The data in question is from a benchmark structure consisting of an 
analytical structural model of a four-story two-bay steel frame structure. Lin et al. was 
able to successfully identify the natural frequencies, damping ratios, mode-shapes, and 
the stiffness and damping matrix of the structure with reasonable accuracy. Furthermore, 
Lin et al. were able to accurately detect the damage locations and severity of the 
benchmark structure using the HHT technique. Using the same data set, Liu et al. (2006) 
also used the HHT to successfully detect and locate the structural damages as well as 
detecting the instant of impact of damage during monitoring.     
Cheraghi et al. (2007) employed the HHT signal analysis method for the evaluation of 
the vibration characteristics of polyvinyl chloride (PVC) pipes using piezoelectric 
sensors. They studied the use of the HHT experimentally and computationally using 
Finite Element simulations to establish the natural frequency and damping ratio of the 
pipe.   
38 
 
2.6 EMD ENERGY DAMAGE INDEX 
Cheraghi et al. (2005) introduced a novel structural health monitoring method based 
on the HHT called the Empirical Mode Decomposition Energy Damage Index (EMD-
EDI). The EMD-EDI is based on the energy of certain IMFs of the vibration signals.   
Once the dynamic characteristics of a healthy structure during free vibration are 
collected through sensors, the acquired signals are then passed through a band-pass filter 
to ensure the existence of the first natural frequency within the data. This is followed by 
extraction of a specific IMF though the Empirical Mode Decomposition. Subsequently, 
the energy of the IMF, for each sensor, is established by: 
 
         
  
 
   (2-1) 
 
The above procedure is repeated for the same structure at its damaged state to 
determine the energy of each signal’s first IMF. The last step is the application of the 
EMD-EDI to each sensor: 
 
         
                 
        
      (2-2) 
 
Once the EMD-EDI is calculated for each sensor, the existence and severity of 
damage may be determined by associating the index values with the existence of damage 
close to the respective sensors. The index values having the greatest amplitudes coupled 
with a progression of increasing index values translates to good results.  
In summary, the methodology followed in the experiments documented herein is 
illustrated in Figure 2-7: Step 1: Monitor the vibration of a system in its healthy state 
using the response of sensors within the system. Step 2: Once again, monitor the 
vibration of the system in its damaged state using the responses of sensors within the 
system. Step 3: Analyze the signals and evaluate the EMD-EDI which is based on the 
39 
 
variation of IMF energy between the healthy and damage system states. Step 4: Establish 
whether damage exists based on the comparison of the EMD-EDIs evaluated in the last 
step.  
 
 
 
 
 
 
 
 
 
 
 
 
Figure 2-7  Damage detection methodology (adapted from Rezaei and Taheri, 2009). 
2.6.1 Application of the EMD Energy Damage Index to SHM 
Several researchers have performed experimental and numerical investigations in an 
effort to validate the effectiveness of the EMD-EDI method for structural health 
monitoring purposes. The majority of such works has concentrated on pipelines since 
they are such an integral part of today’s residential and industrial infrastructure. Today’s 
infrastructure relies on pipelines for conveying natural gas, oil, water, and 
communication and power cables.  
The stability and reliability of pipelines may be affected by ground movement, 
environmental conditions, and damage caused by excavation equipment at any time 
throughout their service life, making pipeline maintenance of prime importance. 
Moreover, damage assessment of pipelines is especially critical for economic and 
community recovery after natural disasters (Park and Inman, 2005). Pipelines can be 
severely damaged by shaking, liquefaction, and landslides during earthquakes (Kitaura 
Process the piezoceramic signals through 
the Empirical Mode Decomposition 
Evaluate the EMD energy of 
the desired IMF of the 
healthy structure 
Evaluate the EMD energy of 
the desired IMF of the 
damaged structure 
Monitor the vibration of 
the healthy structure 
Monitor the vibration of the 
structure in its damaged state. 
Calculate the EMD-EDI for each sensor.  
  Determine and report presence and 
location of damage.  
40 
 
and Miyajima, 1996). An immediate damage assessment may aid in preventing fires, 
explosions, and pollutions from broken gas and sewage lines (Park and Inman, 2005). For 
these reasons, the development of safe and reliable structural health monitoring 
techniques for pipelines is vital.  
Mentioned earlier, Cheraghi et al. (2005) investigated the integrity of the EMD-EDI 
by performing a set of experiments on simply supported plastic pipes with adhesively 
bonded joints. The vibration response of the healthy and damaged pipes was monitored 
using piezoelectric patches bonded at the joint location. Joint damage was in the form of 
a quarter and half circumferential debond. They compared the results of the EMD method 
against two other damage indices based on the Fast Fourier and Wavelet signal analysis 
methods and concluded that EMD method could effectively detect the damage.  
To further study and verify the EMD-EDI, Cheraghi and Taheri (2007) numerically 
evaluated the integrity of the method using Finite Element simulations. The subject of the 
investigation was a cantilevered aluminium pipe with various forms of defects, 
represented by the removal of elements in the pipe wall, simulating corrosion damage. 
The vibration response of the healthy and damaged pipes was monitored using 
piezoelectric patches spaced along the top of the pipe. The dynamic response of the pipe 
was analyzed using both the EMD-based procedure as well as methods based on Fourier 
transform, Wavelet transform, and Wavelet packet transform. Cheraghi and Taheri 
concluded that EMD and wavelet transform methods were the most sensitive and 
effective methods for detecting the damage location. Furthermore, the EMD-based 
method was the best method for identifying damage severity. 
Rezaei and Taheri (2009) employed the EMD-EDI for damage detection of 
cantilevered steel pipes, instrumented with piezoceramic sensors. Various degrees of 
corrosion damage were simulated by partial and full circumferential reductions in wall 
thickness. The influence of several variables on the results was considered. Different test 
cases altering the support stiffness and the type of impact hammer head used to excite the 
pipe were completed. Rezaei and Taheri determined that the EMD-EDI is effective in 
detecting and localizing single and multiple areas of damage as well as qualitatively 
assessing the damage severity.  
41 
 
Subsequently, Rezaei and Taheri (2010) extended the use of the EMD-EDI to pipeline 
girth welds. A steel pipe with a mid-span girth weld was experimentally and numerically 
tested in both pressurized and non-pressurized conditions. In contrast to the earlier 
investigations, Rezaei and Taheri used both piezoceramic sensors and a laser Doppler 
vibrometer to monitor the free vibration of the pipe. Multiple damage cases were 
completed involving cracks, simulated by notches of various locations and depths, at the 
girth weld. The damage evaluation using the EMD-EDI was compared with the 
evaluation of the lag in instantaneous phase, a quantity derived from the HHT and 
proposed as another damage feature. Rezaei and Taheri concluded that their results were 
encouraging and promise the effectiveness of the proposed approaches as inexpensive 
systems for structural health monitoring purposes. 
This thesis explores the integrity of the EMD-EDI for detecting damage in bolted 
joints, due to the limitations of the other vibration-based damage detection methods as 
outlined in Chapter 1. These studies serve to further expand the previous work 
accomplished on this method by focusing on a different type of structural damage than 
has been studied to date. 
 
 
42 
 
CHAPTER 3 
DAMAGE DETECTION OF A STEEL PIPELINE BOLTED JOINT 
 
The following chapter details the experimental tests performed on a simply supported 
steel pipe jointed at the mid-section using an 8-bolt flanged joint. An overview of the 
experimental set-up is shown in Figure 3-1. The objective of this study was to validate 
the effectiveness of using the EMD-EDI to detect various degrees of joint damage. 
Varying degrees of joint damage were simulated by reducing the torque load on the bolts 
and/or removing them completely. Three different combinations of piezoceramic sensors 
and/or impact locations were explored in total. As a result, this chapter is divided into 
three different sections, each detailing an experimental session.  
 
Figure 3-1  Overview of experimental set-up. 
  
43 
 
3.1 GENERAL EXPERIMENTAL SET-UP 
The test specimen used in this study was an ASTM compliant standard steel pipe 
(ASTM International, 2007) commonly used in oil and gas pipelines.   Table 3-1 lists the 
specifications of the schedule 40 Grade B pipe having a nominal pipe size of 6. A total 
pipe length of 4.27 m was chosen so as to fit on the press bed in the steel laboratory. 
Furthermore, the length was kept as long as possible so that the set-up would reflect, as 
much as possible, a real case scenario and so that the boundary conditions would not have 
an overbearing effect on the results.    
  Table 3-1  Session 1-3 ASTM pipe dimensions and material properties. 
Length  4.27 m Density 7850 kg/m
3
 
Outer Diameter 168.3 mm E  200  GPa 
Wall Thickness 6.35 mm v 0.3  
 
For sessions one through three, the mechanical joint utilized in the experiments was 
consistent. The ASTM and ASME/ANSI compliant forged steel flanges chosen were of 
the following type: Nominal Pipe Size (NPS) 6, class: 150, raised face, slip on, and 
material type: A105N.  Accordingly, the red rubber gaskets used were NPS: 6, class: 150, 
1/8 inch, and full face. SAE Grade 5 UNC hex head bolts were used (3/4 inch by 4 in) 
with corresponding sized nuts, and flat and lock washers. To create the mechanical joint 
in the pipe, the pipe was cut in two equal pieces, and the two flanges were welded onto 
the pipe. The flanges were continuously welded to the pipe on the inside of the joint with 
a small overlap over the pipe edge of approximately 23 mm, and tack welded on the 
outside of the joint.  
Illustrated in Figure 3-2, the pipe and joint system was supported at either end using 
simple supports, bolted to the press bed. For this specific joint configuration, the 
maximum bolt torque was determined to be 124.7 N-m by the gasket manufacturer. The 
procedure to tighten the bolts was the industry standard which follows a criss-cross 
pattern across the face of the flange following the bolt numbering in Figure 3-3. Multiple 
passes should be completed at increments of 30%, 60% and then 100% of the maximum 
torque for the joint. For this joint, the magnitudes of these increments are listed in Table 
44 
 
3-2. The last step to tightening the joint is to go around the joint in a clockwise pattern 
and verify that every bolt is at the maximum torque. The joint was tightened using a 
torque wrench.  
 
Figure 3-2  Boundary conditions. 
 
 
 
 
 
 
 
 
 
Figure 3-3  Session 1-3 bolt numbering. 
 
Table 3-2  Session 1-3 bolt torque increments.  
30% of Maximum 37.4 N-m 
60% of Maximum 74.8 N-m 
100% of Maximum 124.7 N-m 
 
1 
5 
3 
7 6 
4 
8 
2 
45 
 
It may also be of some importance, considering the symmetry of the joint, that the 
bolts were always fastened so that the bolt heads lay on the same side as impact location 
C and D (Figure 3-4), which remained consistent for all three sessions. 
To monitor the dynamic response of the system, piezoceramic sensors were bonded to 
the pipe and joint in several different arrangements. The piezoceramic sensors used for 
the entirety of the experimental testing were of type PZT-5H, extracted from 72.4 × 72.4 
mm sheets available from Piezo Systems Inc. (Cambridge, MA). The dimensions of the 
sensors are listed in Table 3-3.  
   Table 3-3  Piezoceramic sensor dimensions.   
Length 24.0 mm 
Width 10.0 mm 
Thickness 1.0 mm 
 
Due to the small width of the sensors, fitting them to the curvature of the pipe’s 
surface did not pose a problem. The material properties of the sensors (obtained from the 
manufacturer) are described using the following matrices. The density of PZT-5H is 
          . 
 
   
  
 
         
 
 
 
 
 
 
                 
                 
                 
         
         
          
 
 
 
 
 
 
 
  
 
 
          
        
        
              
      
 
                
      
      
      
  
where    is the compliance matrix,   is the piezoelectric coupling matrix, and  
  is 
the permittivity of the piezoceramic. The piezoceramic sheets were polarized through the 
46 
 
thickness allowing the sensors to be created by soldering two electrodes to the top and 
bottom surfaces of each piece cut from the piezoceramic sheet. However, soldering 
caused each sensor to have an uneven surface which prevented complete contact between 
the pipe surface and the sensor. Consequently, a steel plate of size 25.0 × 10.0 × 0.5 mm 
was used as an adapter. Each adapter plate had a small semi-circular hole cut at the edge 
to fit around the solder. Placing the adapter plate between the sensor and the pipe surface 
allowed for increased contact.  
Before a sensor was attached, the surface of the pipe was scoured using silicon-
carbide paper. The pipe surface, sensor and adapter plate were then wiped clean of dust 
and oil with isopropyl alcohol. The preparation, resulting in clean and rough surfaces, 
permitted a stronger bond to be created. The piezoceramic sensors and the adapter plates 
were subsequently adhered to the pipe using a two part epoxy, Araldite 2011 (Huntsman 
Advanced Materials Americas Inc., Los Angeles, CA). The sensors were secured to the 
pipe, taped down using masking tape, and left to cure for twelve hours to create a strong 
bond.    
  
47 
 
3.2 SESSION 1 
The following section includes a description of the experimental set-up for the initial 
combination of sensors/impact locations used, the testing procedure, the data analysis, 
and a presentation of the results followed by a discussion.  
3.2.1 Experimental Set-Up 
The set-up for the completion of damage detection was accomplished as follows and 
may be seen in Figure 3-4. Four impact locations were chosen for testing, Location A, B, 
C and D situated on either side of the joint. In order to maintain consistency between 
impacts, a steel ball measuring 12.7 mm in diameter was placed at each of the impact 
locations to provide a consistent impact target. The impact balls were attached to the pipe 
using the same epoxy as used for the sensors. An example impact location is shown in 
Figure 3-5. 
 
 
 
 
 
 
 
 
Figure 3-4  Session 1 experimental set-up.  
 
Figure 3-5  Example impact location. 
To monitor the system vibrations, six piezoceramic sensors were placed on the pipe at 
a distance of 250 mm from the outside of the closest flange measured to the center of the 
250 mm 
1060 mm 
760 mm 760 mm 
1060 mm 
Location D Location B Location C Location A 
Piezoceramic sensors 
48 
 
sensors. The placement of these sensors along the pipe are shown in Figure 3-4 and the 
placement and numbering of the sensors is illustrated in a cross-section of the pipe in 
Figure 3-6, and in a photograph of the experimental set-up in Figure 3-7.  
 
 
 
 
 
 
 
 
Figure 3-6  Session 1 PZT sensor placement and numbering. 
 
Figure 3-7  Session 1 sensor placement. 
The sensor locations were chosen so that a comparison could be made between the 
results from the sensors at the top and bottom of the pipe as well as with the sensors 
situated at 45 degrees. Concerning the distance to the flange, it was thought that the 
sensors should be placed at a distance so that work on the joint would not have the 
chance of inadvertently damaging the sensors.  
 
 
 
  
 
 
 
PZT1 
  PZT2 
 PZT3 
PZT4 
PZT6 
  PZT5 
PZT1 
PZT2 
PZT6 
49 
 
Since the sensors were only placed on one side of the pipe (thought to be sufficient 
since the set-up is symmetrical), the impact locations were chosen so that they would not 
be too close to the sensors (thereby interfering with the signals). Furthermore, the impact 
locations were placed on both sides of the joint in order to study the difference between 
the vibration-induced wave propagation through the joint, to reach the sensor. Moreover, 
having two impact locations per side permitted the exploration of whether impacts closer 
to the joint could have improved the results.   
3.2.2 Testing 
 To create free vibration in the system, the pipe was excited using a piezoelectric 
impulse hammer (model 5800B5, Dytran Instrument Inc., Chatsworth, CA) outfitted with 
an aluminum tip which could excite frequencies up to 5 kHz. The hammer outputs an 
analog voltage which is representative of the input impulse. A typical hammer impulse 
signal is depicted in Figure 3-8 (a). 
  
a)                                  b)  
Figure 3-8  a) Typical impulse hammer signal b) Typical sensor response. 
In order to collect the response of the PZT sensors and the impulse hammer, they were 
connected by BMC cables to a multifunction PCI 6220 data acquisition card 
manufactured by National Instruments Inc., Austin, Texas). The data acquisition (DAQ) 
system, shown in Figure 3-9, provided a limitation to the experiments in that the board 
can only hold up to 8 inputs, which meant that data from only seven sensors could be 
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
-0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5
A
m
p
li
tu
d
e 
(V
o
lt
)
Time (10-3s)
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
0.0 0.5 1.0
A
m
p
lit
u
d
e
 (
V
o
lt
)
Time (s)
50 
 
concurrently read and recorded. Furthermore, the same gasket was used for all testing. A 
typical sensor response is depicted in Figure 3-8 (b).   
 
Figure 3-9  The data acquisition hardware. 
 For session 1, two healthy cases were completed, as well as four damage cases. A list 
of all the test cases considered for session 1 is shown in Table 3-4. Each healthy case and 
each damage case consisted of five impacts at each of the four impact locations in an 
effort to achieve consistent results. Two healthy cases were completed for comparison 
purposes so that the consistency between healthy states could be investigated. For each 
healthy state that was completed, the joint was fully loosened and re-torqued using the 
procedure outlined earlier. 
Table 3-4  Session 1 healthy cases and damage cases.              
  State Bolts Loosened Damage Case 
Healthy Case 1-1 
 
HC 1-1 
Healthy Case 1-2 
 
HC 1-2 
Damage Case 1-1 Bolt #1: Damaged 80% DC 1-1 
Damage Case 1-2 Bolt #1: Damaged 50% DC 1-2 
Damage Case 1-3 Bolt #1 and Bolt #5: Damaged 50% DC 1-3 
Damage Case 1-4 Bolt #1, #5 and #8: Damaged 50% DC 1-4 
 
 For the damage cases, the percent of damage listed in the table is the percentage of the 
maximum torque to which the bolts were loosed. For example, the 80% for damage case 
51 
 
1-1 signifies that Bolt #1was loosened to 100 N-m (80% of 124.7 N-m). 
 Figure 3-10 is a schematic depicting the bolts included in the damage cases. The 
method used to loosen the bolt was as follows: with the aid of a torque wrench set to the 
value that the bolt should be loosened to, the bolt was loosened using an adjustable 
wrench a slight amount, then tested using the torque wrench for its current torque, and 
subsequently adjusted until the desired torque was reached.  
For damage case 1-4, it was discovered that when the third bolt was loosened, Bolt #1 
retightened and therefore each bolt had to be rechecked for the correct bolt torque.  
 
 
 
 
 
 
 
Figure 3-10  Session 1 damage overview. 
3.2.3 Data Analysis  
After monitoring and recording the free vibration of the pipe in its healthy state using 
the six piezoceramic sensors, damage was introduced in the joint by loosening the bolt(s) 
and the free vibration of the pipe was recorded in its damaged state. For all damages, the 
second healthy state was the comparison healthy state used in the calculation of the 
EMD-EDI. As mentioned in the previous section, four damage sessions were studied, 
which included the loosening of one bolt, two bolts, and three bolts to a lower torque.  
All of the data for the experimental testing was recorded using a sampling frequency 
of 10 kHz. Once the data was collected using LabView software, the data for each impact 
was stored in an ASCII text file which included the responses of all the sensors and the 
impulse hammer. Subsequently, the recorded signals from the healthy and damaged pipe 
joints were processed using a Matlab code, developed in house by another member of the 
B5 
B1 
 
B8 
7 6 
2 
3 4 
52 
 
research group, which performed the Empirical Mode Decomposition. This code was 
programmed to read the text file produced by the DAQ system and then process the 
response of all six sensors simultaneously. The code removes unwanted data before 
impact, filters signal noise, compensates for any offset, normalizes the response of any 
sensor with respect to the hammer input force, applies a low-pass filter in order to have 
the first natural frequency in the filtered data, decomposes the signals based on the EMD 
method, derives the IMFs, and finally calculates the energy of the desired IMF for each 
sensor (Rezaei and Taheri, 2009). A typical voltage signal of a sensor and its 
decomposition based on the EMD sifting process is illustrated in Figure 3-11. 
 
Figure 3-11  Typical signal decomposition from EMD (Rezaei and Taheri, 2010). 
 For each excitation, the vibration lasted for seconds before decaying to zero (see a 
typical vibration signal in Figure 3-8 (b)). Because the EMD-EDI is based on the 
comparison of special features in the vibration signals before and after damage creation, 
there is no limitation for the choice of signal duration, as long as the same duration is 
used to evaluate both the healthy and damaged state signals (Rezaei and Taheri, 2009). 
With this in mind, an interval of 0.5 seconds was chosen for the signal processing.  
53 
 
Furthermore, each of the acquired signals from the sensors was normalized based on 
the input (hammer) signal. The difference between the signals after normalization was 
considered to be a good measure of the repeatability of the impact procedure (Rezaei and 
Taheri, 2009).  
To normalize the signals, the following procedure was followed (Rezaei and Taheri, 
2009): The fast Fourier transform (FFT) of each sensor response was divided by the FFT 
of the hammer signal. Afterwards, the result was transferred to the time domain by the 
inverse fast Fourier Transform. Three impacts for each test case were processed and the 
IMF energies attained, and their average used for the EMD energy. Once the EMD 
energy of both the healthy and damages cases was calculated, the EMD-EDI was 
calculated for each damage case.  
3.2.4 Results 
In an effort to more fully understand the system dynamics, the first three natural 
frequencies of the pipe joint system were calculated analytically using the following 
equation for a simply supported beam (Chopra, 2001):  
   
  
  
 
  
   
                                           (3-1) 
 where           is the angular frequency of the n
th
 mode, E is the modulus of 
elasticity, I is the second area moment of inertia,   is the beam length, and m is the mass 
of the beam. The mass of the joint itself was calculated separately and distributed along 
the length of the beam for the sake of calculation. The natural frequencies are listed in 
Table 3-5. 
 Table 3-5  Natural frequencies calculated analytically.              
Natural frequency ( f = ω/2π) 
1 23 Hz 
2 92 Hz 
3 207 Hz 
These were compared with the natural frequencies calculated using the Fast Fourier 
transform of the piezoceramic sensor’s typical output signal shown in Figure 3-12. The 
54 
 
first two natural frequencies compare well while the third frequency does not, possibly 
due to the assumptions made for the calculation (i.e. distribution of the joint mass and 
stiffness).  
             
Figure 3-12  Natural frequencies of pipe-joint system from FFT of PZT7 data  
(signal from session 2, HC 1-4, Location C, aluminium tip). 
As part of the EMD-EDI procedure, the data should be filtered before it is 
decomposed (Cheraghi et al., 2005; Cheraghi and Taheri, 2007). Cheraghi et al. (2005) 
initially suggested that the band-pass filter should be of a magnitude to only retain the 
first frequency. However, it was later discovered through more research (Rezaei and 
Taheri, 2010) that the higher frequency components may need to be retained in the case 
of small localized damages. By analyzing the data using different frequency bands, it was 
found that a low-pass filter with a pass-band frequency of 2000 Hz provided the best 
resolution for identification of damage. This pass-band was chosen based on exhibition of 
damage progression as well as consideration of energy amplitude.  
Figure 3-13 through Figure 3-16 illustrate the EMD-EDI calculated respectively for 
each impact location. Each graph illustrates the indices of all damage cases for each 
sensor. For example, the results for PZT2 were as follows: DC 1-1: index of 58; DC 1-2: 
index of 48; DC 1-3: index of 80; DC 1-4: index of 53. The results shown in these figures 
are based on the frequency band of 0-2000 Hz, and the energy of the first IMF.  
 
A
m
p
li
tu
d
e 
55 
 
 
 
Figure 3-13  EMD-EDI, session 1, location A, IMF1.  
 
 
 
 
Figure 3-14  EMD-EDI, session 1, location B, IMF1. 
 
 
 
 
Figure 3-15  EMD-EDI, session 1, location C, IMF1. 
 
0
20
40
60
80
100
120
PZT1 PZT2 PZT3 PZT4 PZT5 PZT6
E
M
D
-E
D
I
Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
0
20
40
60
80
100
120
PZT1 PZT2 PZT3 PZT4 PZT5 PZT6
E
M
D
-E
D
I
Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
0
20
40
60
80
100
120
PZT1 PZT2 PZT3 PZT4 PZT5 PZT6
E
M
D
-E
D
I
Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
56 
 
 
 
Figure 3-16  EMD-EDI, session 1, location D, IMF1. 
From comparing Figure 3-13 through Figure 3-16, it is clear that impact locations A 
and B provided better damage progression overall. PZT1 and PZT5 respectively show 
damage progression for the two locations. Impact locations C and D did not produce any 
damage progression and the damage index amplitudes were in the same range as that of 
impact location B.  It can therefore be postulated that the EMD based method can better 
detect damage progressions when the vibration signal travels through the joint to reach 
the sensors (see Figure 3-4).   
One can also observe that the damage index amplitudes obtained through the signals 
excited by impacting the pipe at location A are much greater than those obtained by 
excitation at other locations.  There is a great deal of similarity among the damage index 
amplitudes and the relationship among each damage index for impact locations C and D, 
which one would expect, considering the fact that the two impact locations are 0.3 m 
apart. Nevertheless, impact locations A and B do not show this similarity, which might 
suggest errors in some signal with respect to impact location A and B.     
3.2.5 Consistency 
As mentioned in section 3.2.3, once the five impacts for each test case were processed 
and the IMF energies obtained, only the three closest energies of the signals of each 
sensor were retained and their average used. When reducing the number of energies from 
five to three, it was desired to maintain the variation between the resulting EMD energy 
averages below 20% for each sensor. If the variation in the energies was above 20%, 
there was too much scatter in the impact data. Therefore, five additional tests were 
0
20
40
60
80
100
120
PZT1 PZT2 PZT3 PZT4 PZT5 PZT6
E
M
D
-E
D
I
Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
57 
 
conducted for the particular impact location. The variation of each impact induced EMD 
damage energy of a given sensor was calculated from the average using the following 
equation:  
                      
               
    
   
               
    
  (3-2) 
Tables A-1 and Table A-2 (Appendix A) report the calculated variation for each 
healthy and damage case state. Table A-1 lists the values for the pass-band frequency of 
0-2000 Hz, IMF1 and Table A-2 lists the values for the pass-band frequency of 0-2000 
Hz, IMF2 for the sake of examining whether energies calculated based on the second 
IMF could provide more conclusive results. 
3.2.6 Result Repeatability 
In order to address the question of result repeatability, the EMD energy of two 
healthy states was calculated as mentioned earlier. It was assumed that the difference 
between the average EMD energy for each sensor was a sufficient gauge with which to 
measure the repeatability of the impact procedure. The following equation was used to 
calculate the variability between the EMD energy of the healthy states.   
                      
                     
        
   
                     
        
  (3-3) 
 
where:  i and j represent the two states being compared and  
                   = the average energy based on the vibration signals of the first state 
       = the average energy based on the vibration signals of the second state 
         = the average of the average energies of the two states 
The range of comparison values are listed in Table A-3 in Appendix A. The EMD 
energies tabulated for impact location A do not compare well, having differences of at 
least 30%, which may explain the high magnitudes of the index for that location.  
58 
 
3.2.7 Discussion and Conclusions 
Due to the poor consistency of the results concerning the progression of damage, as 
well as the high variability in the values of the EMD energies obtained at the healthy 
state, the decision was made to remove the gasket and repeat the tests to see if the results 
would improve with less damping in the joint.  Furthermore, to improve the damage 
progression results, it was decided that new sensors would be placed in a closer proximity 
to the joint. Without consistent results demonstrating damage progression, the decision 
was made to maintain all four impact locations until better results could suggest the 
elimination of the less useful locations. Moreover, the same reasoning applies for the 
continued use of both IMFs to analyze the data until better results allow a good 
comparison.  
  
59 
 
3.3 SESSION 2 
The following section includes a description of the experimental set-up for the second 
combination of sensors/impact locations used, the testing procedure, the data analysis, 
and a presentation of the results followed by a discussion.  
3.3.1 Experimental Set-Up 
The experimental set-up for session 2 was the same as for session 1, except for the 
lack of gasket and a different sensor arrangement. The gasket was removed in order to 
see if the lack of damping could improve the results. The experimental set-up for session 
2 is illustrated in Figure 3-17. The sensor locations were chosen with the thought that 
sensors closer to the joint may provide better results.  
 
 
 
 
 
 
 
 
Figure 3-17  Session 2 experimental set-up. 
 
Six new piezoceramic sensors were attached to the pipe and flange. To monitor the 
pipe vibrations, two piezoceramic sensors, PZT7 and PZT8, were placed on the pipe at a 
distance of 25.0 mm from the outside of the closest flange to the center of the sensors. To 
monitor the flange vibrations, four piezoceramic sensors were placed on the flange on the 
same side as the previous sensors. PZT9 and PZT10 were place on either side of Bolt 1 
and PZT11 and PZT12 were placed on either side of Bolt 2 (refer to Figure 3-3).  
The placement of these sensors along the pipe is shown in Figure 3-17 and the 
placement and numbering of the sensors is illustrated in a cross-section of the pipe in 
Figure 3-18, and in a photograph in Figure 3-19. The sensor locations were chosen so that 
a comparison could be made between those results from on the pipe and on the flange, as 
well between the results obtained from the sensors placed on the bottom and top of the 
pipe/flange.   
Location D Location B Location C Location A 
25 mm 
760 mm 760 mm 
1060 mm 1060 mm 
Piezoceramic sensors 
60 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure 3-18  Session 2 PZT sensor placement and numbering. 
 
Figure 3-19  Session 2 sensor placement. 
3.3.2 Testing 
In contrast to session 1, two hammer tips, aluminium and plastic, were employed in 
order to determine which might yield better results.  Unlike the aluminium tip, the hard 
plastic tip is only capable of exciting frequencies below 2 kHz. For session 2, ten impacts 
were completed for each impact location for all cases to ensure that impact errors were 
PZT7 
PZT8 
 PZT10 
PZT11 
PZT12 
 
 
 
  
 
 
 
PZT10 
PZT7 
 PZT8 
 
PZT12 PZT11 
61 
 
not contributing to the lack of damage progression.  A list of all test cases considered for 
session 2 is shown below in Table 3-6.  
Table 3-6  Session 2 cases.              
  State Bolts Loosened Damage Case 
Healthy Case 2-1 
 
HC 2-1 
Healthy Case 2-2 
 
HC 2-2 
Healthy Case 2-3 
 
HC 2-3 
Healthy Case 2-4 
 
HC 2-4 
Damage Case 2-1 Bolt #1: Damaged 80% DC 2-1 
Damage Case 2-2 Bolt #1: Damaged 66% DC 2-2 
Damage Case 2-3 Bolt #1 and Bolt #5: Damaged 66% DC 2-3 
Damage Case 2-4 Bolt #1 and Bolt #5: Damaged 50% DC 2-4 
Damage Case 2-5 Bolts #1, #5 and #8: Damaged 50% DC 2-5 
Damage Case 2-6 
Bolt #1: Completely Loose 
Bolts #5 and #8: Damaged 50% 
DC 2-6 
Damage Case 2-7 
Bolts #1 and #5: Completely Loose 
Bolt #8: Damaged 50% 
DC 2-7 
Damage Case 2-8 Bolts #1, #5 and #8: Completely Loose DC 2-8 
Damage Case 2-9 Bolts #1, #5, #8 and #4: Completely Loose DC 2-9 
 
Figure 3-20 is a schematic depicting the bolts included in the damage scenario listed 
above. Four healthy cases were completed for comparison purposes so that the 
consistency between healthy states could be investigated.  For each healthy state that was 
completed, the joint was fully loosened and re-torqued using the procedure outlined in the 
experimental set-up discussed earlier. The four healthy states listed in Table 3-6 were 
performed again due to the poor results obtained for session 1. For the damage cases for 
which a bolt was completely loose, the bolt was completely removed from the joint. 
 
 
 
 
62 
 
 
 
 
 
 
 
 
 
 
Figure 3-20  Session 2 damage overview. 
3.3.3 Data Analysis  
 The data analysis for session 2 was completed in the same manner as for session 1, 
except that for session 2, five impact signals were chosen for the calculation of EMD 
average energy.  
3.3.4 Results 
For session 2, it was decided that in order to fully explore the range of bandwidth 
frequencies that might be suitable for this joint configuration, the full range of 
frequencies from 0-4000 Hz would be processed. Thirty-six pass-bands were explored 
using increments of 500 Hz. This range was chosen because the plastic tip excites 
frequencies up to 2 kHz and the aluminum hammer tip is capable of exciting frequencies 
of 5 kHz.  
By analyzing the data using the different frequency bands, it was determined that a 
low-pass filter with a pass-band frequency of 2500 Hz provided the best resolution for 
identification of damage. This pass-band was chosen based on exhibition of damage 
progression as well as consideration of the index amplitude. Figure 3-21 through Figure 
3-24 illustrate the aluminum tip results calculated for each sensor for each respective 
impact location. The results shown in these figures are based on the first IMF. 
B5 
B1 
B4 
B8 
7 6 
2 
3 
63 
 
 
 
Figure 3-21  EMD-EDI, session 2, location A, 0-2500 Hz,  
IMF1, aluminium tip. 
 
   
 
 
Figure 3-22  EMD-EDI, session 2, location B, 0-2500 Hz,  
IMF1, aluminium tip. 
 
 
 
Figure 3-23  EMD-EDI, session 2, location C, 0-2500 Hz,  
IMF1, aluminium tip. 
0
20
40
60
80
100
PZT7 PZT8 PZT10 PZT11 PZT12
E
M
D
-E
D
I
Sensor
DC 2-1
DC 2-2
DC 2-3
DC 2-4
DC 2-5
DC 2-6
DC 2-7
DC 2-8
DC 2-9
0
20
40
60
80
100
PZT7 PZT8 PZT10 PZT11 PZT12
E
M
D
-E
D
I
Sensor
DC 2-1
DC 2-2
DC 2-3
DC 2-4
DC 2-5
DC 2-6
DC 2-7
DC 2-8
DC 2-9
0
20
40
60
80
100
PZT7 PZT8 PZT10 PZT11 PZT12
E
M
D
-E
D
I
Sensor
DC 2-1
DC 2-2
DC 2-3
DC 2-4
DC 2-5
DC 2-6
DC 2-7
DC 2-8
DC 2-9
64 
 
 
 
Figure 3-24  EMD-EDI, session 2, location D, 0-2500 Hz,  
IMF1, aluminium tip. 
 
Only one impact location graph is shown for the plastic tip as the results for the other 
impact locations are very similar in variability.   
 
 
Figure 3-25  EMD-EDI, session 2, location C, 0-2500 Hz,  
IMF1, plastic tip. 
 
3.3.5 Consistency 
The same process was used for session 2 as for session 1 to measure result 
consistency. In order to document the range of variability, Tables A-4 and A-5 (Appendix 
A) list the variations for each healthy and damage case. Table A-4 lists the values for the 
aluminum tip data and Table A-5 lists the values for the plastic tip data.  
However, it must be noted that the signals were initially analyzed and accepted 
/rejected using the pass-band frequency of 0-2000 Hz since this was the best pass-band 
frequency for session 1.  Therefore, it was the variability of the EMD energies using the 
pass-band of 0-2000 Hz that were verified as being below 20%. This is why a few high 
errors exist in Table A-5, which lists the values for the plastic tip data for the pass-band 
0
20
40
60
80
100
PZT7 PZT8 PZT10 PZT11 PZT12
E
M
D
-E
D
I
Sensor
DC 2-1
DC 2-2
DC 2-3
DC 2-4
DC 2-5
DC 2-6
DC 2-7
DC 2-8
DC 2-9
0
20
40
60
80
100
PZT7 PZT8 PZT10 PZT11 PZT12
E
M
D
-E
D
I
Sensor
DC 2-1
DC 2-2
DC 2-3
DC 2-4
DC 2-5
DC 2-6
DC 2-7
DC 2-8
DC 2-9
65 
 
frequency of 0-2500 Hz. This stands to reason because the plastic tip only excites reliable 
frequencies up to 2000 Hz. 
3.3.6 Result Repeatability 
In order to re-address the question of result repeatability, four healthy cases were 
completed as mentioned earlier. It was assumed that the difference between the average 
EMD energies of the healthy cases was a sufficient gauge with which to measure the 
repeatability of the impact procedure. Eqn. 3-3 was used to calculate the variability 
between healthy states.   
The range of comparison values are listed in Table A-6 (aluminium tip) and Table A-
7 (plastic tip) in Appendix A.  The results are much improved from those of session 1, 
with values over 20% occurring only for PZT10 and PZT12 for the plastic tip. Once 
again, this high values may be due to the fact that the plastic tip only excites reliable 
frequencies up to 2000 Hz. 
3.3.7 Discussion and Conclusions 
Considering the above data, the frequency range of 0-2500 Hz yielded the best results 
in terms of damage index amplitude and damage progression with the aluminium tip 
yielding better results than the plastic tip.  
From the comparison of impact locations (Figure 3-21 through Figure 3-24), impact 
location C produced the best results when all sensors were considered, yet it was out 
performed by impact location A for PZT7, PZT8 and PZT10 considering both index 
amplitude and damage progression.  Impact location D produced the least clear results in 
terms of damage progression.  It is interesting to note that the results for impact location 
B, although closer to the joint, were similar yet smaller in amplitude than those of impact 
location A.  This trend is reversed considering the opposite side of the joint (C and D), 
for which the impact location closest to the joint, location C, provided the most useful 
results.  
PZT10 demonstrated the best amplitude among all the sensors and showed damage 
progression for at least the last five damage cases for all impact locations (for the 
aluminum tip only). It is also clear from Figures 3-21 to 3-25 that PZT11 and PZT12 only 
66 
 
demonstrated good damage progression and amplitude for impact location C, the impact 
location closest to the sensors, (see Figure 3-17).  Regarding whether the best location for 
the sensors is on the pipe itself or on the flange, the results are as of yet inconclusive.  
Based on these results, it was decided that the next session would once again include 
both hammer tips to give further evidence substantiating the conclusion that the 
aluminum tip yields better results than the plastic tip. Furthermore, because of the 
variance in results between the four impact locations, the next session would include 
sensors on either side of the flange in an effort to discover which side of the pipe is better 
for impact with respect to the side that the sensors are on. The next session would also 
include two new impact locations closer to the joint on either side, and not consider the 
impact locations A and D.  
It is also worthy of note that during this session, the introduced damage occurred only 
on top of the pipe, with bolts that are side by side. Therefore, in an effort to have a wider 
range of results, the next session will include loosened bolts that are not side by side, and 
that are located on either side of the pipe. Moreover, because the majority of the results 
from session 2 did not show damage progression for the first 5 damage progressions, only 
fully loosened bolts will be considered for the next session.   
Finally, the decision to perform testing without a gasket yielded improved results 
overall. A possible inference would be that the damping caused by the presence of the 
gasket, and the type of gasket, may play a large role in the usefulness of this method.   
  
67 
 
3.4  SESSION 3  
The following section includes a description of the experimental set-up for the third 
combination of sensors/impact locations used, the testing procedure, the data analysis, 
and a presentation of the results, followed by a discussion.  
3.4.1 Experimental Set-Up 
The experimental set-up for session 3 was the same as for session 2, except for a 
modified sensor arrangement. Similarly to session 2, no gasket was used for the tests. The 
experimental set-up for session 3 is illustrated in Figure 3-26. 
 
 
 
 
 
Figure 3-26  Session 3 experimental set-up.  
Four new piezoceramic sensors, PZT13 through PZT16, were created and attached to 
the opposite side of the flange, mirroring the positions of PZT9 through PZT12 from 
session 2 that were also retained for this session. The placement of these sensors along 
the pipe is shown in Figure 3-26 and the placement and numbering of the sensors are also 
illustrated in a cross-section of the pipe in Figure 3-27, and in a photograph in Figure 3-
28.   
The sensor locations were chosen so that a comparison could be made between the 
results from either side of the flange to determine if one side would yield better results. 
The tightening procedure for session 3 was the same as for session 2.  
 
 
 
 
760 mm 
460 mm 460 mm 
760 mm 
 Location E  Location B Location C Location F 
Piezoceramic sensors 
68 
 
 
 
 
  
 
 
 
PZT10 
 
  
 
PZT12 PZT11 
 
 
 
 
 
 
 
 
a)                                                    b) 
Figure 3-27  Session 3 PZT sensor placement and numbering a) Face of left flange b) Face of 
right flange. 
 
 
Figure 3-28  Session 3 sensor placement. 
PZT14 
PZT13 
PZT16 
PZT15 
 
 
 
  
 
 
 
PZT13 
 
  
     PZT14 
PZT15 PZT16 
69 
 
3.4.2 Testing 
For session 3, seven impacts were completed for each impact location for the healthy 
state, and five impacts were completed for each impact location for each of the three 
damage states. The number of impacts was reduced from the number used in session 2 
(i.e. 10) due to the consistency achieved in session 2. Similarly to session 2, both 
aluminium and plastics impact hammer tips were employed in order to help determine 
which might yield better results. A list of all test cases considered for session 3 is shown 
in Table 3-7. Figure 3-29 is a schematic depicting the bolts included in the damage cases 
listed in Table 3-7. 
Table 3-7  Session 3 cases.              
  State Bolts Loosened Damage Case 
Healthy Case 3-1 
 
 
Damage Case 3-1 Bolt #3: Removed from Joint DC 3-1 
Damage Case 3-2 Bolt #3 and Bolt #4: Removed from Joint DC 3-2 
Damage Case 3-3 Bolts #1, #3, #4: Removed from Joint DC 3-3 
 
For this session, only one healthy state was completed since it was felt that the issue 
of repeatability was dealt with sufficiently in session 2. Additionally, only complete 
removal of bolts was performed due to the difficulty in detecting a smaller amount of 
clamping force loss in the joint.  
 
 
 
 
 
 
 
 
 
Figure 3-29  Session 3 damage overview. 
B1 1 
5 
 
7 6 
4 
8 
2 
B3 B4 
70 
 
3.4.3 Data Analysis  
 The data analysis for session 3 was completed in the same manner as for session 1 
and 2 except that for session 3, five healthy case impact signals were chosen for the 
calculation of EMD average energy, and three of the damaged case signals. 
3.4.4 Results 
For session 3, the bandwidth frequencies in Table 3-8 were explored in an effort to 
determine which one would be the most suitable for this joint configuration. For each 
frequency range, both aluminum tip and plastic tip data was analyzed for both IMF 1 and 
IMF 2 as stated earlier. Until now, the experimental research performed for this method 
has concentrated on the first IMF (Cheraghi and Taheri, 2007; Rezaei and Taheri, 2009). 
It was therefore decided to explore the results from the second IMF.  
Table 3-8  Session 3 frequencies ranges. 
0-1000 Hz 
0-2000 Hz 
0-2500 Hz 
1000-2000 Hz 
1000-2500 Hz 
1500-2000 Hz 
0-500 Hz 
500-2500 Hz 
By analyzing the data according to the frequency ranges in Table 3-8, it was 
determined that a low-pass filter with a pass-band frequency of 2000 Hz provided the 
best resolution for identification of damage.  
Figure 3-30 to Figure 3-33 illustrate the aluminum tip results for the first IMF. 
Likewise, Figure 3-34 through Figure 3-37 illustrates the plastic tip results for the first 
IMF. In each figure, the sensor results are illustrated and organized according to the 
sensors placed on the left side of the joint and those on the right (in reference to Figure 3-
26).  
 
 
71 
 
 
Figure 3-30  EMD-EDI, session 3, location B, 0-2000 Hz,  
IMF1, aluminium tip. 
 
 
 
Figure 3-31  EMD-EDI, session 3, location C, 0-2000 Hz,  
IMF1, aluminium tip. 
 
 
 
Figure 3-32  EMD-EDI, session 3, location E, 0-2000 Hz,  
IMF1, aluminium tip. 
 
 
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
           Left Side       Right  Side 
        Left Side        Right Side 
         Left Side       Right Side 
72 
 
 
Figure 3-33  EMD-EDI, session 3, location F, 0-2000 Hz,  
IMF1, aluminium tip. 
 
 
 
 
Figure 3-34  EMD-EDI, session 3, location B, 0-2000 Hz,  
IMF1, plastic tip. 
 
 
 
Figure 3-35  EMD-EDI, session 3, location C, 0-2000 Hz,  
IMF1, plastic tip. 
 
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensors
DC 3-1
DC 3-2
DC 3-3
        Left Side      Right Side 
           Left Side       Right Side 
          Left Side      Right Side 
73 
 
 
 
Figure 3-36  EMD-EDI, session 3, location E, 0-2000 Hz,  
IMF1, plastic tip. 
 
 
 
Figure 3-37  EMD-EDI, session 3, location F, 0-2000 Hz,  
IMF1, plastic tip. 
 
 
The results for the second IMF are shown as Figures A-1 through A-8 in Appendix A. 
A comparison of the results obtained based on the two IMFs demonstrate considerable 
similarity, although each IMF may show better progression for specific sensors for 
specific locations differently than the other IMF.  It was therefore decided to add the 
IMFs to see if a better resolution of damage could be achieved. The results for the 
combined IMFs are shown as Figures A-9 through A-16 in Appendix A. The addition of 
the IMFs yields very similar results overall but slightly improved compared to the either 
the first or second IMF. 
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
        Left Side        Right Side 
         Left Side      Right Side 
74 
 
3.4.5 Consistency 
The same process was used for session 3 as for sessions 1 and 2 to measure result 
consistency. Tables A-8 through A-11 in Appendix A list the variations in the EMD 
energy for each healthy and damage case. The values were all below 20%.  
3.4.6 Discussion and Conclusions 
The filtering frequency range of 0-2000 Hz yielded the best results in terms of 
damage index amplitude and damage progression. Moreover, the combined IMFs 
demonstrated the best resolution for damage detection.   
From the comparison of data obtained for the impact locations for IMF1, IMF2 and 
the combined IMFs, it was observed that impact location B and C produced the best 
results concerning damage progression.  Furthermore, the side of the flange that hosted 
the sensors relative to the impact location did not seem to have any significant influence 
on the results.   
Concerning the influence of hammer tips, the two impulse hammer tips yielded 
approximately similar results overall, with the plastic tip producing more consistent 
results when considering the variability in EMD energies (Table A-8 through Table A-
11). However, taking into account the results from session 2 that indicated that a 
frequency bandwidth that surpasses 2 kHz could produce improved damage resolution, it 
was concluded that the aluminum tip would be used for the subsequent testing.  
It is interesting to note that for the results produced by combining the IMFs (A-9 
through A-16), PZT10, PZT10, PZT12, and PZT14 yielded the most consistent damage 
indicators.  
Last but not least, the pass-band frequency of 0-2000 Hz appears to be the most 
useful and applicable filtering frequency across all three sessions; it will therefore be 
used as the first pass-band frequency for the future analyses.  
 
 
75 
 
CHAPTER 4 
DAMAGE DETECTION OF AN IN-SERVICE CONDENSATION  
PIPELINE JOINT 
Julie Briand, Davood Rezaei and Farid Taheri
 
Proceedings of SPIE: Nondestructive Characterization for Composite Materials, 
Aerospace Engineering, Civil Infrastructure, and Homeland Security 2010 
Volume 7649, 76491T, doi:10.1117/12.847633 
 
4.1  ABSTRACT 
The early detection of damage in structural or mechanical systems is of vital 
importance. With early detection, the damage may be repaired before the integrity of the 
system is jeopardized, resulting in monetary losses, loss of life or limb, and 
environmental impacts. Among the various types of structural health monitoring 
techniques, vibration-based methods are of significant interest since the damage location 
does not need to be known beforehand, making it a more versatile approach. The non-
destructive damage detection method used for the experiments herein is a novel 
vibration-based method which uses an index called the EMD Energy Damage Index, 
developed with the aim of providing improved qualitative results compared to those 
methods currently available. As part of an effort to establish the integrity and limitation 
of this novel damage detection method, field testing was completed on a mechanical pipe 
joint on a condensation line, located in the physical plant of Dalhousie University. 
Piezoceramic sensors, placed at various locations around the joint were used to monitor 
the free vibration of the pipe imposed through the use of an impulse hammer. Multiple 
damage progression scenarios were completed, each having a healthy state and multiple 
damage cases. Subsequently, the recorded signals from the healthy and damaged joint 
were processed through the EMD Energy Damage Index developed in-house in an effort 
to detect the inflicted damage. The proposed methodology successfully detected the 
inflicted damages. In this paper, the effects of impact location, sensor location, frequency 
bandwidth, intrinsic mode functions, and boundary conditions are discussed. 
76 
 
Keywords: Vibration-based damage detection, structural health monitoring, 
piezoceramic sensors, mechanical pipeline joint. 
4.2   INTRODUCTION 
Bolted joints are widely used in aerospace, mechanical and structural systems. 
Wherever bolted systems exist in environments which impose vibration, shocks or 
thermal cycling on the threaded fasteners, they are at risk of self-loosening caused by 
such dynamic loads (Hess, 1998), especially if proper joint assembly guidelines have not 
been followed. If the loosened bolts are not detected early, ignorance of the problem may 
lead to joint failure resulting in monetary losses, environmental impacts and possibly loss 
of life or limb. According to the Industrial Fasteners Institute (1995), self-loosening in 
conjunction with fatigue failure is the most frequent cause of failure of dynamically 
loaded bolted joints.  
Structural health monitoring methods may be used for early detection of a joint’s loss 
of integrity. In the field of structural health monitoring, Vibration Based Damage 
Detection (VBDD) is a non-destructive evaluation technique rapidly expanding in 
popularity. The basis for most VBDD methods is that the occurrence of damage in a 
system alters the dynamic properties of the system. Accordingly, system damage may be 
identified by measuring deviations in the structural dynamic characteristics associated 
with the changes in the physical system caused by the damage. For instance, Moniz et al. 
(2005) used a multivariate, attractor based approach for feature extraction from the time 
series information resulting from a system’s vibration. This approach was used to detect 
loosened bolts at either end of a composite beam bolted to a plate. Furthermore, 
Ritdumrongkul et al. (2004) applied the electrical impedance method to two aluminum 
beams bolted together, modeled using the Spectral Element Method. By knowing that the 
electrical impedance is related to the mechanical impedance of the structure, 
Ritdumrongkul et al. (2004)
 
were able to detect the loosened bolt through changes in the 
electrical impedance caused by changes in the structural dynamic properties.   
 A considerable amount of research in the field of VBDD has been concentrated on 
assessment of modal parameters as damage metrics. Concerning the loosening of 
threaded fasteners, Todd et al. (2004) conducted experiments on a beam with threaded 
77 
 
fasteners constraining the ends and used changes in the resonant frequencies and mode 
shapes to attempt to identify bolt loosening. In another study, one approach that Caccese 
et al. (2004) used to detect bolt load loss was using a low frequency modal analysis. 
However, in practical applications modal parameters are not always successful in 
identifying damage due to their sensitivity to boundary conditions, sensor location, sensor 
spacing, nonlinearities, and environmental effects such as temperature and humidity 
(Pines and Salvino, 2006). 
To overcome the above mentioned drawbacks, another category of VBDD methods 
known as time series analysis techniques has evolved. The concept behind these methods 
is to extract damage metrics based on the analysis of the vibration signals acquired from 
the structure before and after damage. The Fourier transform and Wavelet Transform are 
the two well known techniques in this area that have been previously employed by 
researchers. For instance, Krawczuk (2002) introduced an optimization process based on 
a genetic algorithm for small crack detection in beams using the Fourier transform of the 
dynamic responses of the structure. Quek et al.
 
(2001) examined the distribution of 
wavelet coefficients of a beam’s deflection profiles through which sudden changes were 
proposed as a crack indicator. 
Another time-series analysis, known as the Hilbert-Huang Transform (HHT), was 
introduced by Huang et al. (1998), which is an adaptive technique that decomposes the 
vibration signals in both time and frequency domains. This method has received 
significant attention for its use in structural health monitoring and has been successfully 
applied in the diagnosis of damage in building structures (Xu and Chen, 2004), (Lin et 
al., 2005). Additionally, Cheraghi et al. (2007) and Cheraghi and Taheri (2007) employed 
the decomposition part of the HHT known as Empirical Mode Decomposition (EMD) 
and proposed a damage index called the EMD Energy Damage Index (EMD-EDI) for 
structural health monitoring; subsequently verifying its applicability through numerical 
and experimental studies. The EMD-EDI utilizes an energy term obtained from EMD and 
compares the healthy and damaged states of a structure in order to identify the damage. 
In the present paper, the integrity of the EMD-EDI (Cheraghi and Taheri, 2007; 
Cheraghi et al., 2007) has been further investigated through an experimental study 
performed on an in-service condensation pipeline with a 12-bolt mechanical joint. The 
78 
 
joint/pipe was excited with an impulse hammer and the resulting free vibration was 
captured via piezoceramic sensors bonded on the system in the vicinity of the joint. 
Damage was created by loosening successive bolts to a fully loose condition. The signals 
obtained from the piezoceramic sensors were processed through a MATLAB code 
developed in-house to carry out the signal processing, the Empirical Mode 
Decomposition and the calculation of  the EMD energy term for each healthy and damage 
state. The effects of impact location, sensor location, frequency bandwidth, intrinsic 
mode functions, and boundary conditions are discussed in relation to the successful 
detection of damage. 
4.3   METHODOLOGY  
In 1998, Huang et al. (1998) introduced a new adaptive signal processing method 
called the Hilbert-Huang Transform (HHT), which nearly satisfies all the requirements 
for processing linear, non-linear, stationary, and non-stationary signals. The HHT 
consists of two main parts: EMD and the Hilbert transform. The EMD method 
decomposes a real signal into a collection of simpler modes (IMFs). The IMFs are unique 
intrinsic oscillating modes within the data and are associated with energy at different time 
scales and contain important data characteristics (Huang et al., 1998). The HHT applies 
the original Hilbert transform to the extracted IMFs and derives the local energy and 
instantaneous frequencies, thereby providing an energy-frequency-time distribution of the 
data. 
The Empirical Mode Decomposition employs a sifting process to extract the IMFs. 
Each IMF satisfies two conditions. First, the number of extrema and the number of zero 
crossings are either equal or differ at most by one. Secondly, the average of the 
envelopes, defined by the local maxima and local minima, is zero. The first IMF 
component contains the shortest period component of the signal. To derive other IMFs, 
the first IMF is removed from the signal and the residue is considered as the new signal 
and the sifting process is performed to obtain the second IMF component. This procedure 
is repeated for all subsequent residues to derive the longer period components. After 
extracting all IMFs, the original signal is decomposed into   empirical modes,   , and a 
79 
 
residue,   , which can be either the mean trend or a constant. This process can be 
represented by the following mathematical relation:  



n
1i
ni rc)t(X  (4-1) 
In the present work, the damage index introduced by Cheraghi et al. (2007), and 
Cheraghi and Taheri (2007) was used for damage identification. This damage index is 
based on the energy of the vibration signal’s IMFs. According to this method, the 
dynamic response of the healthy structure during its free vibration is collected through 
sensors; the acquired signals are then passed through a band-pass filter to ensure the 
existence of the first natural frequency within the data. This is followed by extraction of 
the IMFs through EMD. Finally, the energy of the desired IMF for each sensor is 
established by: 
         
  
 
   (4-2) 
The above procedure is subsequently repeated for the same structure during its 
damaged state to once again determine the energy of each signal’s first IMF. The last step 
is the application of the EMD-EDI to the extracted IMF energy of each sensor’s signal: 
         
                 
        
      (4-3) 
Once the EMD-EDI is calculated for each sensor, the existence and severity of the 
damage may be determined by associating high index values with the existence of 
damage close to the respective sensors. 
Cheraghi and Taheri (2007) suggested applying a band-pass filter to the data to keep 
only the first frequency component. However, in the experimental investigations carried 
out by Rezaei and Taheri
 
(2009), it was concluded that this method only works well for 
systems where the first frequency is the dominant vibrating frequency of the structure. 
Otherwise, more frequency components should be included within the band-pass filter in 
order to avoid the elimination of damage sensitive frequency components.  
In summary, the methodology followed in the experiment documented in this paper is 
as follows: Step 1: Monitor the vibration of a system in its healthy state using the 
80 
 
response of sensors within the system. Step 2: Once again, monitor the vibration of the 
system in its damaged state using the responses of sensors within the system. Step 3: 
Analyze the signals and evaluate the EMD-EDI which is based on the variation of IMF 
energy between the healthy and damage system states. Step 4: Establish whether damage 
exists based on the comparison of the EMD-EDIs evaluated in the last step.  
4.4   EXPERIMENTAL STUDY 
A 12-bolt mechanical joint (shown in Figure 4-1), located in the underground tunnels 
at the Central Services Building at Dalhousie University, NS, was chosen for the case 
studies due to ease of access. The joint was located on an insulated condensation line, 
1.524 m to the right of an elbow and 1.092 m to the left of an anchor (support), 
respectively shown in Figure 4-2 (a) and (b). The pipe had a diameter of 203 mm and the 
bolts were 33.34 mm in diameter. To employ the first and third steps of the EMD damage 
detection methodology, the set-up for this experimental study, detailed in Figure 4-1, was 
accomplished as follows.  
To capture the dynamic response of the joint, 8 piezoceramic (PZT) sensors were 
bonded to the flange and pipe in the vicinity of the joint. The sensors were labeled PZT1 
through PZT8 and are denoted by the numbers 1 through 8 in Figure 4-1. The sensors 
were created from type PZT-5H sheets, 1.0 mm thick, available from Piezo Systems Inc. 
(Cambridge, MA, USA). The sensors were 24.0 mm long by 10.0 mm wide rectangles 
and polarized through the thickness. In order to monitor and compare the response of 
different locations on the pipe and flange, the sensors were bonded in the following 
locations: PZT1 was attached to the left vertical face of the flange between Bolt1 and 
Bolt12. PZT2 and PZT3 were respectively bonded to the top and bottom of the pipe at a 
distance of 82.6 mm from the joint centre to the edge of the PZT (see Figure 4-1 (c)). 
Mirroring the placement of PZT2, PZT4 was attached to the top of the pipe to the right of 
the joint at a distance of 82.6 mm from the joint centre to the edge of the PZT sensor (see 
Figure 4-1 (c)). Likewise, PZT5 was placed on the vertical face of the right flange to 
mirror the placement of PZT1. In order to obtain data for the side of the pipe, PZT7 and 
PZT8 were respectively placed to the right and left sides of the joint on the side of the 
pipe at a distance of 82.6 mm from the joint centre to the edge of the PZT sensors.  
81 
 
To induce free vibration in the system, the pipe was impacted at two locations on the 
pipe with a piezoelectric impulse hammer (model 5800B5) manufactured by Dytran 
Instrument Inc. (Chatsworth, CA, USA). An aluminum hammer tip was utilized for all 
testing. The two impact locations chosen for testing were Impact Locations B and D, 
shown in Figure 4-1 (c), directly opposite to one another. A 12.7 mm chrome alloy steel 
ball was bonded to the pipe at each impact location as a target in order to maintain 
consistency during testing. Impact locations B and D were respectively located on the 
right and left sides of the joint at a distance of 298.5 mm from the joint centre. The two 
locations were placed on either side of the joint in an attempt to determine what effect the 
different boundary conditions would have on the results. 
 
a)                                                           b) 
 
c) 
Figure 4-1  Piezoceramic sensor locations for part 2 a) Left side view of joint b) Right side view 
of joint c) Top view of joint. 
82 
 
 
a)                                                              b) 
Figure 4-2  Pipeline boundary conditions a) Elbow at left side of joint b) Anchor at right side of 
joint. 
Furthermore, the impact location’s distance to the joint was chosen in an effort to 
compromise between the desire not to corrupt the sensor signal (from PZT2, PZT3, 
PZT4, PZT7 and PZT8) by an impact in close proximity to the sensors and the 
knowledge, gained from prior testing, which indicated that impacts further from the joint 
yielded less accurate results. As the condensation pipe originally was covered in 
approximately 44.5 mm of insulation, the insulation in the vicinity of the impact locations 
and sensor placement was removed. The PZT sensors and impact targets were attached to 
the pipe using Araldite 2014, after the surface area had been scoured using sandpaper and 
wiped free of dust and other residue. 
During vibration, the response of the impulse hammer and the sensors were 
simultaneously collected through two NI 9215 modules inserted in an NI cDAQ-9172 
data acquisition chassis manufactured by National Instruments Inc. (Austin, Texas, USA) 
and saved as an ASCII text file. With only eight data acquisition channels available, it 
was only possible to monitor the response of seven sensors concurrently, since the 
hammer signal was also required for analysis and required the eighth connector. 
Consequently, the results for scenarios 2 and 3 only give data for PZT2 through PZT8. It 
may be noted in the results section that scenario 1 only has three sensors whereas 
scenarios 2 and 3 have seven. This is due to scenario 1 occurring prior to the other 
83 
 
scenarios at which time only three sensors had been placed. The remaining sensors were 
placed after this scenario in an effort to more fully capture the joint response.  
The data collected for each healthy state and damage state, typically comprised of 10 
or 5 text files per state, was subsequently analyzed using a MATLAB code developed in-
house which served to: (a) normalize the data with respect to the hammer signal, (b) 
process the data using Empirical Mode Decomposition, and (c) calculate the EMD-EDI 
for each damage case. The onsite experimental set-up detailed above is depicted in Figure 
4-3.  
 
Figure 4-3  Experimental set-up. 
To accomplish step 2 of the damage detection methodology, listed in Table 4-1 are the 
three damage scenarios that were completed during testing. In order to preserve the 
joint’s functionality during the experiment, a state of damage was created by having a 
professional steamfitter loosen bolt(s). Of the twelve bolts on the joint, only six were 
accessible, therefore damage in the form of loosened bolts was applied to these six bolts, 
labeled B12 through B5 as illustrated in Figure 4-1. For each damage scenario, each bolt 
was fully loosened, past the stage of finger-tight. After each scenario was completed, the 
loosened bolts were retightened based on the steamfitter’s judgment; a torque wrench was 
purposely not used, in order to follow actual practice in the real situation. Therefore, 
testing for a new healthy state had to be completed after each scenario. 
84 
 
To obtain an accurate average energy for each sensor and for each damage case, 
multiple impacts were required for each state. For all scenarios, 10 impact scenarios were 
completed for each healthy case, whereas only 5 impact scenarios were completed for 
each damage case. The reasoning for the greater number of impacts for the healthy case 
was to ensure a solid base for the index evaluation. Also of note is that for scenario 1, 
only impact location B was used, whereas for scenarios 2 and 3 both impact locations B 
and D were used; the change in the number of impact locations is due to scenario 1 
occurring prior to the other scenarios when impact location D had not yet been placed. 
        Table 4-1  Damaged joint scenarios. 
Scenario State Bolts Loosened Damage Case 
Scenario 1 
Healthy Pipe 1   
Damage Case 1-1 Bolt #1 DC 1-1 
Damage Case 1-2 Bolt #1 and #2 DC 1-2 
Scenario 2 
Healthy Pipe 2   
Damage Case 2-3 Bolt #12 DC 2-3 
Damage Case 2-4 Bolt #12 and #1 DC 2-4 
Damage Case 2-5 Bolt #12, #1and #2 DC 2-5 
Scenario 3 
Healthy Pipe 3   
Damage Case 3-6 Bolt #5 DC 3-6 
Damage Case 3-7 Bolt #5 and #3 DC 3-7 
4.5   RESULTS AND DISCUSSION 
After monitoring and recording the free vibration of the pipe in its healthy state 
through the piezoceramic sensors, damage was introduced in the joint by loosening 
subsequent bolts and the free vibration of the pipe was recorded for each damaged state. 
As mentioned in the previous section, three damage scenarios were studied which 
included one bolt, two bolts, and in one scenario, three bolts completely loosened. 
Subsequently, in order to complete step 4 and 5 of the damage detection process, the 
recorded signals from the healthy and damaged pipe joint were processed and the EMD-
EDI were evaluated in an effort to establish the presence of the inflicted damage.  
85 
 
4.5.1 Result Analysis 
For the healthy state of scenario 1, five signals were chosen for each impact location 
and for the damage cases of scenario 1, three signals were used. As a part of the EMD-
EDI procedure, the data should be filtered before decomposing. By analyzing the data 
using different frequency bands, it was found that a low-pass filter with a cutoff 
frequency of 2000 Hz provided the best resolution for identification of damage for 
scenario 1. Figure 4-4 and Figure 4-5 illustrate the EMD-EDI calculated for sensors 
PZT1, PZT2 and PZT3, for scenario 1. The results shown in Figure 4-4 and Figure 4-5 
are respectively based on the energy of the first and second IMFs. The damage case 
labels used in all of the legends for the following graphs are explained in Table 4-1. 
 
Figure 4-4  Scenario 1, impact location B, IMF 1. 
As can be seen in Figure 4-4, of the three sensors used for scenario 1, all were 
successful in detecting the presence and progression of damage. Moreover, PZT1 and 
PZT2 provided the best damage identification in reference to index amplitude. The 
performance of PZT1 and PZT2 may be due to their close proximity to the damaged 
bolts, whereas PZT3 was located on the bottom surface of the pipe and far from the 
damaged area. After analyzing the results using the EMD-EDI based on the first IMF, the 
second IMF was also investigated as a case study. Figure 4-5 illustrates that the second 
IMF was also successful in damage identification of the joint.   
0
10
20
30
40
50
60
PZT1 PZT2 PZT3
E
M
D
-E
D
I
Sensor
DC 1-1
DC 1-2
86 
 
 
Figure 4-5  Scenario 1, impact location B, IMF 2. 
The same signal selection scheme used in scenario 1 was followed for the second 
scenario. However, for scenario 3, three signals were chosen for each impact location for 
both the healthy and damaged cases. By analyzing the data for scenario 2 and 3 using 
different frequency bands, it was found that frequency bands of 0-2000 Hz, and 1000-
2000 Hz both provided the best resolution for identification of the damage when both 
IMFs were taken into consideration. However, for the purposes of this paper, only the 
results from the frequency band of 0-2000 Hz will be shown. 
The bandwidth of 0-2000 Hz was chosen based on the exhibition of damage 
progression and damage index amplitude observed for the first scenario. Accordingly, a 
bandwidth of 0-2000 Hz was used to process all the data for scenarios 2 and 3. Figure 4-6 
and Figure 4-7 respectively illustrate the EMD-EDI calculated for impact locations B and 
D for sensors 2 through 8. The results shown in these figures are based on the energy of 
the first IMF.  
 
Figure 4-6  Scenario 2, impact location B, IMF 1. 
0
10
20
30
40
50
60
PZT1 PZT2 PZT3
E
M
D
-E
D
I
Sensor
DC 1-1
DC 1-2
0
10
20
30
40
50
60
70
80
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 2-3
DC 2-4
DC 2-5
87 
 
A comparison of Figure 4-6 and Figure 4-7 with Figure 4-4 reveals an interesting 
phenomenon, in that PZT2, the sensor giving the most consistent results for scenario 2, 
was also the sensor that gave the most promising results for scenario 1. Also, impact 
location D provided relatively better damage progression overall with four out of seven 
PZTs (PZT2, PZT4, PZT5, PZT8) showing full damage progression, whereas impact 
location B only showed full progression with three PZTs (PZT2, PZT6, PZT7). 
Furthermore, the amplitude of the damage indices was greater overall for impact location 
D. Both locations showed a damage progression for PZT2 with impact location B also 
illustrating relatively high amplitude and clear damage progression for PZT6 and PZT7, 
whereas location D also demonstrated comparable results for PZTs 4, 5 and 8. Once 
again, PZT3, the sensor furthest from the damage, did not give a clear indication of the 
inflicted damage for either impact location.  
 
Figure 4-7  Scenario 2, impact location D, IMF 1. 
 
Figure 4-8  Scenario 2, impact location B, IMF 2. 
After analyzing the results using the EMD-EDI based on the first IMF, the second 
IMF was also investigated as a case study. Figure 4-8 and Figure 4-9 illustrate that the 
0
20
40
60
80
100
120
140
160
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 2-3
DC 2-4
DC 2-5
0
20
40
60
80
100
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 2-3
DC 2-4
DC 2-5
88 
 
second IMF was also successful in damage identification, if not damage progression. 
However, a high damage index is the damage criterion with more significance because in 
most cases, the damage would be fixed before it progresses to a more dangerous 
condition. 
 
 
Figure 4-9  Scenario 2, impact location D, IMF 2. 
Figure 4-10 Figure 4-11 respectively illustrate the EMD-EDI calculated for impact 
locations B and D for sensors 2 through 8 for scenario 3. The results shown in the 
following two figures are based on the frequency band of 0-2000 Hz, and the energy of 
the first IMF. As can be seen from comparing Figure 4-10 Figure 4-11, impact location B 
provided better damage progression with five out of seven PZTs (PZT3, PZT4, PZT5, 
PZT7, PZT8) showing damage progression, whereas location D only showed progression 
with PZT8. However, similarly to scenario 2, the damage index amplitude for location D 
is significantly higher overall with each sensor clearly giving an indication of the 
inflicted damage. Both locations showed a damage progression for PZT 8 with location B 
showing additional damage progression for PZTs 3, 4, 5 and 7.  
0
10
20
30
40
50
60
70
80
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 2-3
DC 2-4
DC 2-5
89 
 
 
Figure 4-10  Scenario 3, impact location B, IMF 1. 
 
Figure 4-11  Scenario 3, impact location D, IMF 1. 
After analyzing the results using the EMD-EDI based on the first IMF, the second 
IMF was also investigated as a case study. Figure 4-12 and Figure 4-13 illustrate that the 
second IMF was also successful in damage identification, if not damage progression. 
 
Figure 4-12  Scenario 3, impact location B, IMF 2. 
0
10
20
30
40
50
60
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 3-6
DC 3-7
0
40
80
120
160
200
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 3-6
DC 3-7
0
10
20
30
40
50
60
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 3-6
DC 3-7
90 
 
 
Figure 4-13  Scenario 3, impact location D, IMF 2. 
It is interesting to note that the sensors closest to the damage, namely PZT7 and PZT8, 
yielded among the best results for scenario 3 except in the case of impact location D, 
IMF1. Even though the pipe is impacted at its apex, the impact excites many modes of 
vibration, not necessarily in the vertical direction, thereby producing strain in various 
sensors around the circumference of the pipe. It may be theorized from the results of the 
three scenarios that the sensors located closest to the damage have the highest probability 
of detecting the presence of damage.  
4.5.2 Discussion 
From Figure 4-4 through Figure 4-13, it can be observed that the EMD-EDI amplitude 
for several of the sensors (PZT2, PZT7 and PZT8) in the vicinity of their respective 
damages is in general more consistent with varying impact locations and IMFs and 
clearly demonstrates the capability of the proposed damage index in identifying and 
localizing the damage. Taking each damage index individually, each sensor was able to 
detect each damage state as compared to the healthy state. Furthermore, at varying times, 
several sensors were able to detect not only the damage, but the damage progression of 
the successively loosened bolts. From an analysis of the above results, it is clear that the 
placement of the sensors on and around the flanges has an effect on the resulting EMD-
EDI amplitude and progression. However, no clear pattern is as yet discernable regarding 
why certain sensors demonstrate a greater sensitivity to the damage under certain 
conditions.  
0
10
20
30
40
50
60
PZT2 PZT3 PZT4 PZT5 PZT6 PZT7 PZT8
E
M
D
-E
D
I
Sensor
DC 3-6
DC 3-7
91 
 
Additionally, a comparison of the above results between IMF1 and IMF2 illustrate that 
an analysis based on IMF1 for impact location D yields greater index amplitudes than for 
impact location B. It is hypothesized that the higher index magnitudes were due to the left 
end of the pipe (see Figure 4-1) not being constrained, but instead with more freedom to 
vibrate than the anchor support allowed the pipe on the other side of the joint. It is worth 
noting that this greater index amplitude does not appear for the analysis based on the 
second IMF. These results serve to indicate that while the second IMF may indeed 
demonstrate good qualitative results in some cases, in other cases the results are much 
less useful than those of IMF1 and are therefore not helpful in damage identification for 
this experimental set-up. In real situations of damage detection for health monitoring, 
once an indication of damage at a joint is noted, one would not wait to see damage 
progression; therefore, a high EMD-EDI amplitude is the most favorable indicator which 
would serve to indicate based on the previous results, that when testing, impacting the 
side of the joint with a less restrained boundary condition will give relatively better 
results than the more constrained side.  
Several items should be taken as variables in this experiment that possibly had an 
untoward influence on the results. First, no method was available to monitor the fluid 
flow and therefore it is unknown how much flow variation occurred during the tests, and 
what affect it may have had on the results. Second, due to corrosion on the pipe, the bond 
of PZT3, PZT7 and PZT8 to the pipe may not have been able to transmit the appropriate 
strains from the pipe to the sensors. Finally, without a torque wrench available, the bolts 
were retightened after each scenario relying only on the steamfitter’s experience. 
Moreover, as it was impossible to access the entire joint with the tools available, it was 
not possible to verify the torque of each bolt in the joint for the healthy condition which 
could introduce more uncertainty in the results. 
4.6   CONCLUSIONS 
The objective of this study was to experimentally confirm the use of the EMD-EDI for 
the evaluation of the damaged condition of an in service 12-bolt joint mechanical joint 
located on a condensation pipeline. The EMD-EDI is based on the variation of the IMF 
energy, decomposed from the vibration signals using EMD, prior to and following the 
92 
 
occurrence of damage in a system. Three damage scenarios were analyzed, with each 
damage scenario progressing from a healthy state, to having one bolt completely 
loosened, then two and then in the second scenario, three.  
This experimental study documented and herein demonstrated that the EMD-EDI can 
successfully detect the presence of damage. However, it is important to note that sensor 
placement, boundary conditions and therefore the impact locations have a significant 
influence on the results which should be carefully considered when choosing sensor and 
impact location placement. Moreover, additional work is required to determine if the 
proximity of a sensor to a loosened bolt increases the ability of the sensor to detect 
damage. Furthermore, the band-pass filter used for signal filtering should be applied 
carefully to avoid filtering out frequency components that contain meaningful 
information required for the proposed damage detection method. 
4.7   ACKNOWLEDGEMENTS 
The authors gratefully acknowledge the financial support of the Natural Sciences and 
Engineering Council of Canada (NSERC), of Petroleum Research Atlantic Canada 
(PRAC) and the assistance of Chris Leslie, Chief Operating Engineer with Facilities 
Management at Dalhousie University, NS, Canada.  
 
  
93 
 
4.8   REFERENCES 
Caccese, V., Mewer, R., & Vel, S. (2004). Detection of bolt load loss in hybrid 
composite/metal bolted connections. Engineering Structures, 26(7), 895.  
Cheraghi, N., & Taheri, F. (2007). A damage index for structural health monitoring based 
on the empirical mode decomposition. J. of Mechanics of Materials and Structures, 
2(1), 43-62.  
Cheraghi, N., Riley, M. J., & Taheri, F. (2007). Application of Hilbert-Huang transform 
for evaluation of vibration characteristics of plastic pipes using piezoelectric sensors. 
Structural Engineering Mechanics, 25(6), 653-674.  
Hess, D. P. (1998). Vibration and shock induced loosening. In Handbook of bolts and 
bolted joints, J. H. Bickford, & S. Nassar (Eds.), (pp. 757). Marcel Dekker, Inc.: 
New York.  
Huang, N. E., Shen, Z., Long, S. R., Wu, M. C., Shih, H. H., Zheng, Q., Nai-Chyuan, Y., 
Tung, C. C., & Liu, H. H. (1998). The empirical mode decomposition and the 
Hilbert spectrum for nonlinear and non-stationary time series analysis. Philosophical 
Transactions of the Royal Society A: Mathematical, Physical and Engineering 
Sciences, 454(1971), 903-995.  
Industrial Fasteners Institute: Division VI Aerospace Fasteners. (1995). Test application 
handbook relating to mechanical fasteners. Industrial Fasteners Institute: Cleveland, 
Ohio.  
Krawczuk, M. (2002). Application of spectral beam finite element with a crack and 
iterative search technique for damage detection. Finite Elements in Analysis and 
Design, 38(6), 537-548.  
Lin, S., Yang, J. N., & Zhou, L. (2005). Damage identification of a benchmark building 
for structural health monitoring. Smart Materials & Structures, 14(3), S162-S169.  
Moniz, L., Nichols, J., Nichols, C., Seaver, M., Trickey, S., & Todd, M. (2005). A 
multivariate, attractor-based approach to structural health monitoring. Journal of 
Sound and Vibration, 283(1-2), 295-310.  
94 
 
Pines, D., & Salvino, L. (2006). Structural health monitoring using empirical mode 
decomposition and the hilbert phase. Journal of Sound and Vibration, 294(1-2), 97-
124.  
Quek, S. T., Wang, Q., Zhang, L., & Ang, K. K. (2001). Sensitivity analysis of notch 
detection in beams by wavelet technique. International J. of Mechanical Sciences, 
43, 2899-2910.  
Rezaei, D., & Taheri, F. (2009). Experimental validation of a novel structural damage 
detection method based on empirical mode decomposition. Smart Materials and 
Structures, 18(4). 
Ritdumrongkul, S., Abe, M., Fujino, Y., & Miyashita, T. (2004). Quantitative health 
monitoring of bolted joints using a piezoceramic actuator-sensor. Smart Materials 
and Structures, 13(1), 20-29.  
Todd, M. D., Nichols, J. M., Nichols, C. J., & Virgin, L. N. (2004). An assessment of 
modal property effectiveness in detecting bolted joint degradation: Theory and 
experiment. Journal of Sound and Vibration, 275(3-5), 1113-1126. Retrieved from 
http://dx.doi.org/10.1016/j.jsv.2003.10.037  
Xu, Y. L., & Chen, J. (2004). Structural damage detection using empirical mode 
decomposition: Experimental investigation. Journal of Engineering Mechanics, 
130(11), 1279-1288.  
 
 95 
 
CHAPTER 5 
DAMAGE DETECTION OF A PRESSURIZED STEEL PIPELINE JOINT  
USING AN EMPIRICAL MODE DECOMPOSITION-BASED ENERGY  
DAMAGE INDEX 
Julie Briand and Farid Taheri
 
Prepared Article Ready for Submission 
 
5.1  ABSTRACT  
The development of safe and reliable structural health monitoring techniques for 
pipelines is vital due to their importance for the safe and reliable transportation of fluids 
necessary for our infrastructure. The non-destructive damage detection method used for 
the experiments herein is a novel vibration-based method which uses an index called the 
Empirical Mode Decomposition Energy Damage Index, developed with the aim of 
providing improved qualitative results compared to those methods currently available. As 
part of an effort to establish the integrity and limitation of this novel damage detection 
method, experiments were conducted on a pressurized steel pipeline with a bolted joint. 
Piezoceramic sensors, placed at various locations around the joint were used to monitor 
the free vibration of the pipe imposed through the use of an impulse hammer. Multiple 
damage progression scenarios were completed, each consisting of a healthy state and 
multiple damage cases. Subsequently, the recorded signals from the healthy and damaged 
joint were processed using the Empirical Mode Decomposition Energy Damage Index in 
an effort to detect the inflicted damage. The methodology was used with varying degrees 
of success to detect the inflicted damages. In this paper, the effects of impact location, 
sensor location, frequency bandwidth, and the influence of the order of the intrinsic mode 
functions on the results are discussed. 
 
Keywords: Vibration-based damage detection; piezoelectric sensors; structural health 
monitoring; bolted pipeline joint; empirical mode decomposition. 
 
 96 
 
5.2   INTRODUCTION  
Pipelines are an integral part of today’s residential and industrial infrastructure. They 
are essential for the safe and effective transportation of natural gas, oil, water, waste 
products, and communication/power cables. The stability and reliability of pipelines may 
be affected by ground movement, environmental conditions, the interaction with the 
fluids they carry, and damage caused by external hazards at any time throughout their 
service life. These factors make regular pipeline inspection and maintenance especially 
important. 
Moreover, damage assessment of pipelines is critical for economic and community 
recovery after natural disasters (Park and Inman, 2005). An immediate damage 
assessment may aid in preventing fires, explosions, and pollution from broken gas and 
sewage lines (Park and Inman, 2005). Dependent on the contents and pressure of the 
pipeline system, leakage or failure may have potentially catastrophic consequences. As 
one of the critical components of a pressurized system, the failure of bolted joints may 
result in disastrous outcomes (Energy Institute (Great Britain), 2007).  
One of the most common causes of bolted joint failure is vibration induced bolt 
loosening, if the joint is exposed to environments which impose dynamic loading such as 
vibration or shocks on the threaded fasteners (Hess, 1998; Industrial Fasteners Institute, 
1995; Pai and Hess, 2002). This failure mode is exacerbated by the incorrect and 
uncontrolled assembly practices of the majority of bolted joints (Corbett, 1998) leading to 
an insufficient clamping force which contributes to joint failure.  
To date, traditional non-destructive methods used to assess damaged joints include 
visual inspection, the cylindrical guided wave technique (CGWT), phased array 
ultrasonics, and time-of-flight ultrasonics (Energy Institute (Great Britain), 2007). 
However, quick and effective prioritization of actions after a disaster is infeasible with 
manual inspection (Johnson et al., 2004). For this reason and those mentioned above, the 
development of safe and reliable structural health monitoring techniques for pipelines is 
vital for operational reliability, reduction in maintenance costs, and operator/public 
safety. However, the amount of research performed on the structural health monitoring of 
bolted joints has been somewhat limited.  
 97 
 
One of the widely studied qualitative SHM techniques currently under development to 
assess bolted joint damage is the electromechanical impedance-based method (Yan et al., 
2007) which takes advantage of the direct relation between the electrical impedance of a 
bonded PZT patch to the mechanical impedance of the host structure. The damage 
location is ascertained due to the known localized detection area surrounding each sensor 
and the damage severity is qualitatively assessed using a damage index which measures 
the change in the impedance before and after damage (Ritdumrongkul and Fujino, 2006). 
For example, Park et al. (2003) used this local diagnostic method and performed 
preliminary experimental testing on a single bolt lap-joint in an effort to monitor bolt 
preload loss in real time. To improve upon the impedance method and make it more 
economically feasible, Peairs et al. (2004) experimentally investigated the use of a low 
cost impedance method on a model pipeline with bolted joints which uses a digital signal 
analyzer with an Fast Fourier Transform (FFT) function instead of the very expensive 
impedance analyzer.  
Ritdumrongkul et al. (2004) also used the impedance method in conjunction with a 
spectral element method (SEM) model for damage characterization. Laboratory 
experiments were performed on a suspended two-joint and four-joint aluminum beam 
where damage was simulated by loosening bolts and the numerical models were used to 
quantitatively locate and assess the level of damage.  
However, even though it has been successfully illustrated that the impedance method 
is successful in damage detection of bolted joints, the equipment required to perform this 
testing is expensive (Peairs et al., 2004) and therefore unfeasible in situations with lower 
budgets.  
Vibration Based Damage Detection (VBDD) is another category of non-destructive 
evaluation techniques that has been rapidly growing in popularity over the past several 
decades. Damage detection using VBDT methods is based on the fact that modal 
properties such as frequencies, mode shapes, and the modal damping of a structure are 
functions of physical properties like mass, stiffness, and damping. When defects or 
damage occur in a structure, some or all of these physical properties may change and lead 
to alterations in the dynamic properties of the system. For a comprehensive review of the 
 98 
 
advances in vibration based SHM, the reader is directed to Doebling et al. (1996; 1998) 
and Carden and Fanning (2004).  
In recent years, structural health monitoring via modal characteristics has emerged as 
one category of vibration damage detection that has received attention by many 
researchers. Modal based damage detection methods rely on alterations in the modal 
parameters of a structure such as natural frequencies/mode shapes, modal strain energy 
etc. for an indication of damage.  
Modal parameters have been studied as damage sensitive features with different 
degrees of success. For example, Shi (2000) employed modal strain energy for the 
localization and quantification of damage numerically and experimentally on a single-
bay, two-story portal frame structure. Damage was created in the structure by loosening 
the joints. In another work, Todd et al. (2004) experimentally and theoretically assessed 
modal property effectiveness to detect bolted joint degradation in the supports of a beam.  
However, the use of modal methods is limited due to their sensitivity to boundary 
conditions, sensor spacing, environmental effects such as temperature and moisture and 
nonlinearities (Pines and Salvino, 2006). Additionally, most modal based damage 
identification methods require numerical simulations of the specified structure which 
decreases the practicality and effectiveness of the approach (Rezaei and Taheri, 2009). 
Furthermore, Todd et al. (2004) also stated that only significant preload losses are 
typically detectable when vibration-based techniques such as modal analysis are 
employed (Fasel et al., 2009), since the resulting frequencies are too low to identify small 
changes in preload loss, and localized damage (Doebling et al., 1998).  
Non-modal based methods include autoregressive approaches, neural networks, time 
series analyses and various other pattern recognition approaches (Nichols, 2003). 
Research using auto-regressive modeling was conducted by Fasel et al. (2009). 
Experiments and numerical analyses were performed to detect bolt preload reduction 
using a chaotic structural excitation with a damage detection algorithm based on auto-
regressive modeling. Signals were applied to the structure using macro fiber composite 
(MFC) patches.                                                     
For instance, Michaelides et al. (2008) used a statistical time series model called the 
Statistical Power Spectral Density-based method to a thin plate with joints.  Nichols 
 99 
 
(2003) explored the use of the steady-state dynamic analysis for damage detection instead 
of the traditional transient or stochastic vibration analysis. Experimental testing was 
performed on a scaled three story frame for which damage was created by bolt removal. 
Continuing on with the research into steady-state dynamics analysis, Todd et al. (2004) 
also used time series analysis to detect preload loss in a bolted joint in an aluminium 
frame structure.  
Another time-series technique, the Hilbert-Huang transform (HHT), was developed by 
Huang et al. (1998) and has been successfully applied to structural health monitoring 
techniques in mechanical and civil systems (Loutridis, 2004; Pines and Salvino, 2006). 
The HHT is an adaptive signal processing technique that is able to produce signal 
decomposition in time and frequency domains, thus providing valuable information for 
damage detection purposes (Rezaei and Taheri, 2010).  
Empirical Mode Decomposition (EMD), the key component in the HHT, was used by 
Loutridis (2004) to detect and monitor the growth of a tooth root crack in a gear system. 
Loutridis illustrated that the defect evolution can be monitored by computing the energy 
of the intrinsic mode that has the greatest damage sensitivity, manifested through an 
increase in its envelope amplitude. Pines and Salvino (2006) experimentally investigated 
the capability of HHT by applying it to the damage detection of a three story building 
model. It was concluded that the results on the laboratory experimental structure were 
promising; they could determine the presence and location, and severity of damage with 
respect to a baseline. Cheraghi and Taheri (2008) completed a theoretical investigation of 
the vibration characteristics of a six degree of freedom mechanical system to ascertain the 
capability and integrity of EMD with respect to assessing performance in structures.   
Cheraghi et al. (2005) introduced a novel structural health monitoring method based 
on the HHT, called the Empirical Mode Decomposition Energy Damage Index (EMD-
EDI). This method combines a time-series analysis technique with a local diagnosis 
approach through the use of piezoceramic sensors. Cheraghi et al. investigated the 
integrity of the EMD-EDI by performing a set of experiments on simply supported plastic 
pipes with adhesively bonded joints. The vibration response of the healthy and damaged 
pipes was monitored using piezoelectric patches bonded at the joint location. It was 
concluded that the index could effectively detect the joint damage.   
 100 
 
The EMD-EDI was further investigated by Cheraghi and Taheri (2007) who 
numerically evaluated the integrity of the method using Finite Element simulations. The 
subject of the investigation was a cantilevered aluminium pipe with various forms of 
corrosion damage represented by the removal of material (elements) in the pipe wall. 
Subsequently, Rezaei and Taheri (2009) experimentally employed the EMD-EDI for 
damage detection of cantilevered steel pipes with various degrees of simulated corrosion 
damage. Rezaei and Taheri (2010) then extended the use of the EMD-EDI to detect a 
flaw in pipeline girth welds in both pressurized and non-pressurized conditions.  
Rezaei and Taheri (2010; 2009) determined that the EMD-EDI was effective in 
detecting and localizing single and multiple areas of damage as well as qualitatively 
assessing the damage severity. As such, this method holds promise for an effective an 
inexpensive system for structural health monitoring. 
5.3   MOTIVATIONS AND OBJECTIVES  
The objective of the research summarized in this paper was to further examine the 
effectiveness and integrity of the EMD-EDI method applied to bolted joints. With the 
limited amount of structural health monitoring research performed on bolted joints, it was 
thought to be an ideal topic to extend the research previously conducted concerning the 
evaluation of the integrity of the EMD-EDI method. In an effort to determine the optimal 
parameters for damage detection, the effects of two different impact locations and six 
different sensor locations were investigated. Furthermore, the effect of frequency 
bandwidth, used for signal filtering, and the influence of the order of the intrinsic mode 
functions on the results were investigated. 
5.4   EMPIRICAL MODE DECOMPOSITION  
In 1998, Huang et al. (1998) developed the Hilbert-Huang transform (HHT) in an 
effort to create an improved analysis method of nonlinear and non-stationary signals and, 
according to Loutridis (2004), to create a method for accurate time and frequency 
localization. The key component of the HHT is a method called the Empirical Mode 
Decomposition (EMD), which is used in conjunction with the Hilbert transform.  
 101 
 
The first step of the HHT process is the decomposition of a signal using EMD into 
components. EMD assumes that every signal is composed of a number of unique intrinsic 
oscillating modes called Intrinsic Mode Functions (IMFs). The IMFs are associated with 
energy at different time scales and contain important characteristics of the data (Huang et 
al., 1998). The second step is to apply the Hilbert transform to the decomposed IMFs and 
construct an energy-frequency-time distribution, known as the Hilbert Spectrum, in order 
to ascertain the local energy and instantaneous frequency.    
The Empirical Mode Decomposition employs a sifting process to extract the IMFs, 
thereby providing a description of the frequency contents of a signal. The procedure of 
the sifting process is schematically illustrated in Figure 5-1. To extract the IMFs for a 
given time signal x(t), two cubic splines are fitted through the local maxima and minima 
to produce the upper and lower envelopes, respectively. The upper and lower envelopes 
should cover all the data. Then the mean of these two splines or envelopes (m1) is 
subtracted from the original signal. The difference between the signal and the mean is 
called the first component (h1): 
At this stage h1 is treated as the original time signal and the sifting process is repeated to 
get the next component by: 
                     (5-2) 
where k is the operation (sifting) cycle number. At each step the sifting process produces 
a more symmetric signal with respect to zero mean. The above procedure is schematically 
illustrated in Figure 5-2. If the sifting process continues to the extreme, it will remove the 
physically meaningful amplitudes and fluctuations. So, Huang et al. (1998) proposed a 
criterion based on limiting the size of the standard deviation, SD, computed from two 
successive sifting results. This stoppage criterion can be expressed as: 
     
                   
 
       
    
 
 
   
 (5-3) 
           (5-1) 
 102 
 
If the SD is smaller than a predetermined value (usually between 0.2 and 0.3), the sifting 
process is stopped and h1k  is called the first IMF component (c1) of the signal: 
khc 11   (5-4) 
 
Figure 5-1  A schematic representation of one sifting cycle a) The original signal b) The signal in 
thin solid line; the upper and lower envelopes in dot-dashed lines; the mean in thick solid line c) 
The difference between the signal and mean (adapted from Rezaei and Taheri, 2009). 
The first IMF component, c1, contains the finest scale or the shortest period component of 
the signal. To derive other IMFs, the first component is removed from the signal: 
           (5-5) 
 103 
 
The residual, r1, contains the larger scales or longer period components. Then, the 
residual is considered as the new signal and the sifting process is performed on it to 
obtain the second IMF component. This procedure is repeated for all subsequent residuals 
to derive the longer period components. At each repetition the signal is modified with 
respect to the obtained IMF as following: 
           (5-6) 
After extracting all IMFs, the original signal is decomposed into n empirical modes and a 
residue, rn , which can be either the mean trend or a constant, as mathematically 
represented by: 
           
 
   
 (5-7) 
5.5   EMD ENERGY DAMAGE INDEX 
The research documented in the following paper used the novel structural health 
monitoring method introduced in 2005 by Cheraghi et al. (2005). This method is based on 
the HHT and is called the Empirical Mode Decomposition Energy Damage Index (EMD-
EDI). The EMD-EDI is based on the energy of certain IMFs of the vibration signals.      
In this method, once the dynamic characteristics of a healthy structure during free 
vibration are collected through sensors, the acquired signals are then passed through a 
band-pass filter to ensure the existence of the first natural frequency within the data. This 
is followed by the extraction of a specific IMF though Empirical Mode Decomposition. 
Subsequently, the energy of the IMF, for each sensor, is established by: 
            
  
 
 (5-8)  
The above procedure is then repeated for the same structure during its damaged state 
to once again determine the energy of each signal’s first IMF. The last step is the 
application of the EMD-EDI is to the extracted IMF energy of each sensor’s signal: 
 104 
 
         
                 
        
      (5-9)  
The above damage index provides the percentage difference in the structure’s energy 
in its healthy and damage states. Once the EMD-EDI is calculated for each sensor, the 
existence and severity of the damage may be determined by relative comparison of the 
indices: high index values reveal the existence of damage close to the respective sensors. 
In summary, the methodology followed in the experiment documented herein is as 
follows which is also illustrated in the following flow chart (Figure 5-2):  
Step 1:  Monitor the vibration of a system in its healthy state using the response of 
sensors within the system.  
Step 2:  Once again, monitor the vibration of the system in its damaged state using 
the responses of sensors within the system.  
Step 3:  Analyze the signals and evaluate the EMD-EDI; which is based on the 
variation of IMF energy between the healthy and damage system states.  
Step 4:  Establish whether damage exists based on the comparison of the EMD-EDIs 
evaluated in the last step.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure 5-2  Damage detection methodology (adapted from Rezaei and Taheri, 2009). 
Process the piezoceramic signals through 
the Empirical Mode Decomposition 
Evaluate the EMD energy of 
the desired IMF of the 
healthy structure 
Evaluate the EMD energy of 
the desired IMF of the 
damaged structure 
Monitor the vibration of 
the healthy structure 
Monitor the vibration of the 
structure in its damaged state. 
Calculate the EMD-EDI for each sensor.       
 Determine and report presence and 
location of damage.  
 105 
 
5.6   EXPERIMENTAL SET-UP 
The test specimen used in this study was an ANSI/ASME compliant (1985) standard 
steel pipe commonly used in oil and gas pipelines. The specifications of the schedule 40 
SA106 Grade B pipe, with a nominal pipe size of six, are listed in Table 5-1. An 
overview of the experimental set-up is illustrated in Figure 5-3. As shown, the jointed 
pipe is suspended from either end with nylon rope from two steel saw horses. To comply 
with normal operating pressures in oil and gas pipelines, the pipe joint system was loaded 
with an internal water pressure of 10 MPa during all testing. The internal pressure of 10 
MPa was attained through use of a GDS Instruments Limited 125 MPa High Pressure 
Controller, depicted in Figure 5-3. 
   Table 5-1  ANSI/ASME pipe dimensions and material properties. 
Length  4.036 m Density 7850 kg/m
3
 
Outer Diameter 168.3 mm E  200  GPa 
Wall Thickness 7.10 mm v 0.3  
 
For the bolted joint, ASTM and ASME/ANSI compliant forged steel flanges were 
chosen of the following type:  NPS: 6, class: 600, raised face, slip on, and material type 
A105N.  The spiral wound metallic gaskets used were NPS: 6, class: 300, and style WRI 
304-GRI 304 IR. SAE Grade 8 UNC hex head bolts were used (25.4 mm by 152.4 mm) 
with corresponding sized nuts, and grade 8 flat washers as well as lock washers.  
To create the mechanical joint in the pipe, the pipe was cut in two approximately equal 
lengths, 2019 mm (left half) and 2016 mm (right half), and the two flanges were welded 
onto the pipe.  The flanges were continuously welded to the pipe on the inside of the joint 
with a small overlap over the pipe edge of approximately 13 mm, and were also 
continuously welded to the pipe on the outside of the joint. The ends of the pipe were 
sealed using square steel plates shown in Figure 5-4. The plates (176 mm by 176 mm by 
13 mm) were welded to the outside of the pipe. 
For this specific joint configuration, the maximum bolt torque was determined to be 
786 N-m. This value was calculated using the ASME - International Boiler and Pressure 
Vessel Code by the gasket manufacturer. Before bolting, the bolts and nut faces were   
 106 
 
 
Figure 5-3  Experimental set-up overview. 
 
 
a)                                                                     b) 
Figure 5-4  a) Boundary condition b) Pressure connection.   
 
 Nylon rope_  
 End Plate_  
 Saw Horse_  
 Pipe_  
 End Plate_  
 Connection_  
 Water Supply_  
 107 
 
lubricated using anti-seize lubricant manufactured by Permatex (Milton, Ontario). The 
procedure to tighten the bolts was the industry standard, which followed a criss-cross 
pattern across the face of the flange following the bolt numbering in Figure 5-5. Multiple 
passes had to be completed at increments of 30%, 60% and then 100% of the maximum 
torque for the joint.  
For this joint, the magnitudes of these increments are listed in Table 5-2. The last step 
is tightening the bolts in a clockwise pattern to verify that every bolt is at the maximum 
torque. The joint was tightened using a torque wrench.  
 
 
 
 
 
 
 
 
 
Figure 5-5  Bolt sequence. 
Table 5-2  Torque increments.  
30% of Maximum 236 N-m 
60% of Maximum 472 N-m 
100% of Maximum 786 N-m 
In order to tighten the joint, the two halves were first bolted together to a hand-tight 
condition on the ground, with the pipes supported by blocks of wood. Subsequently, a 
forklift was used to set them on two sawhorses after which the joint was tightened using 
the procedure outlined above. It may also be of some importance, considering the 
symmetry of the joint, that the bolts were always fastened so that the bolt heads lay on the 
same side as the impact locations A and S, as shown in Figure 5-6.  
 
 108 
 
The piezoceramic sensors used for the entirety of the experimental testing were of type 
PZT-5H, extracted from 72.4 × 72.4 mm sheets available from Piezo Systems Inc. 
(Cambridge, MA). The dimensions of the sensors are listed in  Table 5-3.    
  Table 5-3  Piezoceramic sensor dimensions   
Length 24.0 mm 
Width 10.0 mm 
Thickness 1.0 mm 
 
Due to the small width of the sensors, fitting them to the curvature of the pipe’s 
surface did not pose a problem. The material properties of the sensors (obtained from the 
manufacturer) are described using the following matrices. The density of PZT-5H is 
          . 
 
   
  
 
         
 
 
 
 
 
 
                 
                 
                 
         
         
          
 
 
 
 
 
 
 
  
 
 
          
        
        
              
      
 
                
      
      
      
  
where    is the compliance matrix,   is the piezoelectric coupling matrix, and  
  is 
the permittivity of the piezoceramic. The piezoceramic sheets were polarized through the 
thickness allowing the sensors to be created by soldering two electrodes to the top and 
bottom surfaces of each piece cut from the piezoceramic sheet. However, soldering 
caused each sensor to have an uneven surface which prevented complete contact between 
the pipe surface and the sensor. Consequently, a steel shim of size 25.0 × 10.0 × 0.5 mm 
was used as an adapter. Each adapter plate had a small semi-circular hole cut at the edge 
 109 
 
to fit around the solder. Placing the adapter plate between the sensor and the pipe surface 
allowed for increased contact.  
Before a sensor was attached, the surface of the pipe was scoured using silicon-
carbide paper and the pipe surface, sensor and adapter plate were wiped clean of dust and 
oil with isopropyl alcohol. The preparation, resulting in clean and rough surfaces, 
permitted the achievement of a stronger bond. The piezoceramic sensors were 
subsequently adhered to the adapters and the adapters to the pipe using a two part epoxy, 
Araldite 2011 (Huntsman Advanced Materials Americas Inc., Los Angeles, CA). The 
sensors were secured to the pipe by using masking tape, and left to cure for twelve hours 
to create a strong bond. Figure 5-6 (a) shows a typical sensor attached to the pipe.  
 
a) b) 
Figure 5-6  a) Typical bonded piezoceramic sensor b) Impact locations. 
The set-up for the completion of the damage detection was accomplished as follows, 
as shown in Figure 5-7. Two impact locations were chosen for testing, Locations A and 
S, respectively situated on the apex of the pipe and the side of the pipe (see Figure 5-6 
(b)). In order to ensure consistency of the impact location, a steel ball measuring 12.7 mm 
in diameter was attached at each of the impact locations to provide a target. The impact 
balls were attached to the pipe using the same epoxy as the sensors. These impact 
locations were chosen in an effort to determine the differences, if any, between impacts 
around the circumference of the pipe.  
  S_  
  A_  
 110 
 
Six piezoceramic sensors were attached to the pipe and flange. The placement and 
numbering of the sensors is illustrated in a cross-section of the pipe in Figure 5-8. 
 
 
 
 
 
Figure 5-7  Experimental set-up. 
 
 
 
 
 
 
 
 
 
Figure 5-8  PZT arrangement. 
To monitor the flange vibrations, four piezoceramic sensors, PZT3 through PZT6, 
were placed on one flange face at the locations of 12, 3, 6 and 9 o’clock.  An additional 
sensor, PZT2, was placed on top of the flange, directly above PZT3. To monitor the pipe 
vibrations initially, the remaining piezoceramic sensor, PZT1, was placed on the pipe at a 
distance of 250 mm measured from the center of the sensor to the outside of the closest 
flange (Figure 5-7). This sensor set-up was utilized for only the first damage scenario as 
All dimensions in millimeters  
  PZT3 
  PZT2 
 PZT4 
 PZT5 
  PZT6 
   S 
  A 
   PZT1 
 PZT1 PZT2 
 250 
      760 
    2045      2042 
 111 
 
described in Table 5-4. Due to the low amplitudes of the resulting EMD-EDI, the 
placement of PZT1 was altered for the remaining three damage scenarios. PZT1 was 
consequently placed directly at the base of the weld connecting the flange and pipe as 
illustrated in Figure 5-7 and Figure 5-9.  
The placement and numbering of all the sensors is illustrated in Figure 5-9. The 
sensor locations were chosen so that a comparison might be made between those results 
from on the pipe and on the flange, as well as between the different sensors located on the 
flange face.   
 
Figure 5-9  Piezoceramic arrangement. 
5.7   EXPERIMENTAL TESTING & ANALYSIS 
To create free vibration in the system, the pipe was excited using a piezoelectric 
impulse hammer (model 5800B5, Dytran Instrument Inc., Chatsworth, CA) outfitted with 
an aluminum tip, which can excite frequencies up to 5 kHz. The hammer outputs an 
 PZT2_  
 PZT3_  
 PZT4_  
 PZT1.  
(original location)_  
 PZT1.  
(reused location)_  
 PZT6_  
 PZT5_  
 112 
 
analog voltage which is representative of the input impulse. A typical hammer impulse 
signal is depicted in Figure 5-10 (a).  
 
      a) 
 
      b) 
Figure 5-10  a) Typical impulse hammer signal b) Typical sensor response. 
During vibration, the response of the impulse hammer and the sensors were 
simultaneously collected through two NI 9215 modules mounted on an NI cDAQ-9172 
data acquisition chassis, manufactured by National Instruments Inc. (Austin, Texas, 
USA). With only eight data acquisition channels available, it was only possible to 
monitor the response of seven sensors concurrently, since the hammer signal was also 
required for analysis and required the eighth connector. All of the data for the 
experimental testing was recorded using a sampling frequency of 10 kHz. A typical 
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
-0.2 0.0 0.2 0.4
A
m
p
li
tu
d
e 
(V
o
lt
)
Time (10-3 s)
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
0.0 0.2 0.4 0.6 0.8 1.0
A
m
p
li
tu
d
e 
(V
o
lt
)
Time (s)
 113 
 
sensor response is depicted in Figure 5-10 (b). The data acquisition system is illustrated 
in Figure 5-11.  
  
Figure 5-11  The data acquisition hardware. 
Once the data was collected using the LABView software, the data for each impact 
was stored in an ASCII text file, which included the responses of all the sensors and the 
impulse hammer. Subsequently, the recorded signals from the healthy and damaged pipe 
joint were processed using a Matlab code, developed in house, which performed the 
Empirical Mode Decomposition.  
This code was programmed to read the text file produced by the DAQ system and then 
process the response of all sensors simultaneously. The code removes unwanted data 
before impact, filters signal noise, compensates for any offset, normalizes the response of 
any sensor with respect to the hammer input force, applies a low-pass filter in order to 
have the first natural frequency in the filtered data, decomposes the signals based on the 
EMD method, derives the IMFs, and finally calculates the energy of the desired IMF for 
each sensor (EMD energy) (Rezaei and Taheri, 2009). 
For each excitation, the vibration lasted for seconds before decaying to zero (see 
Figure 5-10 (b)). Because the EMD-EDI is based on the comparison of special features in 
the PZT signals before and after damage creation, there is no limitation for the choice of 
 Impulse Hammer. 
 Laptop with DAQ software. 
 Power Supply. 
 DAQ system.  
 114 
 
signal duration as long as the same duration was used to evaluate both the healthy and 
damaged state signals (Rezaei and Taheri, 2009). With this in mind, an interval of 0.5 
seconds was chosen for the signal processing.  
The testing procedure used for the experiments followed the methodology outlined in 
Figure 5-2. Four different damage scenarios were completed, each having a healthy case 
and multiple damage cases. Table 5-4 lists the scenarios involving the healthy cases and 
damage cases investigated for each of the four scenarios.  
  Table 5-4  Damaged joint scenarios. 
Scenario State Bolts Loosened Damage Case 
Scenario 1 
Healthy Pipe  HC 1-1 
Damage Case 1-1 Bolt #5 DC 1-1 
Damage Case 1-2 Bolt #5 and Bolt #1 DC 1-2 
Damage Case 1-3 Bolts #5, #1 and #12 DC 1-3 
Damage Case 1-4 Bolts #5, #1, #12 and #9 DC 1-4 
 Damage Case 1-5 Bolts #5, #1, #12, #9 and #8 DC 1-5 
Scenario 2 
Healthy Pipe   HC 2-1 
Damage Case 2-6 Bolt #2 DC 2-6 
Damage Case 2-7 Bolt #2 and Bolt #6 DC 2-7 
Damage Case 2-8 Bolts#2, #6 and #10 DC 2-8 
Scenario 3 
Healthy Pipe   HC 3-1 
Damage Case 3-9 Bolt #3 DC 3-9 
Damage Case 3-10 Bolt #3 and Bolt #7 DC 3-10 
Damage Case 3-11 Bolts#3, #7 and #9 DC 3-11 
Scenario 4 
Healthy Pipe   HC 4-1 
Damage Case 4-12 Bolt #11 DC 4-12 
Damage Case 4-13 Bolt #11 and Bolt #12 DC 4-13 
Damage Case 4-14 Bolts#11, #12 and #10 DC 4-14 
Damage Case 4-15 Bolts #11, #12, #10 and #9 DC 4-15 
Damage Case 4-16 Bolts #11, #12, #10, #9 and #6 DC 4-16 
Damage Case 4-17 Bolts #11, #12, #10, #9,  #6 and #2 DC 4-17 
 
 115 
 
For each scenario, once the vibration of the healthy system was collected and 
processed through the EMD, damage was introduced to the system and the vibration of 
the pipe was again monitored, collected and processed. Testing was halted when leakage 
occurred in the joint; therefore for each scenario, leakage occurred during the last damage 
case. Figure 5-12 is a schematic depicting the bolts included in the damage cases for each 
scenario.  
 
 Scenario 1 Scenario 2 
 
 Scenario 3 Scenario 4 
 
Figure 5-12  Damage scenarios. 
In order to achieve consistent results, five impacts/data collections were completed 
for each of the two impact locations for each test case. Furthermore, each of the acquired 
 116 
 
signals from the sensors was normalized based on the input (hammer) signal. The 
difference between the signals after normalization was considered to be a good measure 
of the repeatability of the impact procedure (Rezaei and Taheri, 2009).  
To normalize the signals, the following procedure was followed (Rezaei and Taheri, 
2009): The fast Fourier transform of each sensor response was divided by the fast Fourier 
transform of the hammer signal. Afterwards, the result was transferred to the time domain 
by the inverse fast Fourier Transform. Once the five impacts for each test case were 
processed and the IMF energies attained, their average was calculated to use for the EMD 
energy.    
Once the EMD energy of both the healthy and damages cases was calculated, the 
EMD-EDI was calculated for each damage case.  
5.8   RESULTS AND DISCUSSION  
Cheraghi and Taheri (2007) suggested applying a band-pass filter to the data to keep 
only the first frequency component. However, in the experimental investigations carried 
out by Rezaei and Taheri (2009), it was concluded that this method only works well for 
systems where the first frequency is the dominant vibrating frequency of the structure. 
Otherwise, more frequency components should be included within the band-pass filter in 
order to avoid the elimination of damage sensitive frequency components.  
By analyzing the data using different frequency bands, it was found that a low-pass 
filter with a pass-band frequency of 2000 Hz was one of the frequency ranges that 
provided the best resolution for identification of damage. The following graphs, (Figures 
5-13 through 5-20) use the added EMD-EDI values of IMF1 and IMF2. The added IMFs 
gave better resolution than either IMF by itself. The other frequency ranges that yielded 
similar results are 0-1500 Hz and 500-2000 Hz for scenario 1, 0-1500 Hz and 500-2500 
Hz for scenario 2, 0-1500 Hz and 1000-2500 Hz for scenario 3 and 0-2500 Hz and 500-
2500 Hz for scenario 4. These pass-bands were chosen based on exhibition of damage 
progression as well as consideration of energy amplitude. 
 
 
 
 117 
 
Scenario 1 
 
Figure 5-13  Damage scenario 1, 0-2000 Hz, impact location A, added IMFs. 
 
Figure 5-14  Damage scenario 1, 0-2000 Hz, impact location S, added IMFs. 
 
Scenario 2 
 
 
Figure 5-15  Damage scenario 2, 0-2000 Hz, impact location A, added IMFs. 
0
30
60
90
120
150
180
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
0
50
100
150
200
250
300
350
400
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
0
20
40
60
80
100
120
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 2-6
DC 2-7
DC 2-8
 118 
 
 
Figure 5-16  Damage scenario 2, 0-2000 Hz, impact location S, added IMFs. 
 
Scenario 3 
 
Figure 5-17  Damage scenario 3, 0-2000 Hz, impact location A, added IMFs. 
 
Figure 5-18  Damage scenario 3, 0-2000 Hz, impact location S, added IMFs. 
 
 
 
0
20
40
60
80
100
120
140
160
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 2-6
DC 2-7
DC 2-8
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
 119 
 
Scenario 4 
 
Figure 5-19  Damage scenario 4, 0-2000 Hz, impact location A, added IMFs. 
 
 
Figure 5-20  Damage scenario 4, 0-2000 Hz, impact location S, added IMFs. 
 
From the above graphs, one can observe that the damage indices for impact location S, 
on the side of the pipe, are significantly greater for damage scenarios 1 and 2 which are 
located respectively on the top and bottom of the joint.  Furthermore, for scenario 1 and 
4, both impact locations demonstrate similar damage progression detection whereas for 
scenarios 2 and 3, the results from impacting the pipe at Location A deliver improved 
damage progression detection.  
5.9   CONCLUSIONS  
 The objective of this study was to experimentally confirm the use of the EMD-EDI 
for the evaluation of the damaged condition of a pressurized steel pipeline with a 12-bolt 
flanged joint. The EMD-EDI is based on the theory that IMFs, decomposed from a 
vibration signal, contain damage sensitive features. The damage sensitive feature used by 
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
 120 
 
the EMD-EDI is the variation in the EMD energy of an IMF prior to and following the 
occurrence of damage in a system. Four damage scenarios were analyzed, with each 
damage scenario progressing from a healthy state to having multiple bolts completely 
loose.  
This experimental study demonstrated that the EMD-EDI can successfully detect the 
presence of damage in bolted pipeline joints. The results indicated inconsistency in the 
progression of damage detection. However, it is believed that at least some of this 
inconsistency is due to the variability inherent in using a hand held impulse hammer. 
Further work is currently being conducted using an electric hammer which provides 
consistent impacts. It is also important to note that sensor placement and impact locations 
have a significant influence on the results. As such, further work is required to understand 
the most efficient location choices. Furthermore, it is important to note that the band-pass 
filter used for signal filtering should be applied carefully to avoid filtering out frequency 
components that contain meaningful information. 
5.10 ACKNOWLEDGEMENTS  
The authors gratefully acknowledge the financial support of the Natural Sciences and 
Engineering Council of Canada (NSERC) and Petroleum Research Atlantic Canada 
(PRAC). 
 
 
  
 121 
 
5.11 REFERENCES  
 ANSI/ASME. (1985). ANSI / ASME B36.19M-1985 - american national standard for 
welded and seamless wrought steel pipe ANSI/ASME.  
Carden, E., & Fanning, P. (2004). Vibration based condition monitoring: A review. 
Structural Health Monitoring, 3(4), 355-377.   
Cheraghi, N., Riley, M. J., & & Taheri, F. (2005). A novel approach for detection of 
damage in adhesively bonded joints in plastic pipes based on vibration methods 
using piezoelectric sensors. In Proceedings of the IEEE International Conference on 
Systems, Man and Cybernetics, Hawaii, USA, 10-12 Oct., pp. 3472-3478.  
Cheraghi, N., & Taheri, F. (2008). Application of the empirical mode decomposition for 
system identification and structural health monitoring. International J. of Applied 
Mathematics and Engineering Sciences, 2(1), 61-72.  
Cheraghi, N., & Taheri, F. (2007). A damage index for structural health monitoring based 
on the empirical mode decomposition. J. of Mechanics of Materials and Structures, 
2(1), 43-62.  
Corbett, R. (1998). Stretch control. In Handbook of bolts and bolted joints, J. H. 
Bickford, & S. Nassar (Eds.), (pp. 571). Marcel Dekker, Inc.: New York.  
Doebling, S. W., Farrar, C. R., Prime, M. B., & Shevitz, D. W. (1996). Damage 
identification and health monitoring of structural and mechanical systems from 
changes in their vibration characteristics: A literature review. Report No. LA-
13070-MS, Los Alamos National Laboratory. 
Doebling, S. W., Farrar, C. R., & Prime, M. B. (1998). A summary review of vibration-
based damage identification methods. The Shock and Vibration Digest, 30(2), 91-
105.  
Energy Institute (Great Britain). (2007). Guidelines for the management of the integrity of 
bolted joints for pressurised systems. (2nd ed.). Energy Institute: London.  
Fasel, T., Kennel, M., Todd, M., Clayton, E., & Park, G. (2009). Damage state evaluation 
of experimental and simulated bolted joints using chaotic ultrasonic waves. Smart 
Structures and Systems, 5(4), 329-344.  
 122 
 
Hess, D. P. (1998). Vibration and shock induced loosening. In Handbook of bolts and 
bolted joints, J. H. Bickford, & S. Nassar (Eds.), (pp. 757). Marcel Dekker, Inc.: 
New York.  
Huang, N. E., Shen, Z., Long, S. R., Wu, M. C., Shih, H. H., Zheng, Q., Nai-Chyuan, Y., 
Tung, C. C., & Liu, H. H. (1998). The empirical mode decomposition and the hilbert 
spectrum for nonlinear and non-stationary time series analysis. Philosophical 
Transactions of the Royal Society A: Mathematical, Physical and Engineering 
Sciences, 454(1971), 903-995.  
Industrial Fasteners Institute: Division VI Aerospace Fasteners. (1995). Test application 
handbook relating to mechanical fasteners. Industrial Fasteners Institute: Cleveland, 
Ohio.  
Johnson, E. A., Lam, H. F., Katafygiotis, L. S., & Beck, J. L. (2004). Phase I IASC-
ASCE structural health monitoring benchmark problem using simulated data. 
Journal of Engineering Mechanics, 130(1), 3-15.  
Loutridis, S. (2004). Damage detection in gear systems using empirical mode 
decomposition. Engineering Structures, 26(12), 1833-1841.  
Michaelides, P. G., & Fassois, S. D. (2008). Damage detection and localization in the 
joints of a thin plate via a statistical power spectral density based method. In 
Proceedings of the 4th European Workshop on Structural Health Monitoring, 
Krakow, Poland, July, 937-944.  
Nichols, J. M. (2003). Using state space predictive modeling with chaotic interrogation in 
detecting joint preload loss in a frame structure experiment. Smart Materials 
Structures, 12(4), 580-601.  
Pai, N. G., & Hess, D. P. (2002). Three-dimensional finite element analysis of threaded 
fastener loosening due to dynamic shear load. Engineering Failure Analysis, 9(4), 
383-402. Retrieved from http://dx.doi.org/10.1016/S1350-6307(01)00024-3  
Park, G., & Inman, D. J. (2005). Impedance-based structural health monitoring. In 
Damage prognosis for aerospace, civil and mechanical systems, D. J. Inman, C. R. 
Farrar, V. Lopes Jr. & V. Steffen Jr. (Eds.), (pp. 275). John Wiley and Sons, Ltd.: 
Great Britain.  
 123 
 
Park, G., Muntges, D. E., & Inman, D. J. (2003). Self-repairing joints employing shape-
memory alloy actuators. Journal of the Minerals, Metals and Materials Society, 
55(12), 33-37.  
Peairs, D. M., Park, G., & Inman, D. J. (2004). Improving accessibility of the impedance-
based structural health monitoring method. Journal of Intelligent Material Systems 
and Structures, 15(2), 129-140. Retrieved from 
http://dx.doi.org/10.1177/1045389X04039914  
Pines, D., & Salvino, L. (2006). Structural health monitoring using empirical mode 
decomposition and the hilbert phase. Journal of Sound and Vibration, 294(1-2), 97-
124.  
Rezaei, D., & Taheri, F. (2010). Health monitoring of pipeline girth weld using empirical 
mode decomposition. Smart Materials and Structures, 19(5). 
Rezaei, D., & Taheri, F. (2009). Experimental validation of a novel structural damage 
detection method based on empirical mode decomposition. Smart Materials and 
Structures, 18(4). 
Ritdumrongkul, S., Abe, M., Fujino, Y., & Miyashita, T. (2004). Quantitative health 
monitoring of bolted joints using a piezoceramic actuator-sensor. Smart Materials 
and Structures, 13(1), 20-29.  
Ritdumrongkul, S., & Fujino, Y. (2006). Identification of the location and level of 
damage in multiple-bolted-joint structures by PZT actuator-sensors. Journal of 
Structural Engineering, 132(2), 304-311.  
Shi, Z. (2000). Structural damage detection from modal strain energy change. Journal of 
Engineering Mechanics, 126(12), 1216-1223.  
Todd, M., Erickson, K., Chang, L., Lee, K., & Nichols, J. (2004). Using chaotic 
interrogation and attractor nonlinear cross-prediction error to detect fastener preload 
loss in an aluminum frame. Chaos, 14(2), 387-399. doi:10.1063/1.1688091  
Todd, M. D., Nichols, J. M., Nichols, C. J., & Virgin, L. N. (2004). An assessment of 
modal property effectiveness in detecting bolted joint degradation: Theory and 
experiment. Journal of Sound and Vibration, 275(3-5), 1113-1126. Retrieved from 
http://dx.doi.org/10.1016/j.jsv.2003.10.037  
 124 
 
Yan, Y., Cheng, L., Wu, Z., & Yam, L. (2007). Development in vibration-based 
structural damage detection technique. Mechanical Systems and Signal Processing, 
21(5), 2198-2211.  
 
125 
 
CHAPTER 6 
DAMAGE DETECTION OF A PVC PIPELINE JOINT 
 
The following chapter details the experimental tests performed on two PVC pipes 
jointed together using an 8-bolt flanged joint. Two boundary conditions were 
investigated: free-free and cantilevered. An overview of the experimental set-up is shown 
in Figure 6-1. The objective of this study was to validate the effectiveness of using the 
EMD-EDI to detect various degrees of joint damage with a type of material able to 
provide a greater damping effect than steel. It was also decided to explore the effect of 
different boundary conditions on the results. Varying degrees of joint damage were 
simulated by removing bolts.  
 
a) b) 
Figure 6-1  Overview of experimental set-up a) Free-free boundary condition b) Cantilevered 
boundary condition. 
  
126 
 
6.1 EXPERIMENTAL SET-UP 
The test specimen used in this study was a  PVC pipe manufactured by IPEX (Ipex 
Inc., 2010) commonly used for applications such as industrial and process piping, pulp 
and paper, food processing, water and sewage treatment, and irrigation. The 
specifications of the 150 mm nominal pipe are listed in Table 6-1. A total pipe length of 
4.6 m was chosen so as to conform to the previous experimental research performed thus 
far. 
Table 6-1  PVC pipe dimensions and material properties.  
Length  4.6 m Density 1420 kg/m
3
 
Outer Diameter 168.3 mm E  2.758  GPa 
Wall Thickness 7.1 mm v 0.38  
    
Flat faced, schedule 80 IPEX Vanstone socket PVC flanges were chosen in 
conjunction with full faced neoprene gaskets. Correspondingly the gaskets had an NPS of 
6 and were 1/8 inches thick. SAE Grade 5 UNC hex head bolts were used (19.0 mm by 
102 mm) with corresponding sized nuts, and flat washers. The fasteners used were the 
same as the fasteners used for the experiments on the 8-bolt joint in Chapter 3 except that 
no spring washers were used. To create the mechanical joint in the pipe, the pipe was cut 
in two approximately equal pieces of 2315 mm (left half) and 2291 mm (right half) 
(depicted in Figure 6-5 and 6-6). Subsequently, the two flanges were attached to the pipe 
using solvent welding using IPEX primer (Xirtec 7) and PVC cement (Xirtec 11).  
Concerning the boundary conditions, two different boundary conditions were tested 
for the two damage scenarios detailed in the testing section. First, the pipe was suspended 
from saw horses using a nylon rope at either end, as shown in Figure 6-2. The nylon 
ropes were positioned at the mid-point along the beam of the saw horse. The second 
boundary condition consisted of the pipe being cantilevered off the edge of a platform, 
shown in Figure 6-3.  
127 
 
 
Figure 6-2 Free-free boundary condition. 
 
 
Figure 6-3  Cantilevered boundary condition. 
For this specific joint configuration, the maximum bolt torque was determined to be 
67.8 N-m. This value was given in the manual of the joint and pipe manufacturer (Ipex 
Inc., 2010). Before bolting, the bolts and nut faces were lubricated using the Anti-seize 
lubricant manufactured by Permatex (Milton, Ontario).  
The procedure to tighten the bolts was the industry standard, which follows a criss-
cross pattern across the face of the flange following the bolt numbering in Figure 6-4. 
Multiple passes should be completed at increments of 30%, 60% and then 100% of the 
128 
 
maximum torque for the joint. For this joint, the magnitudes of these increments are listed 
in Table 6-2. The last step to tightening the joint is to go around the joint clockwise and 
verify that every bolt is at the maximum torque. The joint was tightened using a torque 
wrench. 
 
 
 
 
 
 
 
 
 
Figure 6-4  Bolt numbering. 
 
Table 6-2  Torque increments.  
30% of Maximum 20.3 N-m 
60% of Maximum 40.7 N-m 
100% of Maximum 67.8 N-m 
 
It may also be of some importance, considering the symmetry of the joint, that the 
bolts were always fastened so that the bolt heads lay on the same side as the impact 
location. 
The piezoceramic sensors used to monitor the dynamic response of the systems were 
of the same type and dimensions as used in the previous experiments. The dimensions of 
the sensors are listed in Table 3-3.  
The set-up for the completion of the damage detection was accomplished as follows 
and may be seen in Figure 6-5 and Figure 6-6. Only one impact location was chosen for 
testing, located 280 mm from the edge of the flange socket. In order to maintain 
consistency between impacts, a steel ball measuring 12.7 mm in diameter was placed at 
the impact location to provide a consistent impact target. The impact ball, illustrated in 
Figure 6-7, was attached to the pipe using the same epoxy as for the sensors.  
6 
4 
8 2 
1 
5 
3 
7 
129 
 
 
 
 
 
 
 
 
 
 
Figure 6-5  Free-free experimental set-up.  
 
 
 
 
 
 
 
 
Figure 6-6  Cantilevered experimental set-up.  
 
Figure 6-7  Impact location. 
To monitor the system vibrations, seven new piezoceramic sensors were created and 
attached to the pipe and flange. Four piezoceramic sensors, PZT1 through PZT4, were 
placed on the flange socket at 0°, 90°, 180° and 270°. The sensors were placed at a 
 
 
  
 
 
 
 
Impact Location 
280 mm 
2337  mm 2312 mm 
Piezoceramic sensors 
  515 mm 
  75 mm 
280 mm 
Impact Location 
2337 mm 2312 mm 
   Piezoceramic sensors 
130 
 
PZT6 
PZT7   PZT5 
  PZT1 
  PZT2 
  PZT3 
  PZT4 
distance of 12.7 mm from the socket edge to the edge of the sensor. The three remaining 
sensors were placed on top of the flange, directly above PZT1, PZT2 and PZT4. The 
placement of the sensors along the pipe is shown in Figure 6-5 Figure 6-6. The placement 
and numbering of the sensors is illustrated in a cross-section of the pipe in Figure 6-8, 
and in a photograph in Figure 6-9. The sensor locations were chosen so that a comparison 
might be made between those results from on the pipe and on the flange, as well between 
the results from different areas around the flange edge.   
 
 
 
 
 
 
 
 
Figure 6-8  PZT sensor placement and numbering. 
 
 
Figure 6-9  Sensor placement. 
 
131 
 
6.2 TESTING 
The experimental set-up used the same data acquisition hardware and software as 
described in Chapter 5. A typical hammer impulse signal is depicted in Figure 6-10 (a). A 
typical sensor response is depicted in Figure 6-10 (b).   
  
a)                                                                        b)  
Figure 6-10  a) Typical impulse hammer signal b) Typical sensor response. 
Two different damage scenarios were completed, each having a healthy case and 
multiple damage cases. Each of these damage scenarios was completed for the two 
different boundary conditions. The aluminum hammer tip was used for all scenarios. 
Seven impacts were completed for each impact location for all cases. According to the 
earlier experiments, seven impacts were chosen due to the probability being low that 
additional impacts would be necessary due to a lack of impact consistency. A list of all 
test cases considered is shown in Table 6-3.  
    Figure 6-11 is a schematic depicting the bolts included in the damage scenarios 
listed above. For each damage case, the indicated bolt was completely removed from the 
joint. An additional healthy case (HC 3-1) was also completed for each boundary 
condition for comparison purposes to investigate healthy state consistency.  
 
 
 
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
-0.2 0.0 0.2 0.4 0.6 0.8 1.0
A
m
p
li
tu
d
e 
(V
o
lt
)
Time (10-3s)
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
0.00 0.05 0.10
A
m
p
li
tu
d
e 
(V
o
lt
)
Time (s)
132 
 
    Table 6-3  Damaged joint scenarios. 
Scenario State Bolts Removed Damage Case 
Scenario 1 
Healthy Pipe 1  HC 1-1 
Damage Case 1-1 Bolt #2 DC 1-1 
Damage Case 1-2 Bolt #2 and Bolt #8 DC 1-2 
Damage Case 1-3 Bolts #2, #8 and #5 DC 1-3 
Damage Case 1-4 Bolts #2, #8, #5 and #4 DC 1-4 
Scenario 2 
Healthy Pipe 2  HC 2-1 
Damage Case 2-5 Bolt #3 DC 2-5 
Damage Case 2-6 Bolt #3 and Bolt #8 DC 2-6 
Damage Case 2-7 Bolts#3, #8 and #5 DC 2-7 
Damage Case 2-8 Bolts #3, #8, #5 and #6 DC 2-8 
 
 
 
 
 
 
 
 
a)                                                              b) 
Figure 6-11  Joint damage scenarios a) Scenario 1 b) Scenario 2.   
6.3 DATA ANALYSIS  
The data analysis was completed in the same manner as for the experiments 
documented in Chapters 1 and 2. For all test cases, five impact signals were chosen for 
the calculation of the EMD average energy. 
B5 
B8 
B6 B3 
  PZT1 
  PZT2 
  PZT3 
  PZT4 
PZT6 
PZT7 
   PZT5 
B5 
B8 
B4 
   B2 
  PZT1 
  PZT2 
  PZT3 
  PZT4 
PZT6 
PZT7 
 PZT5 
133 
 
6.4 RESULTS 
By analyzing the data using different frequency bands, it was found that a low-pass 
filter with a pass-band frequency of 2000 Hz provided the best resolution for the 
identification of damage for the free-free boundary condition. For the cantilevered 
boundary condition, it was determined that a low-pass filter with a pass-band frequency 
of 1500 Hz provided the best resolution for damage identification. These pass-bands were 
chosen based on the exhibition of damage progression, as well as consideration of the 
energy amplitude. However, these two pass-bands only marginally improved the EMD-
EDI with respect to other pass-bands. It was also determined that the first IMF 
demonstrated the best damage resolution overall.  
Figure 6-12 and Figure 6-13 illustrate the EMD-EDI calculated respectively for 
damage scenario 1 and 2 for the free-free boundary condition. Likewise, Figure 6-14 and 
Figure 6-15 illustrate the EMD-EDI calculated respectively for damage scenario 1 and 2 
for the cantilevered boundary condition. 
Boundary Condition: Free-Free 
 
Figure 6-12  Damage scenario 1, 0-2000 Hz, IMF1. 
 
0
50
100
150
200
250
300
1 2 3 4 5 6 7
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
134 
 
 
Figure 6-13  Damage scenario 2, 0-2000 Hz, IMF1. 
Boundary Condition: Cantilevered  
 
Figure 6-14  Damage scenario 1, 0-1500 Hz, IMF1. 
 
Figure 6-15  Damage scenario 2, 0-1500 Hz, IMF1. 
6.5 CONSISTENCY 
The same process was used for these experiments as explained in Chapter 3 to 
measure result consistency between impacts. In order to document the range of 
variability, Tables C-1 and C-2 in Appendix C list the variations for each healthy and 
0
50
100
150
200
250
300
350
1 2 3 4 5 6 7
E
M
D
-E
D
I
PZT Sensor
DC 2-5
DC 2-6
DC 2-7
DC 2-8
0
50
100
150
200
250
300
1 2 3 4 5 6 7
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
0
50
100
150
200
250
300
1 2 3 4 5 6 7
E
M
D
-E
D
I
PZT Sensor
DC 2-5
DC 2-6
DC 2-7
DC 2-8
135 
 
damage case. Table C-1 lists the values for the free-free boundary condition data and 
Table C-2 lists the values for the cantilevered boundary condition data. As seen, the EMD 
energies between different impacts were not consistent. During the experiments, it was 
extremely difficult to achieve impact consistency using the impulse hammer on the PVC 
pipe. This is in direct contrast to the earlier experimental work conducted on the steel 
pipes for which the impacts were much more consistent and the maximum difference 
remained under a value of 20%. The variation in consistency helps to explain the poor 
index results.  
6.6 SET-UP RELIABILITY 
In order to address the question of set-up reliability, three healthy cases were 
completed for each boundary condition as mentioned earlier. Eqn. 3-3 was used to 
calculate the variability between healthy states. The range of comparison values are listed 
in Table C-3 in Appendix C. It may be concluded from the values in the Table that it is 
very difficult to achieve consistency between the set-ups. The difference in EMD 
energies varies significantly, attributing to the poor resulting damage progression.  
6.7 CONCLUSIONS 
From comparing Figure 6-12 through 6-15, one can observed that overall the sensors 
located on the flange socket provided lower amplitude indices than the sensors on the 
flange itself. Furthermore, it is clear that PZT 7 gave the clearest indication of damage 
progression as evidenced by Figure 6-13, 6-14 and 6-15. Concerning boundary 
conditions, it appears that the cantilevered boundary condition resulted in a better 
prediction of damage progression. Unfortunately, due to the large inconsistencies in the 
excitation induced by the impact hammer, as discussed in section 6.5, these conclusions 
may be inherently flawed.  
136 
 
CHAPTER 7 
CONCLUSION 
 
7.1  CONCLUSIONS 
The objective of this thesis was to explore the application of the Empirical Mode 
Decomposition Energy Damage Index, coupled with the use of piezoelectric sensors, to 
bolted pipeline joints. The EMD-EDI is a relatively new method developed for detecting 
the presence of damage in structures, including corrosion, cracks and support stiffness 
losses. The motive behind this research was to expand upon the previous work conducted 
using this method by focusing on a different type of structural damage than has been 
studied to date. 
The EMD-EDI is based on the detection of changes within the dynamic characteristics 
of a structure due to damage. In the developed procedure, the free vibration response of 
the structure is monitored and collected through piezoceramic sensors, before and after 
the occurrence of damage. Subsequently, the Empirical Mode Decomposition, the key 
component in the Hilbert-Huang Transform, is used to decompose the free vibration 
signatures of the structure into a collection of oscillatory modes called Intrinsic Mode 
Functions. The EMD-EDI is then calculated based on the energy change of the IMFs 
measured from the healthy-state baseline. Accordingly, the index serves as a reflection of 
the deviations in the structural integrity of the system, thereby providing a means to 
detect the presence of damage.  
To continue the effort focused on establishing the integrity and limitation of this novel 
damage detection method, a series of experimental testing on mechanically bolted joints 
was performed. Four different bolted joint systems were tested under different conditions 
including: a steel pipe with an 8-bolt flanged joint, a pressurized steel pipe with a 12-bolt 
flanged steel joint, an in-service steel pipe with a 12-bolt flanged joint and a PVC pipe 
with an 8-bolt flanged joint. The induced joint damage was a loss of clamping force 
137 
 
brought on through bolt loosening and/or bolt removal. Each of these experiments is 
briefly summarized below along with the appropriate conclusions.  
7.1.1 Damage Detection of a Steel Pipeline Bolted Joint 
Chapter 3 entailed the experimental testing performed on a simply supported steel 
pipe jointed at the mid-section using an 8-bolt flanged, gasketed joint. Varying degrees of 
joint damage were simulated by reducing the torque load on the bolts and/or removing 
them completely. Three different combinations of piezoceramic sensors and/or impact 
locations were explored in total.  
The first session situated six sensors around the circumference of the pipe, at 250 mm 
from the joint, and used four impact locations for comparison purposes. By analyzing the 
data using different frequency bands, it was found that a low-pass filter with a pass-band 
frequency of 2000 Hz provided the best resolution for identification of damage. However, 
the results were poor in the context of identifying conclusive damage progressions.  
The second session entailed removing the gasket from the first session and 
positioning new sensors in closer proximity to the joint. Two piezoceramic sensors were 
paced 25 mm from the flange, and four were placed directly on the flange face. In 
contrast to the last session, the frequency range of 0-2500 Hz yielded the best results in 
terms of damage index amplitude and damage progression with the excitation by an 
aluminium tip hammer yielding better results than the plastic tip.  
It was concluded that the aluminum tip yielded improved damage detection 
capabilities than the plastic hammer tip. Furthermore, it was concluded that due to the 
variance in results between the four impact locations, further work was required. One 
important observation was that the majority of the results from the second session did not 
show damage progression for the first 5 damage cases. Therefore, only fully loosened 
bolts were considered in the later experiments. Finally, the decision to perform testing 
without a gasket yielded improved results overall. A possible reason would be that the 
presence of the gasket, and the type of gasket, may play a large role in the usefulness of 
this method.   
 
138 
 
The third session entailed placing four new sensors on the flange face on the other 
side of the joint with respect to the sensor locations from session 2. Additionally, two 
new impact locations were chosen closer to the joint in an effort to determine if a closer 
impact would result in improved damage detection.  
The frequency range of 0-2000 Hz yielded the best results in terms of damage index 
amplitude and damage progression, with the added IMFs demonstrating the best 
resolution for damage detection.  From a comparison of the different impact locations, it 
was observed that Impact Locations B and C gave the best results concerning damage 
progression. Therefore, it can be concluded that an impact location too close to the joint 
may negatively affect the index results. Furthermore, the side of the flange on which the 
sensors are mounted relative to the impact location does not seem to make any significant 
difference in the results.   
Concerning the comparison between hammer tips, the two impulse hammer tips 
yielded approximately similar results overall with the plastic tip giving more consistent 
results concerning the variability in EMD energies among impacts. However, taking into 
account the results from session 2 that indicate a frequency bandwidth that surpasses 2 
kHz may be required for improved damage resolution, it was concluded that the 
aluminium tip was the best choice for future damage detection trials.  
7.1.2 Damage Detection of an In-Service Condensation Pipeline Joint 
This experiment was performed in order to apply the EMD-EDI in a practical 
application of an in-service pipeline. A 12-bolt mechanical joint located in the 
underground tunnels at the Central Services Building at Dalhousie University, NS, was 
chosen for testing due to ease of access. To capture the dynamic response of the joint, 
eight piezoceramic sensors were bonded to the flange and pipe in the vicinity of the joint. 
Three different damage scenarios were explored, each involving a healthy case and 
multiple damage cases each involving complete removal of one or more bolts.  
By analyzing the data using different frequency bands, it was found that a low-pass 
filter with a cutoff frequency of 2000 Hz provided the best resolution for identification of 
damage for each of the scenarios. The frequency band of 1000-2000 Hz also provided 
139 
 
good resolution for the second scenario. This choice was based on the evaluation of both 
IMFs. Several conclusions were made after the analysis of the results was completed.  
It was concluded that the EMD-EDI can successfully detect the presence of damage. 
However, sensor placement, boundary conditions and impact locations have a significant 
influence on the results which should be carefully considered when choosing sensor and 
impact location. Moreover, the experimental results did not clearly indicate a relationship 
between the ability of a sensor to detect damage and its proximity to the damage, 
therefore further work in this area is required. The final conclusion was that the band-
pass filter used for signal filtering should be applied carefully to avoid filtering out 
frequency components that contain meaningful information. 
7.1.3 Damage Detection of a Pressurized Steel Pipeline Joint 
Chapter 5 entailed experimental testing on a suspended steel pipe pressurized to 10 
MPa with a 12-bolt flanged joint at the mid-point. Five PZT sensors were placed on the 
flange face and rim, and one sensor was placed on the pipe. Two impact locations were 
used, situated on the top and side of the pipe. Four different damage scenarios were 
completed, each having a healthy case and multiple damage cases.  
By analyzing the data using different frequency bands, it was found that a low-pass 
filter with a pass-band frequency of 2000 Hz was one of the frequency ranges that 
provided the best resolution for identification of damage. The added IMFs gave better 
resolution than either IMF by itself. Further graphs supporting these conclusions may be 
found in Appendix B.  
However, the results indicated inconsistency in the detection of damage progression. It 
is believed that a portion of this inconsistency is due to the variability inherent in using a 
hand held impulse hammer. It was also concluded that sensor placement and impact 
locations have a significant influence on the results. As such, further work is required to 
understand the most efficient locations for impact and senor placement.  
  
140 
 
7.1.4 Damage Detection of a PVC Pipeline Joint 
Chapter 6 entailed experimental testing performed on a PVC pipe jointed at the mid-
section using an 8-bolt flanged joint. Two damage scenarios were investigated including 
a damage pattern concentrated on bottom of the joint, as well as a random pattern. The 
influence of boundary conditions on the results was also investigated using free-free and 
cantilevered boundary conditions. To monitor the vibration of the system, four sensors 
were placed on the flange socket, and three sensors were placed on the flange rim. For the 
various damage cases, joint damage was simulated by completely removing the bolts 
from the flanges.  
By analyzing the data using different frequency bands, it was found that a low-pass 
filter with a pass-band frequency of 2000 Hz provided the best resolution for the 
identification of damage for the suspended boundary condition. For the cantilevered 
boundary condition, it was determined that a low-pass filter with a pass-band frequency 
of 1500 Hz provided the best resolution for damage identification. 
The conclusions reached for this experiment were as follows. The sensors located on 
the flange socket provided lower amplitude damage indices than the sensors on the 
flange. Concerning boundary conditions, it appears that the cantilevered boundary 
condition resulted in a better prediction of damage progression. Unfortunately, due to the 
large inconsistencies in the excitations induced by the hand-held impact hammer, these 
conclusions may not be entirely valid.   
 
  
141 
 
7.2   RECOMMENDATIONS FOR FUTURE WORK  
Due to the presence of factors that caused problems during testing as well as 
inconsistencies in the results, below are some recommendations to improve upon the 
damage detection methodology of the EMD-EDI.  
The first recommendation is to use an appropriate device to accurately measure bolt 
tension for future experimental damage cases that involve reducing bolt tension instead of 
completely removing the bolt from the joint. Taking into account the interaction between 
tensioned bolts, when one bolt is loosened to any degree it would affect the tension in the 
other bolts in the joint. Therefore, it is important to use a device to measure bolt tension 
and accurately gauge the actual damage introduced to the joint, as opposed to a perceived 
degree of damage. Furthermore, measuring the bolt tension will also allow greater 
accuracy when initially torquing the joint, thus yielding a more clearly defined healthy 
state. Examples of methods to employ are direct tension indicators and ultrasonic 
extensometers.     
The implementation of a consistent excitation method is the second recommendation. 
The experimental work documented in this thesis required the use of a hand held impulse 
hammer. This method of excitation introduced inconsistency into the results. 
Additionally, in industrial applications, the use of an impulse hammer would be 
inefficient due to the time necessary for testing and the manpower required for personnel 
site visitation. Furthermore, depending on the location of the joint, manually exciting the 
pipe by hand may be impossible. Consequently, other excitation methods should be 
considered. For future experimental work, a consistent excitation method such as an 
electric hammer should improve results, especially in the case of the PVC pipeline.  
Another recommendation for future work is the development of a method, or a rule-of-
thumb, to determine the correct frequency range to use in the EMD-EDI. The band-pass 
filter used for signal filtering needs to be chosen carefully to avoid filtering out frequency 
components that contain meaningful information. However, it is inefficient to analyze the 
data for many frequency bands, and manually evaluate the results for the best frequency 
range. Accordingly, further insight into the relationship between the joint, damage type 
and the index will significantly decrease analysis time. Furthermore, the creation of a 
142 
 
Matlab code that would automatically determine the best frequency range to use would 
be extremely efficient. Additionally, the creation of an automated system in which the 
Matlab code would automatically apply the damage index upon the system acquiring 
damage data would be ideal.  
The fourth recommendation is that more in-situ testing should be conducted for further 
validation of the method in a variety of situations. The majority of the testing completed 
so far has been conducted in a controlled laboratory setting. In-situ testing involves many 
variables including environmental conditions, external loadings, and unknown factors 
that may negatively affect the application of the method by inducing uncontrolled 
dynamic loading. An example of this would be the influence of noise produced by a 
system and its interaction with the excitation vibration signals.   
The final recommendation is the use of an improved sensor system. It is a drawback to 
the current methodology that an inspector’s presence is required at the site of the damage. 
As mentioned above, in real applications this may not be possible. Therefore, a wireless 
sensor system would be ideal for providing damage detection in cases that have 
inaccessible structures.  
 
 
 
 
143 
 
REFERENCES 
 
Adeli, H., & Hung, S. (1995). Machine learning – neural networks, genetic algorithms, 
and fuzzy systems. John Wiley and Sons, Ltd., New York. 
ANSI/ASME. (1985). ANSI / ASME B36.19M-1985 - american national standard for 
welded and seamless wrought steel pipe ANSI/ASME.  
ASTM International. (2007). ASTM A53 - standard specification for pipe, steel, black 
and hot-dipped, zinc-coated, welded and seamless. West Conshohocken, PA: ASTM 
International.  
Bibel, G. (1998). Tightening groups of fasteners in a structure and the resulting elastic 
interaction. In Handbook of bolts and bolted joints, J. H. Bickford, & S. Nassar 
(Eds.), (pp. 451). Marcel Dekker, Inc.: New York.  
Bickford, J. H. (1995). An introduction to the design and behavior of bolted joints. 
Marcel Dekker, Inc.: New York.  
Bickford, J. H., & Nassar, S. (Eds.). (1998). Handbook of bolts and bolted joints. Marcel 
Dekker, Inc.: New York.  
Bickford, J. H. (1998a). Behavior of a bolted joint during assembly. In Handbook of bolts 
and bolted joints, J. H. Bickford, & S. Nassar (Eds.), (pp. 425). Marcel Dekker, Inc.: 
New York.  
Bickford, J. H. (1998b). Torque control of assembly. In Handbook of bolts and bolted 
joints, J. H. Bickford, & S. Nassar (Eds.), (pp. 559). Marcel Dekker, Inc.: New York.  
Brownjohn, J. M. W. (2007). Structural health monitoring of civil infrastructure. Phil. 
Trans. R. Soc. A, 365, 589-622.  
Caccese, V., Mewer, R., & Vel, S. (2004). Detection of bolt load loss in hybrid 
composite/metal bolted connections. Engineering Structures, 26(7), 895.  
144 
 
Carden, E., & Fanning, P. (2004). Vibration based condition monitoring: A review. 
Structural Health Monitoring, 3(4), 355-377.  
Chang, F. K. (1999). Structural health monitoring: 2000. Technomic Publishing 
Company, Lancaster, Pennsylvania.  
Chattopadhyay, A., Das, S., & Coelho, C. (2007). Damage diagnosis using a kernel-based 
method. Insight - Non-Destructive Testing and Condition Monitoring, 49(8), 451-
458.  
Cheraghi, N., Riley, M. J., & & Taheri, F. (2005). A novel approach for detection of 
damage in adhesively bonded joints in plastic pipes based on vibration methods 
using piezoelectric sensors. In Proceedings of the IEEE International Conference on 
Systems, Man and Cybernetics, Hawaii, USA, 10-12 Oct., pp. 3472-3478.  
Cheraghi, N., & Taheri, F. (2008). Application of the empirical mode decomposition for 
system identification and structural health monitoring. International J. of Applied 
Mathematics and Engineering Sciences, 2(1), 61-72.  
Cheraghi, N., & Taheri, F. (2007). A damage index for structural health monitoring based 
on the empirical mode decomposition. J. of Mechanics of Materials and Structures, 
2(1), 43-62.  
Cheraghi, N., Riley, M. J., & Taheri, F. (2007). Application of hilbert-huang transform 
for evaluation of vibration characteristics of plastic pipes using piezoelectric sensors. 
Structural Engineering Mechanics, 25(6), 653-674.  
Chopra, A. K. (2001). Dynamics of structures: Theory and applications to earthquake 
engineering (2nd ed.). Prentice-Hall, Inc.: Upper Saddle River, NJ.  
CINDE. (2010). Canadian institute for NonDestructive evaluation. Retrieved May 27,  
from http://www.cinde.ca/ndt.shtml  
Coelho, C. K., Das, S., Chattopadhyay, A., Papandreou-Suppappola, A., & Peralta, P. 
(2007). Detection of fatigue cracks and torque loss in bolted joints. In Proceedings 
of SPIE: Health Monitoring of Structural and Biological Systems 2007, California, 
USA, 19-22 March.  
145 
 
Corbett, R. (1998). Stretch control. In Handbook of bolts and bolted joints, J. H. 
Bickford, & S. Nassar (Eds.), (pp. 571). Marcel Dekker, Inc.: New York.  
Doebling, S. W., Farrar, C. R., Prime, M. B., & Shevitz, D. W. (1996). Damage 
identification and health monitoring of structural and mechanical systems from 
changes in their vibration characteristics: A literature review. Report No. LA-
13070-MS, Los Alamos National Laboratory. 
Doebling, S. W., Farrar, C. R., & Prime, M. B. (1998). A summary review of vibration-
based damage identification methods. The Shock and Vibration Digest, 30(2), 91-
105.   
Energy Institute (Great Britain). (2007). Guidelines for the management of the integrity of 
bolted joints for pressurised systems. (2nd ed.). Energy Institute: London.  
Farrar, R., Doeblíng, S. W., & Charles, D. (2001). Vibration-based structural damage 
identification. Philosophical Transactions of the Royal Society A: Mathematical, 
Physical and Engineering Sciences, 359(1778), 131-149.  
Farrar, C., & Worden, K. (2007). An introduction to structural health monitoring. 
Philosophical Transactions of the Royal Society A: Mathematical, Physical and 
Engineering Sciences, 365(1851), 303-315.  
Fasel, T., Kennel, M., Todd, M., Clayton, E., & Park, G. (2009). Damage state evaluation 
of experimental and simulated bolted joints using chaotic ultrasonic waves. Smart 
Structures and Systems, 5(4), 329-344.  
Fasel, T. R., Olson, C. C., Todd, M. D., & Kundu, T. (2008). Optimized guided wave 
excitations for health monitoring of a bolted joint. In Proceedings of SPIE: Health 
Monitoring of Structural and Biological Systems 2008, California, USA, 10-13 
March.  
Hess, D. P. (1998). Vibration and shock induced loosening. In Handbook of bolts and 
bolted joints, J. H. Bickford, & S. Nassar (Eds.), (pp. 757). Marcel Dekker, Inc.: 
New York.  
 
146 
 
Huang, N. E., Shen, Z., Long, S. R., Wu, M. C., Shih, H. H., Zheng, Q., Nai-Chyuan, Y., 
Tung, C. C., & Liu, H. H. (1998). The empirical mode decomposition and the hilbert 
spectrum for nonlinear and non-stationary time series analysis. Philosophical 
Transactions of the Royal Society A: Mathematical, Physical and Engineering 
Sciences, 454(1971), 903-995.  
Industrial Fasteners Institute: Division VI Aerospace Fasteners. (1995). Test application 
handbook relating to mechanical fasteners. Industrial Fasteners Institute: Cleveland, 
Ohio.  
Ipex Inc. (2010). Industrial products, process piping systems, xirtec PVC & corzan 
CPVC. Retrieved March, 2010, from http://www.ipexinc.com/  
Jiang, Y., Zhang, M., & Lee, C. (2003). A study of early stage self-loosening of bolted 
joints. Journal of Mechanical Design, 125(3), 518.  
Johnson, E. A., Lam, H. F., Katafygiotis, L. S., & Beck, J. L. (2004). Phase I IASC-
ASCE structural health monitoring benchmark problem using simulated data. 
Journal of Engineering Mechanics, 130(1), 3-15.  
Junker, G. (1972). Criteria for self-loosening of fasteners under vibration. Aircraft 
Engineering and Aerospace Technology, 44(10), 14.  
Kitaura, M., & Miyajima, M. (1996). Damage to water supply pipelines. Soils and 
Foundations, January, 325-333.  
Krawczuk, M. (2002). Application of spectral beam finite element with a crack and 
iterative search technique for damage detection. Finite Elements in Analysis and 
Design, 38(6), 537-548.  
Kull, F. (1998). Bolt fatigue. In Handbook of bolts and bolted joints, J. H. Bickford, & S. 
Nassar (Eds.), (pp. 699). Marcel Dekker, Inc.: New York.  
Lee, L. S., Karbhari, V. M., & Sikorsky, C. (2004). Investigation of integrity and 
effectiveness of RC bridge deck rehabilitation with CFRP composites. No. SSRP-
2004/08: University of California., Retrieved from 
http://svctwww1.dot.ca.gov/hq/esc/earthquake_engineering/Research/SSRP_200.pdf  
147 
 
Leo, D. J. (2007). Engineering analysis of smart material systems. John Wiley and Sons, 
Ltd.: New Jersey.  
Lin, S., Yang, J. N., & Zhou, L. (2005). Damage identification of a benchmark building 
for structural health monitoring. Smart Materials & Structures, 14(3), S162-S169.  
Liu, J., Wang, X., Yuan, S., & & Li, G. (2006). On hilbert-huang transform approach for 
structural health monitoring. Journal of Intelligent Material Systems and Structures, 
17(8-9), 721-728.  
Lopes JR, V., Park, G., Cudney, H., & Inman, D. (2000). Impedance-based structural 
health monitoring with artificial neural networks. Journal of Intelligent Material 
Systems and Structures, 11(3), 206-214.  
Loutridis, S. (2004). Damage detection in gear systems using empirical mode 
decomposition. Engineering Structures, 26(12), 1833-1841.  
Lovell, P. A., & Pines, D. J. (1998). Damage assessment in a bolted lap joint. In 
Proceedings SPIE: Smart Structures and Materials 1998: Smart Systems for 
Bridges, Structures, and Highways, California, USA, 4-5 March, 112-126.  
Mascarenas, D. L., Park, G., & Farrar, C. R. (2005). Monitoring of bolt preload using 
piezoelectric active devices. In Proceedings of SPIE: Health Monitoring and Smart 
Nondestructive Evaluation of Structural and Biological Systems IV, California, USA, 
7-9 March, 129-136.  
Melhem, H.G. & Nagaraja, S. (1996). Machine learning and its application to civil 
engineering systems. Civil Engineering and Environmental Systems, 13(4), 259-279. 
Metzger, D. (2004). Bolted joint integrity management: Implementing industry best 
practice. In Proceedings of the 12th International Conference on Nuclear 
Engineering, Virginia, USA, 25-29 April, 179-184.  
Michaelides, P. G., & Fassois, S. D. (2008). Damage detection and localization in the 
joints of a thin plate via a statistical power spectral density based method. In 
Proceedings of the 4th European Workshop on Structural Health Monitoring, 
Krakow, Poland, July, 937-944.  
148 
 
Milanese, A., Marzocca, P., Nichols, J., Seaver, M., & Trickey, S. (2008). Modeling and 
detection of joint loosening using output-only broad-band vibration data. Structural 
Health Monitoring, 7(4), 309.  
Moniz, L., Nichols, J., Nichols, C., Seaver, M., Trickey, S., & Todd, M. (2005). A 
multivariate, attractor-based approach to structural health monitoring. Journal of 
Sound and Vibration, 283(1-2), 295-310.  
Nichols, J. M. (2003). Using state space predictive modeling with chaotic interrogation in 
detecting joint preload loss in a frame structure experiment. Smart Materials 
Structures, 12(4), 580-601.  
Nichols, J. M., Trickey, S. T., Seaver, M., Motley, S. R., & Eisner, E. D. (2007). Using 
ambient vibrations to detect loosening of a composite-to-metal bolted joint in the 
presence of strong temperature fluctuations. Journal of Vibration and Acoustics, 
Transactions of the ASME, 129(6), 710-717.  
Noble, R. (2006). Integrity of bolted joints; hydrocarbon leak reduction by joint-integrity 
management. APPEA Journal, 46(1), 603-609.  
Noman, A. (2008). Application of vibration based methods and statistical pattern 
recognition techniques to structural health monitoring (M.A.Sc. dissertation). Dept. 
of Building, Civil and Environmental Engineering, Concordia University, Canada. 
Retrieved from ProQuest Dissertations & Theses Database. (UMI No. AAT 
MR45523)  
Okugawa, M., & Tanaka, T. (2007). Effect on detection sensitivity for smart washer 
configuration, and ambient temperature characteristics on bolted joint. In 
Proceedings of SPIE: Health Monitoring of Structural and Biological Systems 2007, 
California, USA, 19-22 March.  
Pai, N. G., & Hess, D. P. (2002). Three-dimensional finite element analysis of threaded 
fastener loosening due to dynamic shear load. Engineering Failure Analysis, 9(4), 
383-402. Retrieved from http://dx.doi.org/10.1016/S1350-6307(01)00024-3  
Park, G., & Inman, D. J. (2005). Impedance-based structural health monitoring. In 
Damage prognosis for aerospace, civil and mechanical systems, D. J. Inman, C. R. 
Farrar, V. Lopes Jr. & V. Steffen Jr. (Eds.), (pp. 275). John Wiley and Sons, Ltd.: 
Great Britain.  
149 
 
Park, G., Muntges, D. E., & Inman, D. J. (2003). Self-repairing joints employing shape-
memory alloy actuators. Journal of the Minerals, Metals and Materials Society, 
55(12), 33-37.  
Park, S., Yun, C., & Roh, Y. (2005). PZT-induced lamb waves and pattern recognitions 
for on-line health monitoring of jointed steel plates. In Proceedings of SPIE: Smart 
Structures and Materials 2005: Sensors and Smart Structures Technologies for 
Civil, Mechanical, and Aerospace Systems, California, USA, 7-10 March, 364-375.  
Park, S., Yun, C., Roh, Y., & Lee, J. (2006). PZT-based active damage detection 
techniques for steel bridge components. Smart Materials and Structures, 15(4), 957-
966.  
Park, G., Muntges, D. E., & Inman, D. J. (2001). Self-monitoring and self-healing jointed 
structures. Key Engineering Materials, 204-205, 75-84.  
Peairs, D. M., Park, G., & Inman, D. J. (2002). Reducing the cost of impedance-based 
structural health monitoring. In Proceedings of SPIE: Smart Nondestructive 
Evaluation for Health Monitoring of Structural and Biological Systems, California, 
USA, 18-19 March, pp. 301-310.  
Peairs, D. M., Park, G., & Inman, D. J. (2004). Improving accessibility of the impedance-
based structural health monitoring method. Journal of Intelligent Material Systems 
and Structures, 15(2), 129-140. Retrieved from 
http://dx.doi.org/10.1177/1045389X04039914  
Pines, D., & Salvino, L. (2006). Structural health monitoring using empirical mode 
decomposition and the hilbert phase. Journal of Sound and Vibration, 294(1-2), 97-
124.  
Quek, S. T., Tua, P. S., & Wang, Q. (2003). Detecting anomalies in beams and plate 
based on the hilbert-huang transform of real signals. SMART MATERIALS AND 
STRUCTURES, 12(Part 3), 447-460. Retrieved from stacks.iop.org/SMS/12/447  
Quek, S. T., Wang, Q., Zhang, L., & Ang, K. K. (2001). Sensitivity analysis of notch 
detection in beams by wavelet technique. International J. of Mechanical Sciences, 
43, 2899-2910.  
150 
 
Rezaei, D. (2010). Experimental and computational validation of a novel structural 
health monitoring strategy based on the hilbert-huang transform. (Doctoral 
dissertation). Dept. of Civil and Resource Engineering, Dalhousie University, 
Canada. 
Rezaei, D., & Taheri, F. (2010a). Health monitoring of pipeline girth weld using 
empirical mode decomposition. Smart Materials and Structures, 19(5).  
Rezaei, D., & Taheri, F. (2010b). Damage identification in beams using empirical mode 
decomposition. Journal of Structural Health Monitoring, in press.  
Rezaei, D., & Taheri, F. (2009). Experimental validation of a novel structural damage 
detection method based on empirical mode decomposition. Smart Materials and 
Structures, 18(4).  
Riley, M. J. (2005). Investigation into the feasibility of implementing piezoelectric 
patches for sensing and actuation of structural components (Doctoral dissertation). 
Dept. of Civil Engineering, Dalhousie University, Canada.  
Ritdumrongkul, S., Abe, M., Fujino, Y., & Miyashita, T. (2004). Quantitative health 
monitoring of bolted joints using a piezoceramic actuator-sensor. Smart Materials 
and Structures, 13(1), 20-29.  
Ritdumrongkul, S., & Fujino, Y. (2006). Identification of the location and level of 
damage in multiple-bolted-joint structures by PZT actuator-sensors. Journal of 
Structural Engineering, 132(2), 304-311.  
Rutherford, A., Park, G., & Farrar, C. (2007). Non-linear feature identifications based on 
self-sensing impedance measurements for structural health assessment. Mechanical 
Systems and Signal Processing, 21(1), 322-333.  
Rytter, A. (1993). Vibration based inspection of civil engineering structures (Doctoral 
dissertation). Dept. of Building Technology and Structural Engineering, Aalborg 
University, Denmark. Retrieved from ProQuest Dissertations & Theses Database. 
Shi, Z. (2000). Structural damage detection from modal strain energy change. Journal of 
Engineering Mechanics, 126(12), 1216-1223.  
151 
 
Sohn, H., Farrar, C. R., Hemez, F. M., Czarnecki, J. J., Shunk, D. D., Stinemates, D. W., 
& Nadler, B. R. (2003). A review of structural health monitoring literature: 1996–
2001 No. LA-13976-MS, Los Alamos National Laboratory, Los Alamos, NM.  
Stinemates, D. W., Farrar, C. R., Bennett, J. G., & Sohn, H. (2002). Structural health 
monitoring system design using finite element analysis. In Proceedings of SPIE: 
Smart Nondestructive Evaluation for Health Monitoring of Structural and Biological 
Systems, California, USA, 18-19 March, 169-178.  
Sturdevant, G. A. (1998). Hydraulically powered wrenches. In Handbook of bolts and 
bolted joints, J. H. Bickford, & S. Nassar (Eds.), (pp. 479). Marcel Dekker, Inc.: 
New York.   
Todd, M., Erickson, K., Chang, L., Lee, K., & Nichols, J. (2004). Using chaotic 
interrogation and attractor nonlinear cross-prediction error to detect fastener preload 
loss in an aluminum frame. Chaos, 14(2), 387-399.  
Todd, M. D., Nichols, J. M., Nichols, C. J., & Virgin, L. N. (2004). An assessment of 
modal property effectiveness in detecting bolted joint degradation: Theory and 
experiment. Journal of Sound and Vibration, 275(3-5), 1113-1126. Retrieved from 
http://dx.doi.org/10.1016/j.jsv.2003.10.037  
Worden, K., Farrar, C., Manson, G., & Park, G. (2007). The fundamental axioms of 
structural health monitoring. Philosophical Transactions of the Royal Society A: 
Mathematical, Physical and Engineering Sciences, 463(2082), 1639.  
Xu, Y. L., & Chen, J. (2004). Structural damage detection using empirical mode 
decomposition: Experimental investigation. Journal of Engineering Mechanics, 
130(11), 1279-1288.  
Yan, Y., Cheng, L., Wu, Z., & Yam, L. (2007). Development in vibration-based 
structural damage detection technique. Mechanical Systems and Signal Processing, 
21(5), 2198-2211.  
Yu, L., Link, M., Law, S., & Zhang, L. (1999). Structural damage identification 
procedure with application to a frame structure with bolted joints. In Proceedings of 
the 17th International Modal Analysis Conference, Florida, USA, 8-11 February, 
1387-1394.  
 
 
152 
 
APPENDIX A 
SUPPLEMENTARY RESULTS FROM CHAPTER 3 
 
A.1 TABLES  
   
  
153 
 
Table A-1  Session 1 EMD energy variability, 0-2000 Hz, IMF 1.  
Location 
/Sensor 
HC 1-1 HC 1-2 DC 1-1 DC 1-2 DC 1-3 DC 1-4 
Location A (*maximum values are highlighted) 
PZT1 4.16% 4.12% 5.31% 2.86% 3.65% 2.13% 
PZT2 2.29% 3.10% 10.05% 3.31% 3.87% 3.33% 
PZT3 6.08% 3.11% 19.81% 3.61% 4.38% 3.88% 
PZT4 1.10% 3.83% 7.63% 1.92% 2.63% 2.85% 
PZT5 7.85%* 5.88% 8.03% 2.84% 3.00% 3.61% 
PZT6 6.28% 3.34% 3.72% 1.96% 0.54% 2.22% 
Maximum 7.85% 5.88% 19.81% 3.61% 4.38% 3.88% 
              
Location B             
PZT1 3.62% 5.58% 2.51% 3.08% 4.88% 3.89% 
PZT2 1.69% 4.47% 6.88% 4.89% 4.92% 1.55% 
PZT3 5.56% 4.57% 2.51% 0.49% 3.28% 4.34% 
PZT4 1.34% 5.41% 3.01% 2.73% 2.93% 3.00% 
PZT5 5.75% 3.68% 5.33% 2.86% 6.12% 2.37% 
PZT6 3.85% 2.92% 4.13% 1.17% 2.21% 3.17% 
Maximum 5.75% 5.58% 6.88% 4.89% 6.12% 4.34% 
              
Location C             
PZT1 3.60% 2.13% 5.60% 6.19% 1.13% 4.12% 
PZT2 0.43% 1.64% 3.14% 6.22% 2.68% 1.32% 
PZT3 1.45% 2.93% 5.28% 2.85% 0.77% 1.93% 
PZT4 0.27% 1.29% 5.15% 7.45% 2.78% 1.38% 
PZT5 1.52% 3.42% 4.22% 5.30% 1.82% 2.07% 
PZT6 0.99% 2.29% 3.93% 5.56% 2.41% 0.54% 
Maximum 3.60% 3.42% 5.60% 7.45% 2.78% 4.12% 
              
Location D             
PZT1 1.00% 4.12% 1.14% 0.73% 2.70% 2.37% 
PZT2 1.27% 3.10% 1.94% 0.65% 1.47% 2.44% 
PZT3 0.22% 3.11% 1.19% 2.05% 2.62% 3.06% 
PZT4 1.50% 3.83% 1.54% 1.18% 2.08% 3.73% 
PZT5 3.03% 5.88% 0.98% 1.52% 2.56% 2.60% 
PZT6 0.93% 3.34% 0.36% 1.73% 2.48% 3.82% 
Maximum 3.03% 5.88% 1.94% 2.05% 2.70% 3.82% 
              
Average             
PZT1 3.09% 3.99% 3.64% 3.21% 3.09% 3.13% 
PZT2 1.42% 3.08% 5.50% 3.77% 3.24% 2.16% 
PZT3 3.33% 3.43% 7.20% 2.25% 2.76% 3.30% 
PZT4 1.05% 3.59% 4.33% 3.32% 2.61% 2.74% 
PZT5 4.54% 4.72% 4.64% 3.13% 3.38% 2.66% 
PZT6 3.01% 2.97% 3.04% 2.60% 1.91% 2.44% 
154 
 
Table A-2  Session 1 EMD energy variability, 0-2000 Hz, IMF 2.  
Location 
/Sensor 
HC 1-1 HC 1-2 DC 1-1 DC 1-2 DC 1-3 DC 1-4 
Location A (*maximum values are highlighted) 
PZT1 5.60%* 3.80% 3.24% 2.18% 6.12% 7.79% 
PZT2 3.88% 1.77% 6.80% 4.07% 3.30% 10.21% 
PZT3 2.15% 2.77% 11.84% 1.20% 4.24% 7.14% 
PZT4 1.82% 4.88% 6.02% 1.56% 4.22% 8.68% 
PZT5 4.14% 4.65% 5.40% 4.06% 2.22% 9.37% 
PZT6 3.43% 1.98% 5.41% 1.42% 2.83% 10.34% 
Maximum 5.60% 4.88% 11.84% 4.07% 6.12% 10.34% 
              Location B     
   
  
PZT1 4.81% 6.43% 2.20% 3.73% 7.71% 2.13% 
PZT2 7.72% 1.48% 5.24% 4.91% 7.58% 2.37% 
PZT3 5.57% 7.33% 1.22% 3.37% 6.83% 3.78% 
PZT4 4.45% 2.88% 1.80% 6.56% 5.53% 2.06% 
PZT5 7.04% 3.03% 5.01% 2.38% 7.64% 3.33% 
PZT6 7.26% 8.94% 6.55% 3.23% 6.11% 2.69% 
Maximum 7.72% 8.94% 6.55% 6.56% 7.71% 3.78% 
              Location C   
    
  
PZT1 2.30% 6.68% 2.34% 11.48% 2.20% 4.14% 
PZT2 2.66% 10.14% 6.17% 7.51% 3.52% 1.24% 
PZT3 4.88% 8.36% 2.49% 6.49% 4.31% 2.26% 
PZT4 1.64% 8.68% 2.82% 7.78% 2.74% 3.24% 
PZT5 5.17% 2.83% 6.73% 6.34% 0.61% 2.62% 
PZT6 4.04% 4.23% 5.18% 6.34% 6.39% 4.88% 
Maximum 5.17% 10.14% 6.73% 11.48% 6.39% 4.88% 
              Location D   
    
  
PZT1 3.04% 8.78% 5.13% 7.06% 2.33% 2.07% 
PZT2 3.04% 6.67% 1.40% 4.69% 2.29% 2.40% 
PZT3 6.05% 7.08% 3.90% 6.82% 1.70% 2.43% 
PZT4 1.18% 5.40% 4.28% 4.86% 3.02% 8.40% 
PZT5 4.33% 6.69% 5.90% 5.64% 5.13% 7.60% 
PZT6 4.01% 5.89% 5.61% 8.48% 5.28% 7.08% 
Maximum 6.05% 8.78% 5.90% 8.48% 5.28% 8.40% 
       Average             
PZT1 3.94% 6.42% 3.23% 6.11% 4.59% 4.03% 
PZT2 4.33% 5.02% 4.90% 5.29% 4.18% 4.05% 
PZT3 4.66% 6.38% 4.86% 4.47% 4.27% 3.90% 
PZT4 2.27% 5.46% 3.73% 5.19% 3.88% 5.59% 
PZT5 5.17% 4.30% 5.76% 4.60% 3.90% 5.73% 
PZT6 4.68% 5.26% 5.69% 4.87% 5.15% 6.25% 
155 
 
Table A-3  Session 1 result repeatability, healthy states, 0-2000 Hz. 
Location A IMF 1 IMF 2 
 PZT1 75.35% 5.09% 
 PZT2 34.97% 3.89% 
 PZT3 97.22% 58.43% 
 PZT4 68.01% 7.53% 
 PZT5 19.18% 6.90% 
 PZT6 62.44% 9.15% 
     Location B 
 
 
 
PZT1 4.89% 9.42% 
 PZT2 0.23% 2.41% 
 PZT3 81.37% 63.39% 
 PZT4 1.71% 0.81% 
 PZT5 5.24% 1.49% 
 PZT6 2.57% 4.80% 
     Location C 
 
 
 
PZT1 0.15% 1.50% 
 
PZT2 1.44% 4.16% 
 PZT3 1.50% 1.90% 
 PZT4 2.92% 1.69% 
 PZT5 1.32% 5.02% 
 PZT6 1.38% 4.92% 
     Location D 
 
 
 PZT1 3.09% 6.95% 
 
PZT2 3.73% 1.55% 
 
PZT3 2.05% 2.09% 
 PZT4 5.13% 8.43% 
 PZT5 1.71% 2.85% 
 PZT6 1.62% 0.33% 
 
*Variations over 20% are highlighted 
 
 
 
 
 
 
 
 
 
156 
 
Table A-4  Session 2 EMD energy variability, 0-2500 Hz, IMF 1, aluminium tip. 
Location 
/Sensor 
HC 2-1 
(%) 
HC 2-2 
(%) 
HC 2-3 
(%) 
HC 2-4 
(%) 
DC 2-1 
(%) 
DC 2-2 
(%) 
DC 2-3 
(%) 
DC 2-4 
(%) 
DC 2-5 
(%) 
DC 2-6 
(%) 
DC 2-7 
(%) 
DC 2-8 
(%) 
DC 2-9 
(%) 
Loc.  A                           
PZT7 3.58 2.51 3.10 2.69 4.93 4.33 2.25 4.25 4.26 2.59 4.48 5.01 4.81 
PZT8 2.49 3.29 3.41 2.85 2.74 2.26 4.23 4.77 3.66 3.03 6.74 4.03 7.53 
PZT10 4.18 2.62 3.08 3.20 1.02 2.57 1.68 4.75 4.40 3.54 3.74 6.42 6.44 
PZT11 6.59 3.84 5.25 3.48 2.95 2.51 2.74 2.70 4.93 1.47 5.40 4.01 5.81 
PZT12 3.08 4.45 2.68 2.65 4.75 2.65 1.91 3.39 2.58 2.66 6.22 7.38 6.17 
Max 6.59 4.45 5.25 3.48 4.93 4.33 4.23 4.77 4.93 3.54 6.74 7.38 7.53 
                           Loc. B                           
PZT7 3.37 4.09 4.18 2.62 4.65 3.52 2.91 4.66 3.26 7.39 4.26 2.75 5.32 
PZT8 2.56 4.16 3.72 1.39 6.04 3.17 1.65 2.63 1.00 5.44 3.34 4.23 5.60 
PZT10 3.18 4.99 4.45 2.34 3.65 3.72 2.56 3.11 1.43 5.36 3.00 2.33 5.32 
PZT11 1.91 6.52 4.38 3.30 4.27 0.93 2.03 1.58 2.48 2.72 6.56 4.72 6.02 
PZT12 3.85 5.41 5.56 3.85 7.67 4.09 3.34 3.28 2.89 3.04 1.85 1.54 3.79 
Max 3.85 6.52 5.56 3.85 7.67 4.09 3.34 4.66 3.26 7.39 6.56 4.72 6.02 
                           Loc.  C                           
PZT7 5.09 3.19 3.77 4.57 8.73 5.39 5.39 3.19 5.55 3.48 4.79 5.78 3.44 
PZT8 4.27 4.11 1.48 3.14 8.19 1.93 5.15 6.58 6.06 4.16 3.53 6.81 4.35 
PZT10 6.59 2.85 3.35 3.65 9.07 4.76 4.43 3.86 4.08 4.70 3.14 5.95 3.78 
PZT11 
11.1
9 2.65 3.37 2.12 6.94 2.72 3.06 2.80 3.86 3.56 3.61 5.68 5.40 
PZT12 5.16 1.81 4.03 5.79 8.51 2.26 5.91 4.06 5.46 2.88 3.55 2.78 4.49 
Max 11.19 4.11 4.03 5.79 9.07 5.39 5.91 6.58 6.06 4.70 4.79 6.81 5.40 
                           Loc.  D                           
PZT7 6.47 2.79 5.09 3.26 6.73 4.10 6.84 5.99 6.28 5.09 6.92 3.89 6.91 
PZT8 8.68 4.44 6.85 6.13 5.73 3.15 2.64 4.22 3.99 7.24 6.71 4.90 8.29 
PZT10 3.01 5.18 3.27 5.88 8.30 2.19 3.57 3.17 5.97 7.28 5.11 3.30 3.93 
PZT11 2.89 6.35 5.66 5.35 4.15 3.24 5.72 5.03 6.61 3.39 2.64 6.79 3.00 
PZT12 3.76 5.35 4.46 4.88 4.19 4.55 3.45 5.34 5.63 6.21 5.92 6.01 5.36 
Max 8.68 6.35 6.85 6.13 8.30 4.55 6.84 5.99 6.61 7.28 6.92 6.79 8.29 
                           Average                           
PZT7 4.63 3.14 4.03 3.28 6.26 4.34 4.35 4.52 4.84 4.64 5.11 4.36 5.12 
PZT8 4.50 4.00 3.86 3.38 5.68 2.63 3.42 4.55 3.68 4.97 5.08 4.99 6.44 
PZT10 4.24 3.91 3.54 3.77 5.51 3.31 3.06 3.72 3.97 5.22 3.75 4.50 4.87 
PZT11 5.65 4.84 4.66 3.56 4.58 2.35 3.39 3.03 4.47 2.79 4.55 5.30 5.06 
PZT12 3.96 4.26 4.18 4.29 6.28 3.39 3.65 4.01 4.14 3.70 4.38 4.43 4.95 
Maximum values are highlighted 
157 
 
Table A-5  Session 2 EMD energy variability, 0-2500 Hz, IMF 1, plastic tip. 
Location 
/Sensor 
HC 2-1 
(%) 
HC 2-2 
(%) 
HC 2-3 
(%) 
HC 2-4 
(%) 
DC 2-1 
(%) 
DC 2-2 
(%) 
DC 2-3 
(%) 
DC 2-4 
(%) 
DC 2-5 
(%) 
DC 2-6 
(%) 
DC 2-7 
(%) 
DC 2-8 
(%) 
DC 2-9 
(%) 
Loc. A                           
PZT7 7.84 5.62 3.85 5.61 7.13 1.59 1.36 7.13 10.17 16.82 10.32 10.94 15.20 
PZT8 4.68 6.18 6.34 5.80 7.41 2.11 3.10 5.18 8.31 14.86 6.39 14.23 13.92 
PZT10 5.71 3.09 8.45 4.92 3.95 2.82 3.07 6.97 2.63 16.30 4.74 10.01 15.20 
PZT11 5.04 4.04 6.81 5.41 6.43 2.19 4.01 4.70 4.60 8.54 5.38 7.59 13.92 
PZT12 4.46 4.10 10.29 5.80 6.60 1.64 2.01 5.97 3.43 7.97 4.49 4.91 0.00 
Max 7.84 6.18 10.29 5.80 7.41 2.82 4.01 7.13 10.17 16.82 10.32 14.23 15.20 
                          
Loc. B                           
PZT7 3.71 5.79 2.09 1.70 3.85 4.63 4.63 6.75 4.88 4.46 53.90 6.51 8.36 
PZT8 3.50 4.49 1.89 1.77 6.03 4.11 5.06 8.02 6.86 6.29 53.24 6.60 7.76 
PZT10 4.67 3.56 5.05 1.53 2.09 4.76 5.56 6.89 4.21 4.05 34.18 7.24 10.16 
PZT11 5.01 2.53 2.89 2.88 2.31 4.07 3.66 6.00 4.96 5.12 22.92 5.04 6.88 
PZT12 5.93 4.11 9.73 1.03 2.83 3.83 2.84 3.64 2.97 3.46 8.74 2.17 5.55 
Max 5.93 5.79 9.73 2.88 6.03 4.76 5.56 8.02 6.86 6.29 53.90 7.24 10.16 
                           
Loc. C                           
PZT7 2.21 4.04 3.01 2.38 1.08 5.00 3.34 15.32 7.50 13.03 13.86 31.54 26.02 
PZT8 2.30 5.90 2.23 2.95 4.15 13.19 4.18 14.44 6.08 15.58 17.05 25.34 22.76 
PZT10 3.09 5.22 4.62 2.79 2.64 9.64 6.13 4.17 5.09 8.87 16.46 20.12 26.21 
PZT11 3.24 4.28 1.66 3.84 1.54 7.69 4.58 4.77 6.84 7.52 10.14 13.09 5.24 
PZT12 2.15 7.27 5.39 1.71 3.39 2.88 5.48 4.04 2.71 4.97 7.85 3.58 8.26 
Max 3.24 7.27 5.39 3.84 4.15 13.19 6.13 15.32 7.50 15.58 17.05 31.54 26.21 
                           
Loc. D                           
PZT7 2.53 2.04 5.31 3.50 3.35 4.95 3.95 29.39 12.91 9.90 34.94 22.85 7.95 
PZT8 3.94 4.59 9.99 3.34 2.65 5.49 4.61 29.34 13.93 11.99 20.19 25.98 11.48 
PZT10 3.11 2.87 6.10 2.98 4.46 5.10 2.33 21.57 7.02 8.10 29.26 17.02 10.64 
PZT11 2.90 3.05 8.21 4.37 3.09 4.60 4.00 9.40 6.15 5.85 10.71 5.55 4.36 
PZT12 2.31 3.55 17.38 5.52 4.41 5.03 3.54 3.89 3.24 3.96 8.89 10.19 6.29 
Max 3.94 4.59 17.38 5.52 4.46 5.49 4.61 29.39 13.93 11.99 34.94 25.98 11.48 
                            
Avg,                           
PZT7 4.07 4.37 3.57 3.30 3.85 4.04 3.32 14.65 8.87 11.05 28.26 17.96 14.38 
PZT8 3.60 5.29 5.11 3.47 5.06 6.23 4.24 14.24 8.79 12.18 24.22 18.04 13.98 
PZT10 4.15 3.68 6.06 3.06 3.29 5.58 4.27 9.90 4.74 9.33 21.16 13.60 15.55 
PZT11 4.05 3.47 4.89 4.12 3.34 4.64 4.06 6.22 5.64 6.75 12.29 7.82 7.60 
PZT12 3.71 4.76 10.70 3.51 4.31 3.35 3.47 4.39 3.09 5.09 7.49 5.21 5.02 
Maximum values are highlighted 
158 
 
Table A-6  Session 2 result repeatability, healthy cases, 0-2500 Hz, IMF1, aluminium tip. 
   0-2500 Hz 
 
Case 2 
with 
Case 1 
Case 3 
with 
Case 1 
Case 4 
with 
Case 1 
  Case 3 
with 
Case 2 
Case 4 
with 
Case 2 
  Case 4 
with 
Case 3  
    
 
    
Location A 
        
PZT7 0.38% 3.20% 2.69% 
 
3.58% 2.31% 
 
5.88% 
PZT8 1.93% 4.63% 3.85% 
 
6.55% 5.77% 
 
0.79% 
PZT10 7.96% 5.66% 4.83% 
 
13.55% 3.14% 
 
10.46% 
PZT11 0.45% 2.44% 4.85% 
 
1.98% 5.30% 
 
7.28% 
PZT12 0.88% 7.51% 1.66% 
 
8.38% 0.79% 
 
9.16% 
Location B 
        
PZT7 7.25% 3.39% 0.05% 
 
3.87% 7.30% 
 
3.43% 
PZT8 7.64% 6.97% 0.14% 
 
14.54% 7.78% 
 
6.84% 
PZT10 4.09% 0.83% 7.51% 
 
4.92% 3.43% 
 
8.33% 
PZT11 0.42% 2.12% 3.09% 
 
2.54% 2.67% 
 
5.21% 
PZT12 2.61% 10.04% 7.31% 
 
7.45% 4.71% 
 
2.75% 
Location C 
        
PZT7 2.05% 7.59% 6.72% 
 
9.62% 8.76% 
 
0.87% 
PZT8 2.93% 9.41% 5.31% 
 
12.30% 8.22% 
 
4.12% 
PZT10 6.96% 5.63% 5.89% 
 
12.54% 1.08% 
 
11.48% 
PZT11 0.74% 4.33% 5.06% 
 
5.07% 5.81% 
 
0.74% 
PZT12 0.81% 0.53% 2.01% 
 
1.34% 2.82% 
 
1.47% 
Location D 
        
PZT7 2.44% 7.46% 11.54% 
 
9.88% 13.94% 
 
4.11% 
PZT8 1.06% 4.45% 7.25% 
 
5.51% 8.31% 
 
2.81% 
PZT10 12.03% 6.47% 2.93% 
 
18.36% 9.14% 
 
9.38% 
PZT11 5.64% 0.13% 2.46% 
 
5.51% 8.09% 
 
2.59% 
PZT12 2.51% 2.03% 7.99% 
 
4.53% 10.48% 
 
5.98% 
 
 
 
  
159 
 
Table A-7  Session 2 result repeatability, healthy cases, 0-2500 Hz, IMF1, plastic tip. 
 
0-2500 Hz 
 
Case 2 
with 
Case 1 
Case 3 
with 
Case 1 
Case 4 
with 
Case 1 
  Case 3 
with 
Case 2 
Case 4 
with 
Case 2 
  Case 4 
with 
Case 3  
    
 
    
Location A 
        
PZT7 9.02% 6.13% 1.14% 
 
2.90% 10.14% 
 
7.27% 
PZT8 11.64% 1.53% 0.74% 
 
13.14% 12.37% 
 
0.79% 
PZT10 4.58% 7.46% 4.73% 
 
11.99% 0.16% 
 
12.15% 
PZT11 6.93% 5.43% 1.73% 
 
1.51% 5.21% 
 
3.71% 
PZT12 1.39% 16.95% 0.36% 
 
18.30% 1.76% 
 
16.60% 
Location B 
        
PZT7 15.49% 5.50% 2.54% 
 
10.07% 13.00% 
 
2.97% 
PZT8 14.50% 3.66% 3.28% 
 
18.07% 11.27% 
 
6.94% 
PZT10 1.63% 8.47% 5.46% 
 
10.08% 3.83% 
 
13.86% 
PZT11 8.39% 6.72% 4.79% 
 
1.68% 3.62% 
 
1.94% 
PZT12 3.63% 11.30% 2.04% 
 
14.87% 5.66% 
 
9.28% 
Location C 
        
PZT7 3.74% 6.27% 6.69% 
 
9.99% 10.41% 
 
0.42% 
PZT8 4.89% 10.73% 7.28% 
 
15.53% 12.12% 
 
3.48% 
PZT10 12.29% 0.04% 8.12% 
 
12.33% 4.22% 
 
8.16% 
PZT11 3.44% 1.79% 0.56% 
 
1.65% 4.00% 
 
2.35% 
PZT12 1.30% 23.64% 0.99% 
 
22.41% 0.31% 
 
22.70% 
Location D 
        
PZT7 8.65% 3.75% 9.32% 
 
12.36% 17.82% 
 
5.59% 
PZT8 3.85% 12.92% 12.83% 
 
16.70% 16.60% 
 
0.10% 
PZT10 13.46% 6.94% 1.52% 
 
20.21% 14.95% 
 
5.43% 
PZT11 5.26% 4.62% 5.26% 
 
9.86% 10.49% 
 
0.64% 
PZT12 1.77% 36.31% 7.86% 
 
34.77% 9.61% 
 
42.95% 
 Variations over 20% are highlighted 
 
 
  
160 
 
Table A-8  Session 3 EMD energy variability, 0-2000 Hz, IMF 1, aluminium tip. 
Location 
/Sensor 
HC 3-1 DC 3-1 DC 3-2 DC 3-3 
Location A         
PZT10 10.09% 3.50% 6.18% 6.24% 
PZT11 6.70% 3.84% 0.97% 3.59% 
PZT12 8.63% 7.47% 3.35% 2.55% 
PZT13 9.49% 4.48% 2.16% 5.21% 
PZT14 4.02% 0.61% 2.25% 1.63% 
PZT15 8.65% 6.99% 3.02% 2.51% 
PZT16 8.14% 1.94% 4.16% 1.35% 
Maximum 10.09% 7.47% 6.18% 6.24% 
     
Location B 
    
PZT10 6.80% 4.82% 4.45% 4.11% 
PZT11 2.26% 3.05% 2.83% 2.42% 
PZT12 7.02% 3.66% 7.87% 6.72% 
PZT13 7.94% 4.90% 6.90% 7.76% 
PZT14 4.93% 6.61% 4.96% 10.44% 
PZT15 6.03% 7.87% 8.65% 7.72% 
PZT16 2.97% 2.45% 7.49% 4.53% 
Maximum 7.94% 6.61% 7.87% 10.44 
     Location C 
    
PZT10 5.51% 5.99% 0.43% 3.18% 
PZT11 4.39% 5.14% 5.45% 5.91% 
PZT12 1.84% 1.35% 7.08% 10.62% 
PZT13 5.28% 4.77% 5.65% 9.62% 
PZT14 7.76% 3.04% 6.61% 7.80% 
PZT15 4.15% 6.93% 6.21% 4.59% 
PZT16 3.03% 3.25% 4.03% 3.17% 
Maximum 7.76% 5.99% 7.08% 10.62 
     Location D 
    
PZT10 8.80% 2.13% 4.99% 4.04% 
PZT11 11.80% 3.04% 4.11% 6.75% 
PZT12 7.08% 3.30% 5.27% 10.21% 
PZT13 8.53% 5.72% 3.38% 6.07% 
PZT14 8.45% 6.91% 4.46% 7.61% 
PZT15 9.32% 7.08% 2.34% 2.91% 
PZT16 5.81% 5.35% 6.34% 7.83% 
Maximum 11.80% 6.91% 5.27% 10.21 
161 
 
Table A-9  Session 3 EMD energy variability, 0-2000 Hz, IMF 1, plastic tip. 
Location 
/Sensor 
HC 3-1 DC 3-1 DC 3-2 DC 3-3 
Location A 
    
PZT10 6.66% 2.09% 2.18% 2.30% 
PZT11 6.09% 2.10% 3.52% 3.80% 
PZT12 5.52% 2.07% 2.90% 2.64% 
PZT13 5.13% 1.57% 0.99% 0.96% 
PZT14 3.99% 2.16% 1.90% 2.32% 
PZT15 7.17% 1.35% 3.65% 2.25% 
PZT16 4.42% 1.96% 3.53% 0.42% 
Maximum 6.66% 2.16% 3.52% 3.80% 
     
Location B 
    
PZT10 4.71% 2.07% 6.69% 2.33% 
PZT11 3.31% 0.93% 6.91% 2.12% 
PZT12 3.22% 4.32% 5.55% 1.94% 
PZT13 1.56% 1.93% 5.86% 2.25% 
PZT14 3.95% 2.23% 7.00% 0.64% 
PZT15 3.03% 2.31% 5.67% 1.12% 
PZT16 4.39% 1.06% 4.25% 1.72% 
Maximum 4.71% 4.32% 7.00% 2.33% 
     Location C 
    
PZT10 6.11% 2.23% 4.02% 0.92% 
PZT11 5.37% 2.53% 4.08% 1.90% 
PZT12 3.29% 4.78% 4.72% 2.22% 
PZT13 3.99% 2.02% 3.21% 1.74% 
PZT14 5.04% 0.64% 2.39% 2.31% 
PZT15 2.83% 1.49% 4.26% 1.62% 
PZT16 4.40% 1.17% 4.50% 1.22% 
Maximum 6.11% 4.78% 4.72% 2.31% 
     Location D 
    
PZT10 3.46% 3.15% 3.30% 1.60% 
PZT11 3.70% 4.02% 4.56% 2.23% 
PZT12 5.43% 3.15% 4.13% 1.34% 
PZT13 3.80% 4.76% 3.34% 1.82% 
PZT14 3.41% 2.84% 4.63% 1.58% 
PZT15 2.16% 3.95% 4.64% 0.95% 
PZT16 3.87% 3.78% 3.67% 2.10% 
Maximum 5.43% 4.76% 4.63% 2.23% 
162 
 
Table A-10  Session 3 EMD energy variability, 0-2000 Hz, IMF 2, aluminium tip. 
Location 
/Sensor 
HC 3-1 DC 3-1 DC 3-2 DC 3-3 
Location A 
    
PZT10 16.29% 11.76% 5.29% 4.75% 
PZT11 14.11% 9.54% 4.74% 0.98% 
PZT12 9.15% 6.92% 4.76% 9.45% 
PZT13 18.67% 3.76% 5.26% 12.84% 
PZT14 18.99% 17.40% 8.41% 10.11% 
PZT15 19.70% 14.54% 6.33% 11.07% 
PZT16 21.14% 19.62% 10.60% 5.13% 
Maximum 18.99% 17.40% 8.41% 12.84% 
     
Location B 
    
PZT10 14.40% 4.50% 10.23% 8.95% 
PZT11 7.70% 3.95% 5.15% 8.76% 
PZT12 2.17% 11.88% 4.89% 5.08% 
PZT13 12.54% 5.62% 9.74% 6.05% 
PZT14 18.14% 13.81% 7.94% 7.35% 
PZT15 21.14% 11.24% 8.03% 8.18% 
PZT16 6.83% 2.30% 4.60% 12.49% 
Maximum 18.14% 13.81% 10.23% 8.95% 
     Location C 
    
PZT10 4.44% 3.75% 9.59% 4.40% 
PZT11 9.50% 8.09% 10.10% 3.91% 
PZT12 11.57% 18.90% 9.79% 6.12% 
PZT13 10.63% 10.79% 10.17% 3.90% 
PZT14 13.12% 10.66% 13.06% 11.53% 
PZT15 17.59% 11.43% 11.62% 18.88% 
PZT16 10.56% 4.68% 13.09% 10.46% 
Maximum 13.12% 18.90% 13.06% 11.53% 
     Location D 
    
PZT10 12.62% 3.16% 7.11% 5.96% 
PZT11 5.87% 17.25% 15.53% 10.75% 
PZT12 12.73% 1.47% 4.13% 3.75% 
PZT13 6.92% 2.12% 9.88% 9.36% 
PZT14 10.65% 7.84% 9.10% 6.91% 
PZT15 13.57% 12.37% 6.47% 5.64% 
PZT16 12.01% 4.75% 10.73% 5.73% 
Maximum 12.73% 17.25% 15.53% 10.75% 
163 
 
Table A-11  Session 3 EMD energy variability, 0-2000 Hz, IMF 2, plastic tip. 
Location 
/Sensor 
HC 3-1 DC 3-1 DC 3-2 DC 3-3 
Location A 
    
PZT10 4.03% 5.66% 3.61% 2.97% 
PZT11 2.48% 2.13% 5.00% 2.01% 
PZT12 5.81% 4.26% 1.53% 4.95% 
PZT13 5.40% 1.08% 4.92% 0.91% 
PZT14 7.73% 2.83% 3.13% 1.58% 
PZT15 3.04% 2.56% 7.00% 4.93% 
PZT16 3.66% 2.94% 1.66% 2.91% 
Maximum 7.73% 5.66% 5.00% 4.95% 
     
Location B 
    
PZT10 3.29% 1.26% 4.08% 6.39% 
PZT11 6.77% 2.77% 3.97% 3.86% 
PZT12 4.52% 5.84% 3.45% 2.70% 
PZT13 6.45% 2.60% 4.35% 4.32% 
PZT14 2.08% 2.73% 8.17% 3.59% 
PZT15 3.10% 3.06% 2.14% 2.46% 
PZT16 3.71% 3.21% 4.77% 6.24% 
Maximum 6.77% 5.84% 8.17% 6.39% 
     Location C 
    
PZT10 3.93% 2.83% 3.42% 2.68% 
PZT11 1.24% 6.75% 1.26% 1.34% 
PZT12 1.52% 4.07% 3.54% 3.04% 
PZT13 4.88% 4.07% 2.10% 2.76% 
PZT14 3.54% 3.66% 0.18% 1.77% 
PZT15 6.83% 5.66% 4.95% 1.86% 
PZT16 3.35% 2.47% 1.09% 0.59% 
Maximum 4.88% 6.75% 3.54% 3.04% 
     Location D 
    
PZT10 5.26% 2.23% 4.90% 4.89% 
PZT11 4.76% 1.42% 2.29% 1.30% 
PZT12 3.02% 5.23% 4.73% 3.21% 
PZT13 3.25% 2.19% 1.72% 2.43% 
PZT14 3.15% 2.29% 3.98% 3.93% 
PZT15 3.92% 2.95% 2.79% 1.39% 
PZT16 2.27% 3.11% 4.73% 1.81% 
Maximum 5.26% 5.23% 4.90% 4.89% 
 
164 
 
A.2 GRAPHS   
 
 
 
Figure A-1  EMD-EDI, session 3, location B, 0-2000 Hz, IMF2, aluminium tip. 
 
 
 
Figure A-2  EMD-EDI, session 3, location C, 0-2000 Hz, IMF2, aluminium tip. 
 
 
 
Figure A-3  EMD-EDI, session 3, location E, 0-2000 Hz, IMF2, aluminium tip. 
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
      Right Side         Left Side 
      Right Side        Left Side 
      Right Side        Left Side 
165 
 
 
 
Figure A-4  EMD-EDI, session 3, location F, 0-2000 Hz, IMF2, aluminium tip. 
 
 
 
Figure A-5  EMD-EDI, session 3, location B, 0-2000 Hz, IMF2, plastic tip. 
 
 
 
Figure A-6  EMD-EDI, session 3, location C, 0-2000 Hz, IMF2, plastic tip. 
 
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
      Right Side        Left Side 
      Right Side       Left Side 
         Right Side        Left Side 
166 
 
 
Figure A-7  EMD-EDI, session 3, location E, 0-2000 Hz, IMF2, plastic tip. 
 
 
 
Figure A-8  EMD-EDI, session 3, location F, 0-2000 Hz, IMF2, plastic tip. 
 
 
 
Figure A-9  EMD-EDI, session 3, location B, 0-2000 Hz, combined IMFs, aluminium tip. 
 
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
10
20
30
40
50
60
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
         Right Side        Left Side 
         Right Side        Left Side 
        Right Side        Left Side 
167 
 
 
 
Figure A-10  EMD-EDI, session 3, location C, 0-2000 Hz, combined IMFs, aluminium tip. 
 
 
 
Figure A-11  EMD-EDI, session 3, location E, 0-2000 Hz, combined IMFs, aluminium tip. 
 
 
 
Figure A-12  EMD-EDI, session 3, location F, 0-2000 Hz, combined IMFs, aluminium tip. 
 
 
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
        Right Side        Left Side 
        Right Side        Left Side 
        Right Side        Left Side 
168 
 
 
Figure A-13  EMD-EDI, session 3, location B, 0-2000 Hz, combined IMFs, plastic tip. 
 
 
 
Figure A-14  EMD-EDI, session 3, location C, 0-2000 Hz, combined IMFs, plastic tip. 
 
 
 
Figure A-15  EMD-EDI, session 3, location E, 0-2000 Hz, combined IMFs, plastic tip. 
 
 
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
        Right Side        Left Side 
        Right Side        Left Side 
        Right Side        Left Side 
169 
 
 
Figure A-16  EMD-EDI, session 3, location F, 0-2000 Hz, combined IMFs, plastic tip. 
 
0
20
40
60
80
100
120
PZT10 PZT11 PZT12 PZT13 PZT14 PZT15 PZT16
E
M
D
-E
D
I
Sensor
DC 3-1
DC 3-2
DC 3-3
        Right Side        Left Side 
170 
 
APPENDIX B 
SUPPLEMENTARY RESULTS FROM CHAPTER 5 
 
B.1 GRAPHS   
 
 
 
Figure B-1  EMD-EDI, damage scenario 1, location A, 0-1500 Hz, combined IMFs. 
 
 
 
 
 
Figure B-2  EMD-EDI, damage scenario 1, location S, 500-2000 Hz, combined IMFs. 
 
0
50
100
150
200
250
300
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
0
50
100
150
200
250
300
350
400
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
171 
 
 
Figure B-3  EMD-EDI, damage scenario 2, location S, 0-1500 Hz, combined IMFs. 
 
 
 
Figure B-4  EMD-EDI, damage scenario 3, location A, 0-1500 Hz, combined IMFs. 
 
 
 
 
Figure B-5  EMD-EDI, damage scenario 3, location S, 0-1500 Hz, combined IMFs. 
 
 
0
20
40
60
80
100
120
140
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 2-6
DC 2-7
DC 2-8
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
172 
 
 
Figure B-6  EMD-EDI, damage scenario 4, location A, 0-2500 Hz, combined IMFs. 
 
 
 
 
Figure B-7  EMD-EDI, damage scenario 4, location S, 0-2500 Hz, combined IMFs. 
 
 
 
 
Figure B-8  EMD-EDI, damage scenario 1, location A, 0-2000 Hz, IMF1. 
 
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
173 
 
 
Figure B-9  EMD-EDI, damage scenario 1, location A, 0-2000 Hz, IMF2. 
 
 
 
Figure B-10  EMD-EDI, damage scenario 1, location S, 0-2000 Hz, IMF1. 
 
 
 
 
 
Figure B-11  EMD-EDI, damage scenario 1, location S, 0-2000 Hz, IMF2. 
 
0
30
60
90
120
150
180
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
0
40
80
120
160
200
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 1-1
DC 1-2
DC 1-3
DC 1-4
DC 1-5
174 
 
 
Figure B-12  EMD-EDI, damage scenario 2, location A, 0-2000 Hz, IMF1. 
 
 
 
Figure B-13  EMD-EDI, damage scenario 2, location A, 0-2000 Hz, IMF2. 
 
 
 
 
Figure B-14  EMD-EDI, damage scenario 2, location S, 0-2000 Hz, IMF1. 
0
20
40
60
80
100
120
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 2-6
DC 2-7
DC 2-8
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 2-6
DC 2-7
DC 2-8
0
30
60
90
120
150
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 2-6
DC 2-7
DC 2-8
175 
 
 
Figure B-15  EMD-EDI, damage scenario 2, location S, 0-2000 Hz, IMF2. 
 
 
 
 
Figure B-16  EMD-EDI, damage scenario 3, location A, 0-2000 Hz, IMF1. 
 
 
 
 
Figure B-17  EMD-EDI, damage scenario 3, location A, 0-2000 Hz, IMF2. 
 
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 2-6
DC 2-7
DC 2-8
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
176 
 
 
Figure B-18  EMD-EDI, damage scenario 3, location S, 0-2000 Hz, IMF1. 
 
 
 
Figure B-19  EMD-EDI, damage scenario 3, location S, 0-2000 Hz, IMF2. 
 
 
 
 
 
Figure B-20  EMD-EDI, damage scenario 4, location A, 0-2000 Hz, IMF1. 
 
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 3-9
DC 3-10
DC 3-11
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
177 
 
 
Figure B-21  EMD-EDI, damage scenario 4, location A, 0-2000 Hz, IMF2. 
 
 
 
 
Figure B-22  EMD-EDI, damage scenario 4, location S, 0-2000 Hz, IMF1. 
 
 
 
 
Figure B-23  EMD-EDI, damage scenario 4, location S, 0-2000 Hz, IMF2. 
 
 
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
0
20
40
60
80
100
1 2 3 4 5 6
E
M
D
-E
D
I
PZT Sensor
DC 4-12
DC 4-13
DC 4-14
DC 4-15
DC 4-16
DC 4-17
178 
 
APPENDIX C 
SUPPLEMENTARY RESULTS FROM CHAPTER 6 
 
C.1 TABLES  
          Table C-1  EMD energy variability, 0-2000 Hz, IMF 1.  
Boundary Condition: Free-free 
Scenario 1 
Sensor HC 1-1 DC 1-1 DC 1-2 DC 1-3 DC 1-4 
 
(%) (%) (%) (%) (%) 
PZT1 19.97 15.34 62.15 63.14 24.13 
PZT2 5.78 16.90 35.71 13.75 7.45 
PZT3 15.90 19.90 53.47 49.24 14.19 
PZT4 11.14 9.59 50.62 19.80 10.58 
PZT5 25.08 15.80 35.30 30.62 24.01 
PZT6 12.41 41.94 14.93 55.71 22.54 
PZT7 16.47 15.90 15.73 21.26 8.63 
Max 25.08 41.94 62.15 63.14 24.13 
  
 
 
   Scenario 2 
Sensor HC 2-1 DC 2-5 DC 2-6 DC 2-7 DC 2-8 
 
(%) (%) (%) (%) (%) 
PZT1 33.25 40.53 19.66 29.90 57.25 
PZT2 22.19 28.49 35.76 17.53 23.91 
PZT3 18.41 18.00 18.29 31.63 40.11 
PZT4 7.61 12.22 10.09 8.27 17.43 
PZT5 40.69 33.45 7.10 23.60 25.21 
PZT6 25.84 17.95 17.96 15.65 14.55 
PZT7 15.99 10.17 4.43 17.76 19.19 
Max 40.69 40.53 35.76 31.63 57.25 
 
 
 
 
 
179 
 
 
 
 
                     Table C-2  EMD energy variability, 0-1500 Hz, IMF 1. 
Boundary Condition: Cantilevered 
Scenario 1 
Sensor HC 1-1 DC 1-1 DC 1-2 DC 1-3 DC 1-4 
 
(%) (%) (%) (%) (%) 
PZT1 26.72 11.34 41.90 7.48 8.23 
PZT2 48.63 40.98 30.49 9.23 17.11 
PZT3 28.64 11.60 22.10 7.05 7.28 
PZT4 48.89 53.72 41.11 9.88 18.20 
PZT5 23.56 11.95 49.47 7.76 7.06 
PZT6 21.59 21.30 28.37 6.12 20.06 
PZT7 32.46 12.53 28.92 6.47 7.78 
Max 48.89 53.72 49.47 9.88 8.23 
   
 
 
  Scenario 2 
Sensor HC 2-1 DC 2-5 DC 2-6 DC 2-7 DC 2-8 
 
(%) (%) (%) (%) (%) 
PZT1 14.27 46.93 19.35 24.96 17.09 
PZT2 19.25 38.43 39.48 18.66 50.58 
PZT3 25.14 24.50 20.36 21.28 7.64 
PZT4 54.54 52.62 53.16 15.53 58.23 
PZT5 83.92 31.50 18.94 42.22 19.30 
PZT6 64.53 36.69 27.65 22.13 11.61 
PZT7 37.69 17.87 27.27 29.14 16.82 
Max 83.92 52.62 53.16 42.22 58.23 
 
 
 
 
 
 
 
 
 
180 
 
 
 
Table C-3  Result repeatability, healthy states, 0-2000 Hz. 
Boundary Condition: Free-free 
Sensor 
Case 1 
& 
Case 2 
& 
Case 1 
& 
Case 2 Case 3 Case 3 
PZT1 14.7% 28.2% 14.0% 
PZT2 16.3% 43.1% 28.8% 
PZT3 9.0% 20.2% 11.4% 
PZT4 6.7% 44.0% 38.4% 
PZT5 6.3% 25.1% 19.1% 
PZT6 18.6% 20.0% 37.2% 
PZT7 31.9% 8.6% 39.5% 
Max 31.9% 44.0% 39.5% 
    
    
Boundary Condition: Cantilevered 
Sensor 
Case 1 
& 
Case 2 
& 
Case 1 
& 
Case 2 Case 3 Case 3 
PZT1 1.0% 13.7% 14.6% 
PZT2 8.4% 11.7% 19.8% 
PZT3 0.7% 4.9% 4.3% 
PZT4 9.1% 11.2% 20.1% 
PZT5 29.5% 48.4% 22.1% 
PZT6 6.4% 17.7% 23.8% 
PZT7 11.3% 2.5% 8.8% 
Max 29.5% 48.4% 23.8% 
 
 
 
181 
 
APPENDIX D 
COPYRIGHT RELEASE REQUEST 
& STUDENT CONTRIBUTION TO MANUSCRIPT