Bartoli - 2007 - STRUCTURAL HEALTH MONITORING BY ULTRASONIC GUIDED WAVES

Electronic Theses and Dissertations
UC San Diego
Peer Reviewed
Title:
Structural health monitoring by ultrasonic guided waves
Author:
Bartoli, Ivan
Acceptance Date:
2007
Series:
UC San Diego Electronic Theses and Dissertations
Degree:
Ph. D., UC San Diego
Permalink:
http://escholarship.org/uc/item/2pn3c67k
Local Identifier:
b6635697
Abstract:
Guided ultrasonic waves provide a highly efficient method for the non-destructive evaluation
(NDE) and the structural health monitoring (SHM) of solids with finite cross-sectional dimensions
(waveguides). Compared to the widely used ultrasonic bulk waves, guided waves provide larger
monitoring ranges and the complete coverage of the waveguide cross-section. Compared to
global vibrations, guided waves offer increased sensitivity to smaller defects due to the smaller
wavelengths involved. These advantages can be fully exploited only once the complexities of
guided wave propagation (multimode, dispersion, frequency-dependent attenuation) are unveiled
and managed for the given test structure. This doctoral dissertation is aimed at developing a
Semi-Analytical Finite Element (SAFE) method for modeling wave propagation in waveguides of
arbitrary cross-section. The method requires the finite element discretization of the cross- section
of the waveguide, and assumes harmonic motion along the wave propagation direction. The
general SAFE technique was extended to account for viscoelastic material damping by allowing for
complex stiffness matrices for the material. The dispersive solutions are obtained in terms of phase
velocity, group velocity (for undamped media), energy velocity (for damped media), attenuation,
and cross-sectional mode shapes. Once the dispersive properties are computed, the wave motion
can be interpreted and the forced response can be predicted. The proposed SAFE formulation was
applied to enhance the use of ultrasonic guided waves in a number of applications and to interpret
the corresponding experimental results. The following three applications were considered : defect
detection in adhesively-bonded joints found in the wing skin-to-spar assemblies of Unmanned
Aerial Vehicles; defect detection in railroad tracks; load monitoring and defect detection in seven-
wire steel strands used in cables and prestressed concrete structures
Copyright Information:
All rights reserved unless otherwise indicated. Contact the author or original publisher for any
necessary permissions. eScholarship is not the copyright owner for deposited works. Learn more
at http://www.escholarship.org/help_copyright.html#reuse
eScholarship provides open access, scholarly publishing

services to the University of California and delivers a dynamic
research platform to scholars worldwide.
UNIVERSITY OF CALIFORNIA, SAN DIEGO
STRUCTURAL HEALTH MONITORING BY

ULTRASONIC GUIDED WAVES
A dissertation submitted in partial satisfaction of the

requirements for the degree of Doctor of Philosophy
in
Structural Engineering
by
Ivan Bartoli
Committee in charge:
Professor Francesco Lanza di Scalea, Chair

Prof. Thomas Bewley
Prof. Charles Farrar
Prof. William Hodgkiss
Prof. J. Enrique Luco
2007
Copyright
Ivan Bartoli, 2007
All rights reserved.

The Dissertation of Ivan Bartoli is approved, and it is acceptable in
quality and form for publication on microfilm:
Chair
University of California, San Diego
2007
iii
To my family
Michela, Gabriella e Domenico
iv
TABLE OF CONTENTS
Signature Page..……………………………………………………………………..iii
Dedication…………………………………………………………………………….iv
Table of Contents …………………………………………………………………..v

List of Tables ……………………………………………………………………….x
List of Figures……………………………………………………………………...xii
Acknowledgements………………………………………………………………..xxiv
Vita……………………………………………………………………………….. xxvii
Abstract of the dissertation……………………………………………………..xxx
1 INTRODUCTION ....................................................................................1
1.1 Research motivation .............................................................................. 1
1.2 Outline of the dissertation.................................................................... 4
2 INTRODUCTION TO ULTRASONIC GUIDED WAVES......................6
2.1 Introduction ........................................................................................... 6
2.2 Equations of motion in Unbounded Medium....................................... 9
2.3 Guided plate waves: Lamb waves ...................................................... 15
2.4 Analytical solutions of Guided Waves for Multilayered systems. ..... 27
2.5 Numerical solutions of guided waves in Multilayered systems. ........ 31
3 MODELING WAVE PROPAGATION IN DAMPED

WAVEGUIDES OF ARBITRARY CROSS-SECTION.......................... 35
3.1 Introduction ......................................................................................... 35

3.2 Viscoelastic models for wave propagation .......................................... 38
v
3.3 SAFE mathematical framework .......................................................... 40
3.3.1 Problem definition ........................................................................ 40

3.3.2 Equations of motion..................................................................... 42
3.3.3 Semi Analytical Finite element method ...................................... 44

3.3.4 Solutions for undamped media .................................................... 50
3.3.5 Solutions for damped media ........................................................ 52
3.3.6 Group and energy velocity .......................................................... 52

3.4 Results ................................................................................................. 55
3.4.1 Plate systems................................................................................ 55

3.4.2 Arbitrary cross-sections: viscoelastic railroad track .................... 76
3.4.3 Hollow Cylinder............................................................................ 78
3.5 SAFE for axial symmetric systems .................................................... 81

3.5.1 Formulation .................................................................................. 82
3.5.2 Numerical validation .................................................................... 91
3.5.3 Elastic copper tube filled with elastic and

viscoelastic bitumen....................................................................... 95
3.5.4 Elastic steel pipe coated by a thin layer

of viscoelastic bitumen .............................................................. 104
3.5.5 Viscoelastic steel strand embedded in concrete ........................ 108
3.6 Forced responces................................................................................ 113

3.6.1 Response to harmonic force ....................................................... 113
3.6.2 Response to arbitrary force ....................................................... 120

3.7 Discussion and conclusions................................................................ 121
3.8 Acknowledgements ............................................................................. 122
vi
4 GUIDED WAVE STRUCTURAL MONITORING OF
COMPOSITE WING SKIN-TO-SPAR BONDED
JOINTS IN AEROSPACE STRUCTURES ......................................... 124
4.1 Introduction ....................................................................................... 124

4.2 Problem statement and modeling formulation ................................. 128
4.3 Dispersion results for different bond states ..................................... 137

4.4 Identification of Carrier Modes......................................................... 144
4.5 Strength of Transmission as a Function of Bond Condition .......... 147

4.6 Experiments on simulated wing skin-to-spar joints ......................... 154
4.6.1 Test Specimens ........................................................................... 154

4.6.2 Experimental setup and procedure ............................................ 156
4.6.3 Experimental results................................................................... 157

4.7 Summary and conclusions ................................................................. 160
4.8 Acknowledgements ............................................................................. 163
5 GUIDED WAVE STRUCTURAL MONITORING OF RAILROAD

TRACKS (MODELLING).................................................................... 164
5.1 Introduction ....................................................................................... 164

5.2 Finite element modeling with application
to Lamb waves in a plate............................................................... 167
5.3 Application to rails ........................................................................... 178
5.4 Conclusions ........................................................................................ 188
6 GUIDED WAVE STRUCTURAL MONITORING OF RAILROAD

TRACKS (EXPERIMENT) ............................................................... 191
6.1 Introduction ....................................................................................... 191
6.2 Defect detection using hammer generation/air-coupled detection ... 194
vii
6.2.1 Effect of Sensor Inclination ....................................................... 194
6.2.2 Effect of Sensor Lift-off Distance .............................................. 199

6.2.3 Transmission Approach .............................................................. 200
6.3 Use of the Discrete Wavelet Transform to Enhance
Defect Detection................................................................................. 202
6.3.1 Defect detection using laser generation/air-coupled detection.. 206

6.3.2 Transverse Crack at the Top of Rail Head ............................. 207
6.3.3 Transverse Crack at the Gage Corner of the Rail Head ........ 209
6.4 Conclusions ........................................................................................ 215
7 GUIDED WAVE STRUCTURAL MONITORING

OF MULTI-WIRE STRANDS ........................................................... 217
7.1 Introduction ....................................................................................... 217
7.2 Wave propagation in helical waveguides.......................................... 221
7.2.1 Geometry of the strand ............................................................. 221

7.2.2 Dispersion properties of unloaded strands................................. 223
7.2.3 Interwire contact stresses in axially
loaded seven-wire strands.......................................................... 230

7.2.4 SAFE dispersion properties of loaded strands
(pre-twisted waveguide)............................................................. 239
7.2.5 Forced solution for the free strand ........................................... 245

7.2.6 Forced solution for the free strand: effect of anchorages ........ 251
7.3 Experimental Study of Prestress Level Monitoring
in Free Strands (Selection of Optimum Wave

Features and Sensor Lay-out) ........................................................... 256
7.3.1 Test Protocols ............................................................................ 260
viii
7.3.2 Results – Test 1 ........................................................................ 261
7.3.3 Results – Test 2 ........................................................................ 266

7.3.4 Results – Test 3 ........................................................................ 268
7.4 Defect detection in strands ............................................................... 276
7.4.1 Experimental setup and procedure ............................................ 276
7.4.2 Statistical defect classification ................................................... 280

7.4.3 Defect detection results – “low” noise ....................................... 282
7.4.4 Defect detection results – “high” noise...................................... 282
7.5 Discussion and Conclusions............................................................... 283
8 CONCLUSIONS AND RECOMMENDATIONS

FOR FUTURE STUDIES ................................................................... 288
8.1 Conclusions ........................................................................................ 288
8.2 Recommendations for future studies................................................. 294
APPENDIX A…………..…………………………………………………………297
REFERENCES…………..…………………………………………………………311
ix
LIST OF TABLES
Table 3.1: Elastic properties for the T300/914 laminate

(elastic constants in GPa)........................................................ 64
Table 3.2: Elastic and viscous properties of the orthotropic plate

examined by Neau (2003) and in the present study
(elastic constants in GPa, Viscosities given at 2.0 MHz). ..... 65
Table 3.3: Elastic properties of the T800/924 lamina (elastic

constants in GPa). ................................................................... 65
Table 3.4: Elastic and viscous properties of the orthotropic plate

examined by Deschamps (1992) and in the present
study (elastic constants in GPa, Viscosities given
at 2.242 MHz). ......................................................................... 65
Table 3.5: Elastic and viscous properties for the UAV wing
skin-to-spar joint (elastic constants in GPa.
*Neau et al., 2002)................................................................... 74
Table 3.6: Bulk ultrasonic velocities and attenuations of the

UAV wing skin-to-spar interface layer. ................................... 74
Table 3.7: Axisymmetric waveguides geometrical and

acoustic material properties...................................................... 92
Table 3.8: Cut-off frequencies for a 4 in. 40 ANSI steel pipe in

the range 0-1000 kHz for an increasing number
of finite elements nel used in the SAFE mesh.
Analytical solution is obtained by a SPBW formulation. ...... 93
Table 3.9: Guided modes wavenumbers for a 4 in. 40 ANSI steel

pipe at 1 MHz for increasing number of finite elements
nel used in the SAFE mesh. Analytical solution is
obtained by a SPBW formulation from Seco et al. (2002) ... 94
Table 4.1: Real and imaginary stiffness coefficients, geometric

and physic properties for bonded multilayer models ............ 135
Table 4.2: Ultrasonic bulk longitudinal and shear velocities
x
and material attenuations for the bond. ............................... 135
Table 7.1: Properties of the seven-wire strand tested on

the SATEC machine (http://www.dsiamerica.com
/products/MultistrandSystem.html)........................................ 257
Table 7.2: Test protocols for prestress level monitoring in

seven-wire strand (designation of ultrasonic sensors
refers to Figure 7.15(b)-(c)). .................................................. 261
Table 7.3: Results of defect detection from outlier analysis:

number of outliers n/300 for the various damage
sizes and two levels of noise.................................................. 283
xi
LIST OF FIGURES
Figure 2.1: Guided plate waves in an isotropic homogeneous plate. .......... 9
Figure 2.2: Bulk wave in a 3D space......................................................... 14
Figure 2.3: Homogeneous elastic layer of infinite in-plane dimensions...... 15
Figure 2.4: Wavenumber dispersion curves for Lamb

waves in an aluminum plate (h=half-thickness of
the plate, cL=6370 m/sec, cT=3160 m/sec,
ρ=2770kg/m3). ......................................................................... 22
Figure 2.5: (Phase velocity c) dispersion curves for Lamb

Waves in an aluminum plate (cT=shear wavespeed) ............. 23
Figure 2.6: Displacement field of the antisymmetric mode

A0 at the frequency f=2MHz for an aluminum
plate with thickness 2h=1mm. ............................................... 23
Figure 2.7: Displacement field of the symmetric mode S0 at the

frequency f=2MHz for an aluminum plate with
thickness 2h=1mm. The displacement is symmetric. ............ 24
Figure 2.8: First two symmetric mode shapes (A0, A1) at various
products of frequency and half thickness showing
how modeshape curvature increases with frequency and
mode number ............................................................................ 24
Figure 2.9: First two symmetric mode shapes (S0, S1) at various
products of frequency and half thickness showing
how modeshape curvature increases with frequency and
mode number. ........................................................................... 25
Figure 2.10: Group velocity cg dispersion curves for Lamb Waves

in an aluminum plate (cT=shear wavespeed).......................... 26
Figure 3.1: (a) Waveguide of cross-section Ω and (b) generic element

of surface Ωe. (c) Discretization of an infinite plate
and (d) degrees of freedom of the mono-dimensional
xii
three-node element.................................................................... 42
Figure 3.2: Dispersion results (Lamb modes) for a 12.7mm

thick, viscoelastic HPPE plate in vacuum: (a) phase
velocity, (b) energy velocity, (c) attenuation below
500 Np/m, (d) attenuation below 3500 Np/m. ...................... 59
Figure 3.3: Dispersion results (SH modes) for a 12.7mm thick,

viscoelastic HPPE plate in vacuum: (a) phase velocity,
(b) energy velocity, (c) attenuation up to 500 Np/m,
(d) attenuation up to 3500 Np/m........................................... 60
Figure 3.4: Dispersion results for a 1mm thick, elastic

transversely-isotropic carbon-epoxy laminate in vacuum:
(a) phase velocity for waves propagating along
principal direction 1, (b) group velocity for waves ................ 66
Figure 3.5: Dispersion results for a 3.6mm thick, viscoelastic

orthotropic carbon-epoxy plate in vacuum for
waves propagating at 60deg from principal direction
1: (a) phase velocity for the symmetric modes ...................... 67
Figure 3.6: Dispersion results for a 1mm thick, elastic

[±45/0/90]S composite laminate in vacuum for
waves propagating along principal direction 1 of the
0deg lamina: (a) phase velocity, (b) group velocity............... 71
Figure 3.7: Dispersion results (Lamb modes) for a 1mm thick,

viscoelastic orthotropic plate in vacuum: (a), (b) and (c)
case of hysteretic viscoelastic model; (d), (e) and (f)
case of Kelvin-Voigt viscoelastic model. .................................. 72
Figure 3.8: Dispersion results (SH modes) for a 1mm thick,

viscoelastic orthotropic plate in vacuum: (a), (b) and (c)
case of hysteretic viscoelastic model; (d), (e) and (f)
case of Kelvin-Voigt viscoelastic model. .................................. 73
Figure 3.9: Dispersion results for UAV wing skin-to-spar adhesive

joint with a disbonded interface for waves
propagating perpendicularly to the spar lengthwise
direction: (a) phase velocity, (b) energy velocity, .................. 75
Figure 3.10: Dispersion results for a 115-lb A.R.E.M.A., viscoelastic

rail for waves propagating along the rail running
direction: (a) phase velocity, (b) energy velocity,
xiii
(c) attenuation. ......................................................................... 78
Figure 3.11: Dispersion results for a 115-lb A.R.E.M.A., viscoelastic

rail: phase velocity for frequencies below 10 kHz, and first
five cross-sectional mode shapes at 5 kHz. mode m1,
mode m2, mode m3, mode m4, mode m5. ............ 79
Figure 3.12: (a) Notation and mesh for the SAFE model of the pipe;
(b) phase velocity dispersion curves; (c) energy
velocity dispersion curves; (d) attenuation curves .................. 80
Figure 3.13: (a) Schematic representation of a multilayered hollow

cylinder, (b) representation of the through-thickness
mono-dimensional finite element mesh, (c) e-th
mono-dimensional quadratic finite element with three ........... 90
Figure 3.14: Torsional modes for a copper ( ρ = 8900 kg m−3 ,

cT = 2240 m/s , κT = 0 Np /λ ) pipe with inner radius
6.80 mm and wall thickness 0.70 mm, filled with
bitumen. (a)-Phase velocity, (b)-energy velocity and ............. 97
Figure 3.15: T(0,2) guided mode for a copper ( ρ = 8900 kg /m3 ,

cT = 2240 m /s , κT = 0 Np /λ ) pipe, with inner radius
6.80mm and wall thickness 0.70mm, filled with
elastic bitumen ( ρ = 970 kg m−3 , cT = 430 m s-1 ........................ 99
Figure 3.16 Torsional Td(0,1) mode phase velocity (a) and attenuation
(b) dispersion curves ( • • • • ) for a copper
( ρ = 8900 kg /m3 , cT = 2240 m /s , κT = 0 Np / λ ) pipe,
with inner radius 6.80 mm and wall thickness ................... 101
Figure 3.17: Torsional Td(0,1) mode for a copper ( ρ = 8900 kg /m3 ,

cT = 2240 m / s , κT = 0 Np / λ ) pipe with inner radius
6.80 mm and wall thickness 0.70 mm filled with
viscoelastic bitumen ( ρ = 970 kg /m3 , cT = 430 m / s , ............. 103
Figure 3.18: Longitudinal modes for an elastic ( ρ = 7800 kg /m3 ,

cL = 5900 m /s , cT = 3190 m /s ) 4-in. scheduled ANSI 40
steel pipe coated with 0.1524 mm of viscoelastic
bitumen ( ρ = 1500 kg /m3 , cL = 1860 m /s , cT = 750 m /s ,........ 105

cL = 5900 m / s , cT = 3190 m / s ) 4-in. scheduled
ANSI 40 steel pipe coated with 0.1524 mm
of viscoelastic bitumen ( ρ = 1500 kg /m3 ,................................ 107
xiv
Figure 3.20 (a) Seven wire strand embedded in grout and concrete
block. (b) Section of the specimen. (c) Three layer
system considered in the SAFE model.................................. 109
Figure 3.21: Phase velocity (a), energy velocity (b) and

attenuation (c) dispersion curves for the Longitudinal
L(0,i) guided modes in a three-layer waveguide:
viscoelastic steel bar ( ∅ = 15.24 mm , ρ = 7932 kg /m3 , .......... 110
Figure 3.22 (a) Lowest phase velocity Longitudinal mode L(0,low-cph)

and (b) lowest attenuation Longitudinal mode
L(0,low-att) in the three-layer steel-grout-concrete
viscoelastic waveguide at 305 kHz ....................................... 111
Figure 3.23: (a) Deformation of a plate subjected to a uniform

harmonic vertical load (b), (c). ............................................. 118
Figure 3.24: (a) Deformation of a plate subjected to a uniform

harmonic horizontal load (b), (c). ......................................... 119
Figure 3.25: (a) Deformation of a bonded joint subjected to a

uniform harmonic vertical load (b). ...................................... 119
Figure 4.1: (a) Wing skin-to-spar bonded assembly. (b) Across bond
and within bond test configuration ....................................... 129
Figure 4.2 Schematic of wave propagation for (a) “within bond”

test configuration, (b) “across bond” test configuration.
(c) Diagram of SAFE model and beam element adopted.... 131
Figure 4.3: (a) Phase velocity, (b) energy velocity, (c) and
attenuation dispersion results for well bonded
[0/±45/0]S composite plate to composite spar joint.
Wave propagation is along 90 degree direction .................... 137
Figure 4.4: (a) Phase velocity dispersion curves for the [0/±45/0]S
skin-to-spar joint (“across the bond” testing
configuration) with damping and without damping.
Through-thickness Poynting vector for S0 at (b) 155 kHz. . 141
Figure 4.5: (a) Phase velocity, (b) energy velocity, and
xv
(c) attenuation curves for the [0/±45/0]S skin-to-spar
joint with a disbonded interface and for the single
[0/±45/0]S plate (“across the bond” testing configuration)... 142
Figure 4.6: The [0/±45/0]S skin-to-spar joint in the “across the

bond” testing configuration: (a) displacement mode shapes
at 205 kHz in the properly-cured bond, (b) strain profiles
for modes S0 and A1 at 205 kHz in the properly-cured ...... 145
Figure 4.7: Px component of the Poynting vector in the

properly-cured bond and in the poorly-cured bond for
(a) S0 at 155 kHz, (b) S0 at 205 kHz, (c) A1 at
155 kHz, and (d) A1 at 205 kHz. Same quantity ................ 149
Figure 4.8: Above spar power flow of S0 and A1 modes for well
cured and poorly cured cases. 90 deg. wave
propagation direction in (a) the [0/±45/0]s and
(b) the [0/±45/90]s bonded composite plate. ........................ 152
Figure 4.9: [0/±45/0]s plate bonded to spar (a), top view (b),
side view ................................................................................. 155
Figure 4.10: Typical waveforms from the “within the bond” test
of the [0/±45/0]S skin-tospar specimen at 205 kHz.............. 158
Figure 4.11: Normalized RMS difference results for each defect in

“across bond” configuration of (a), [0/±45/0]s
(b), [0/±45/90]s and for “within bond” configuration
of (c), [0/±45/0]s (d), [0/±45/90]s test specimen.................. 159
Figure 5.1: ABAQUS model of the plate with meshes examined. .......... 171
Figure 5.2: Abaqus simulations of Lamb waves in a plate excited

by an impulse. (a) Case of the S0 mode; (b) case of
the A0 mode. .......................................................................... 172
Figure 5.3: Group velocity dispersion curves extracted from the

model for varying mesh refinements. Plot also shows
the Rayleigh-Lamb exact solution. ........................................ 173
Figure 5.4: Energy reflection coefficients in a plate from a
xvi
perpendicular edge (a and b) and from a 35 deg
inclined edge (c through f). ................................................... 177
Figure 5.5: Geometry of the rail studied with the different transverse
head defects. ........................................................................... 178
Figure 5.6: Experimental setup for measuring reflections of the

vertical bending mode from transverse head defects. ........... 179
Figure 5.7: Finite element mesh employed for the rail. .......................... 182
Figure 5.8: Time histories and GWT scalograms of the vertical

acceleration at the top of the rail head from the
simulation (a and c) and from the experiment
(b and d). Resulting group velocity shown in (e)................ 183
Figure 5.9: Same-mode reflection coefficients for the vertical bending

mode from the perfectly-transverse defects (a), the
20 deg inclined defects (b), and the 35 deg inclined
defects (c). .............................................................................. 184
Figure 5.10: Mode-converted reflection coefficients for the vertical

bending mode from the perfectly-transverse defects (a),
the 20 deg inclined defects (b), and the 35 deg
inclined defects (c). ................................................................ 187
Figure 6.1: System for detecting transverse cracks in rails by

generating ultrasonic guided waves with a hammer and
detecting the waves with a pair of air-coupled sensors...... 193
Figure 6.2: Detection of transverse crack by reflection

measurements (detection angle 0º). (a) time history;
(b) continuos wavelet transform; (c) wavelet peaks
at 30 kHz................................................................................ 196
Figure 6.3: Snell’s law of refraction applied to air-coupled detection

of guided waves. ..................................................................... 197
Figure 6.4: Formation of resonances in the lift-off air gap. .................... 197
Figure 6.5: Variation of the reflection coefficient for transverse
xvii
crack (100% Head area reduction) as a function of
detection angle. ....................................................................... 198
Figure 6.6: Detection of transverse crack by transmission

measurements (detection angle 0º). (a) and (b) time
histories; (c) and (d) continuous wavelet scalograms. .......... 201
Figure 6.7: Transmission coefficient for transverse crack at 30 kHz

as a function of position of the sensor pair relative
to the defect. .......................................................................... 203
Figure 6.8: Use of the discrete wavelet transform for signal de-noising.
(a) signal reconstructed from level 6; (b) signal
after thresholding wavelet coefficients at level 6;
(c) transmission coefficient after wavelet processing............. 204
Figure 6.9: Laser/air-coupled system for detecting small transverse

cracks. (a) and (b) experimental setup; (c) and (d)
types of defects considered..................................................... 209
Figure 6.10: Detection of 5 mm crack by laser/air-coupled

system (detection angle 0°). (a) and (b) time histories;
(c) and (d) continuous wavelet scalograms; (e) and
(f) de-noised signals after discrete wavelet processing. ......... 210
Figure 6.11: Detection of 1 mm crack by laser/air-coupled

(c) and (d) continuous wavelet scalograms; (e) and (f)
de-noised signals after discrete wavelet processing................ 211
Figure 6.12: Reflection coefficient spectra for horizontal crack as a

function of crack depth (detection angle 6°). ....................... 212
Figure 6.13: Detection of 1-mm corner crack by laser/air-coupled

(c) and (d) continuous wavelet scalograms; (e) and
(f) de-noised signals after discrete wavelet processing. ......... 213
Figure 6.14: Reflection coefficient spectra for corner crack as a

function of crack depth (detection angle 6°). ....................... 214
xviii
Figure 7.1: Geometry of the strand: helical wire wrapped around
a core. ..................................................................................... 222
Figure 7.2: (a) Core wire modeled with SAFE. (b) Acceleration
time history obtained with SAFE numerical simulation
for a steel circular bar (E=195GPa, ν=0.29,
ρ=7700kg/m3, Length=720mm) subjected to ........................ 227
Figure 7.3: (a) Measured time waveform from Laser Ultrasound

testing of single steel wire (central core,
Length=720mm). (b) Normalized (Complex Morlet)
Continuous Wavelet Transform scalogram of time............... 228

testing of single steel wire (helical wire, Length=720mm).
(b) Normalized (Complex Morlet) Continuous
Wavelet Transform scalogram of time waveform.................. 229

testing of unloaded strand. Guided wave excited in
central wire and received in the same central wire
(Test C-C, Strand length=640mm). (b) Normalized ............ 231

testing of unloaded strand. Guided waves excited in
central wire and received in helical wire (Test C-P,
Strand length=640mm). (b) Normalized (Complex............... 232
Figure 7.7: (a) Axial forces, bending and twisting moments acting
on the helical and central wires of a loaded strand.
(b) Equilibrium of force in an element of a helical wire.
(c) Resultant contact force in the transverse cross .............. 236
Figure 7.8: Reference system of Hertz contact problem .......................... 238
Figure 7.9: Three-wire contact problem studied by ABAQUS to

compute the contact stresses in the seven-wire strand
induced by axial loads. (a) Schematic, (b) FE model
with Von Mises equivalent stresses for axial load................ 238
Figure 7.10: (a) Pretwisted waveguide representing a seven-wire strand;

(b) generic element of the surface Ω; (c) Phase velocities
xix
and mode shapes at 100 kHz of the strand. ........................ 244
Figure 7.11: (a) Guided waves excited in seven wire strand by a

PZT actuator and detected with PICO sensors at the
end. (b) Mesh of the strand cross section employed in
SAFE model............................................................................ 247
Figure 7.12: SAFE predictions of the RMS spectrum of the

waveforms recorded by (a) the Pico Central sensor,
(b) the Pico Peripheral sensor and generated by the
PZT actuator. (c) Ratio between RMS spectra of............... 248
Figure 7.13: SAFE simulation of leakage of guided waves into the

strand anchorage: (a) guided wave excited by the
PZT actuator and received by the PICO sensors.
(b) Mesh of the strand cross section coupled to the ........... 253
Figure 7.14: Strand with anchorage: SAFE predictions of the

RMS spectrum of the waveforms recorded by (a) the
Pico Central sensor, (b) the Pico Peripheral sensor
and generated by the PZT actuator. (c) Ratio between ..... 255
Figure 7.15: (a) The 1.82-m, seven-wire strand installed in the

SATEC testing machine for stress monitoring tests;
(b) ultrasonic sensor lay-out; (c) pictures of the
piezoelectric transmitter (PZT 3) on the peripheral wire .... 258
Figure 7.16: Magnetostrictive coil sensor probing entire strand................ 259
Figure 7.17: Energy leakage between central wire and peripheral wire
as a function of applied prestress. Test 1: PICO (C)
transmitting and PZT 3 receiving. Load and unload
ramps. 100% load = 70% U.T.S. Low frequency ................. 263
Figure 7.18: Energy leakage between central wire and peripheral

wire as a function of applied prestress.
Test 1: PICO (C) transmitting and PZT 3 receiving.
Load and unload ramps. 100% load = 70% U.T.S. ............. 264
Figure 7.19: Shift in peak frequency as a function of applied

prestress. Test 1: PICO (C) transmitting, MsS coil
receiving. Load and unload ramps for two cycles.
xx
Max load 41 kips = 70% U.T.S. Frequency range .............. 265
Figure 7.20: Energy leakage between peripheral wire and central

wire as a function of prestress. Test 3: PZT 3
transmitting, PICOs (C) and (P) receiving. Load and
unload ramps. 100% load = 70% U.T.S. Low frequency..... 267
Figure 7.21: Test 3 – Peripheral wire - central wire energy leakage

(low frequencies). Same feature as in Figure 7.20
for two additional load-unload cycles. ................................... 269
Figure 7.22: Energy leakage between peripheral wire and central wire
as a function of prestress. Test 3: PZT 3 transmitting,
PICOs (C) and (P) receiving. Load and unload ramps.
100% load = 70% U.T.S. Low frequency range ................... 272
Figure 7.23: Energy leakage between peripheral wire and central wire
as a function of prestress. Test 3: PZT 3 transmitting
and PICO (C) receiving. Load and unload ramps.
100% load = 70% U.T.S. Low frequency range ................... 273

prestress. Test 3: PZT 3 transmitting, PICO (C)
receiving. Load and unload ramps for three cycles.

prestress. Test 3: PZT 3 transmitting, PICO (C)
receiving. Load and unload ramps for two cycles.
Figure 7.26: Experimental setup for defect detection in a strand

using reflections of guided waves excited and detected
by magnetostrictive transducers (dimensions in mm)........... 276
Figure 7.27: Signals reconstructed after pruning the DWT coefficients

at the first six decomposition levels...................................... 277
Figure 7.28: Components of the Damage Index vector measured from

the variance and from the root-mean-square of
the thresholded wavelet coefficients at the sixth
decomposition level. ................................................................ 284
xxi
Figure 7.29: Mahalanobis squared distance for the baseline
(undamaged) and damaged strand data corrupted
with the low-level noise (a) and the high-level noise (b). ... 285
Figure A.1: (a) Strain Waveform predicted in a 800 mm-long

aluminum plate showing the first arrival of the dispersive
S0 wave. (b) Complex Morlet Wavelet transform
scalogram of time waveform identifying arrival .................... 302
Figure A.2: (a) Strain Waveform predicted in a 800 mm-long

aluminum plate showing the first arrival of the dispersive
A0 and A1 modes. (b) Complex Morlet Wavelet
transform scalogram of time waveform identifying ............... 303
Figure A.3: (a) Wavelet decomposition by filter bank tree;

(b) signal reconstruction from wavelet coefficients;
(c) reconstruction of original signal.. ..................................... 307
Figure A.4: (a) Signal in seven-wire strand after 500 averages;

(b) signal with no averages; (c) reconstructed signal
after pruning the DWT coefficients at the
first six decomposition levels.................................................. 308
xxii
ACKNOWLEDGEMENTS
The research for this dissertation was performed at the University of
California at San Diego under the direction of Professor Francesco Lanza
di Scalea. I would like to express my deepest and most genuine gratitude
to him. I am especially grateful for the opportunity to perform this
research under his guidance. His incessant help and academic passion
allowed me to achieve what I have today.
I would like to thank my committee members Prof. Thomas Bewley,
Prof. Charles Farrar, Prof. William Hodgkiss and Prof. Enrique Luco for
their personal commitments, patience and technical contributions.
I wanted also to thank Prof. Erasmo Viola and Prof. Alessandro
Marzani that strongly encouraged me to come to San Diego and always
helped me in my research. Thanks to all of my colleagues present and
past in the NDE/SHM laboratory: Prof. Piervincenzo Rizzo, Stefano
Coccia, Salvatore Salamone, Elisa Sorrivi, Giusi Vitale, Robert Ronal
Phillips, Ankit Srivastava, Jake Finkler and Gaetano Restivo for their
invaluable contributions, humor and friendship. I am grateful to all my
roommates and friends, a consistent portion of the SE Department, that in
these years have always helped me in difficult moments: Michele Barbato,
Maurizio Gobbato, Chiara Casarotti, Carmen Amaddeo, Ozgur Ozcelick
and Giuseppe Canducci.
A special thanks goes to the people of SRMD laboratory in
particular Gianmario Benzoni, Donato Innamorato and Edward Stovin for
their continuous help during the tests on the rail-road prototype.
xxiii
Lastly, I must thank my loving family. Michela, Gabriella and
Domenico were always here with me and I never felt the distance that
separates us.
The research presented within this thesis was partially funded by
the Los Alamos/UCSD Education Collaboration Task 2 “Structural
Integrity Monitoring of UAV Composite Wings” and by the UCSD/Los
Alamos Cooperative Agreement on Research and Education (CARE). In
particular I would like to thank Prof. Charles Farrar for his assistance and
support.
The work has being also funded by the U.S. Federal Railroad
Administration (grant# DTFR53-02-G-00011). The in-kind support of
Mahmood Fateh, Program Manager and Mr. Gary Carr, Chief of the FRA
Track Research Division is gratefully acknowledged.
This research was finally supported by the California Department of
Transportation under contract # 59A0538. In particular I would like to
thank Dr. Charles Sikorsky, CALTRANS Program Manager.
Chapter 3, in part, has been published in the Journal of Sound and
Vibration, Bartoli, Ivan; Marzani, Alessandro; Lanza di Scalea, Francesco;
Viola, Erasmo (2006). The title of this paper is “Modeling wave
propagation in damped waveguides of arbitrary cross-section”. The
dissertation author was the primary investigator and primary author of
this paper.
Chapter 4, in part, has been published in the Journal of Acoustical
Society of America, Matt, Howard; Bartoli, Ivan; Lanza di Scalea,
xxiv
Francesco (2005). The title of this paper is “Ultrasonic guided wave
monitoring of composite wing skin-to-spar bonded joints in aerospace
structures”. The dissertation author was co-author of this paper.
xxv
VITA
2001 Bachelor of Science, University of Bologna, Italy
2002 – 2003 Research Assistant, University of Bologna, Italy
2006 Master of Science, University of California, San Diego
2003 – 2007 Research Assistant, University of California, San Diego
2007 Doctor of Philosophy, University of California, San Diego
PUBBLICATIONS
Journals
Viola, E., Bartoli, I. and Marzani, A. (2003) “Boundary Conditions Effect

on the Dynamic Instability of Orthotropic Plates,” The Journal Scientific
Israel-Technological- Advantages, Vol. 5, No. 1, pp. 66-83.
Bartoli, I., Lanza di Scalea, F., Fateh, M. and Viola. E. (2005) “Modeling
guided wave propagation with application to the long-range defect
detection in rail road tracks,” NDT & E International, Vol. 38, pp. 325-
334.
Rizzo, P., Bartoli, I., Marzani A., and Lanza di Scalea, F. (2005) “Defect
Classification in Pipes by Neural Networks using Multiple Guided
Ultrasonic Wave Features,” ASME Journal of Pressure Vessel Technology,
Special Issue on the Nondestructive Evaluation of Pipeline and Vessel
Structures, Vol. 127, No. 3, pp. 294-303.
Lanza di Scalea, F., Rizzo, P., Coccia, S., Bartoli, I., Fateh, M., Viola, E.
and Pascale, G. (2005) “Non-Contact Ultrasonic Inspection of Rails and
Signal Processing for Automatic Defect Detection and Classification,”
Insight - Non-Destructive Testing & Condition Monitoring, Vol. 47, No. 6,
pp. 346-353.
xxvi
Lanza di Scalea, F., Bartoli, I., Rizzo, P. and Fateh, M. (2005) “High-
speed Defect Detection in Rails by Non-contact Guided Ultrasonic
Testing,” Transportation Research Record, Journal of the Transportation
Research Board, No. 1916, pp. 66-77.
Matt, H., Bartoli, I. and Lanza di Scalea, F. (2005) “Ultrasonic guided

wave monitoring of composite wing skin-to-spar bonded joints in aerospace
structures,” Journal of the Acoustical Society of America, Vol. 118, pp.
2240-2252.
Bartoli, I., Marzani, A., Lanza di Scalea, F., and Viola, E. (2006)
“Modeling wave propagation in damped waveguides of arbitrary cross-
section,” Journal of Sound and Vibration, Vol. 295 No. 3-5, pp. 685-707.
Lanza di Scalea, F., Rizzo, P., Coccia, S., Bartoli, I., and Fateh, M.,
(2006) “Laser-Air-Coupled Hybrid Detection in Rail Tracks: Status of FRA
Prototype Development at UC San Diego,” Transportation Research
Record, Journal of the Transportation Research Board, No. 1943, pp. 57-
64.
Lanza di Scalea, F., Matt, H. and Bartoli, I. (2007) “The response of

rectangular piezoelectric sensors to Rayleigh and Lamb ultrasonic waves,”
Journal of the Acoustical Society of America, Vol. 121 pp. 175-187.
Rizzo, P., Bartoli, I., Cammarata M. and Coccia S. (2007) “Digital signal
processing for rail monitoring by means of ultrasonic guided waves,”
Insight - Non-Destructive Testing & Condition Monitoring, Vol. 49, No. 6,
pp. 327-332.
Lanza di Scalea, F., Matt, H., Bartoli, I., Coccia, S., Park, G. and Farrar,
C. (2007) “Health Monitoring of UAV wing skin-to-spar joints using guided
waves and macro fiber composite transducers,” Journal of Intelligent
Material Systems and Structures, Vol. 18, 373-388.
Lanza di Scalea, F., Bartoli, I., Rizzo, P., Marzani, A., Sorrivi, E. and
Viola, E., “Structural Health Monitoring of Multi-wire Strands,” in Chapter
10 (D. Inman, ed.) of Encyclopedia of Structural Health Monitoring, C.
Boller, F-K. Chang and Y. Fujino, eds., Johns Wiley & Sons, Chichester,
UK, submitted (2007).
Rizzo, P., Coccia, S., Bartoli, I. and Lanza di Scalea, F. “Non-Contact

Rail Monitoring by Ultrasonic Guided Waves,” in Chapter 10 (D. Inman,
ed.) of Encyclopedia of Structural Health Monitoring, C. Boller, F-K.
Chang and Y. Fujino, eds., Johns Wiley & Sons, Chichester, UK,
submitted (2007).
xxvii
Marzani, A., Viola, E., Bartoli, I., Lanza di Scalea, F. and Rizzo, P. “A
semi-analytical finite element formulation for modeling stress wave
propagation in axisymmetric damped waveguides,” submitted to Journal of
Sound and Vibration (2007).
xxviii
ABSTRACT OF THE DISSERTATION
Structural Health Monitoring by ultrasonic guided waves
by
Ivan Bartoli
Doctor of Philosophy in Structural Engineering
University of California, San Diego, 2007

Professor Francesco Lanza di Scalea, Chair
Guided ultrasonic waves provide a highly efficient method for the

non-destructive evaluation (NDE) and the structural health monitoring
(SHM) of solids with finite cross-sectional dimensions (waveguides).

Compared to the widely used ultrasonic bulk waves, guided waves provide
larger monitoring ranges and the complete coverage of the waveguide
cross-section. Compared to global vibrations, guided waves offer increased

sensitivity to smaller defects due to the smaller wavelengths involved.
These advantages can be fully exploited only once the complexities of
guided wave propagation (multimode, dispersion, frequency-dependent

attenuation) are unveiled and managed for the given test structure.
This doctoral dissertation is aimed at developing a Semi-Analytical
Finite Element (SAFE) method for modeling wave propagation in

waveguides of arbitrary cross-section. The method requires the finite
element discretization of the cross-section of the waveguide, and assumes
xxix
harmonic motion along the wave propagation direction. The general SAFE
technique was extended to account for viscoelastic material damping by
allowing for complex stiffness matrices for the material. The dispersive
solutions are obtained in terms of phase velocity, group velocity (for
undamped media), energy velocity (for damped media), attenuation, and
cross-sectional mode shapes. Once the dispersive properties are computed,

the wave motion can be interpreted and the forced response can be
predicted.
The proposed SAFE formulation was applied to enhance the use of

ultrasonic guided waves in a number of applications and to interpret the
corresponding experimental results.
The following three applications were considered:
• Defect detection in adhesively-bonded joints found in the wing skin-
to-spar assemblies of Unmanned Aerial Vehicles;
• Defect detection in railroad tracks;
• Load monitoring and defect detection in seven-wire steel strands
used in cables and prestressed concrete structures.
xxx
1 INTRODUCTION
1.1 Research motivation
Ultrasonic testing is one of the most widely used methods today for
Non-Destructive Evaluation (NDE) and Structural Health Monitoring
(SHM). Although stress wave theory can be quite complex, ultrasonic NDE
popularity descends from the simplicity of the basic concepts behind it.
Traditional ultrasonic testing, based on shear and longitudinal (bulk)
waves, is performed locally on specimens of small dimensions or by
scanning the ultrasonic sources and receivers across large structural
components. As a consequence, while the sensitivity of ultrasonic waves to
small flaws and defects is unanimously recognized, traditional ultrasonic
inspections are quite time consuming.
Ultrasonic signals, depending on the inspected materials, can be
detected dozens of meters away from the source, particularly in elongated
structures such as pipes, railroad rails, plates and beams. Waves
propagating in such structures are called “guided waves”.
Guided wave NDE/SHM techniques, owing to extensive research in
the 1990’s, have grown to be a reality outside the confines of academic
research (Ditri and Rose 1992, Cawley and Alleyne, 1996). For example, in
the pipeline industry, ultrasonic guided waves represent a fast and
practical method of conducting inspection. Principal advantages, besides
allowing for long range inspection, are the complete coverage of the
1
2
waveguide cross-sections and an increased sensitivity to small defects when
compared to global vibrations owing to the ~kHz probing frequencies.
Guided waves are complex because of the existence of multiple
modes at any given frequency, the frequency-dependent velocities
(dispersion) and the frequency-dependent attenuation. In order to avoid
errors in the interpretation of the guided wave signals and to fully exploit
their potential, various signal-processing techniques have been used as
Short Time Fourier Transform (Gabor, 1946), Continuous Wavelet

Transform (Daubechies, 1992) and Two-Dimensional Fourier Transform
(Alleyne and Cawley, 1991). Yet, accurate prediction of guided wave
modal and forced solutions is still indispensable. Dispersion properties as
phase velocities and group/energy velocities are important for mode
identification. Similarly, the knowledge of the mode attenuation helps
maximizing the inspection range by exploiting modes associated to
minimum energy attenuation.
The main object of the present dissertation is to develop a Semi-
Analytical Finite Element (SAFE) method for predicting modal and forced
guided wave solutions in a computationally efficient manner. Also referred
to as Spectral Finite Element (SFE), the SAFE scheme describes the wave
propagation displacement field by coupling a finite element discretization of

the waveguide cross-section with harmonic exponential functions along the
wave propagation direction. Compared to standard three-dimensional finite
element (FEM) approaches, the SAFE method allows reducing of one order
the numerical dimension of the problem. Compared to analytical wave-
based methods, such as Superposition of Partial Bulk Wave (SPBW)
methods (Lowe, 1995), the SAFE method presents a wider spectrum of

3
applicability since it can operate on waveguides with arbitrary cross-
sectional geometries, for which theoretical solutions can be unavailable. In
addition, the SAFE method does not suffer from modeling waveguides with
a large number of layers, as in the case of composite laminates, for which
the determination of the dispersive properties via SPBW methods becomes
numerically challenging.
In the present study the SAFE method is extended to model
waveguides of any shape and introducing material damping. Viscoelastic

materials are included by considering constitutive linear viscoelastic
equations in the development of the governing equations. The damped
formulation allows the model to represent the true guided wave
attenuation as a function of frequency. This extension is particularly
relevant for NDE/SHM applications that are described in the remaining
part of the dissertation. Prediction of the modes attenuation is especially
crucial in high-loss materials such as viscoelastic fiber-reinforced polymer
composites. Forced solutions are carried out to predict the arrival times of
wave packets, to maximize the excitation of specific guided modes sensitive
to particular defects and to improve the detection capability of sensors
and sensor arrays.
Applications of the SAFE technique are presented for three cases

that can all benefit from robust structural monitoring systems. These are
the monitoring of adhesive-bonds in aerospace structures, the monitoring of
railroad tracks and that of multi-wire strands found in civil structures.

4
1.2 Outline of the dissertation
The thesis is structured as follows. In Chapter 2 an introduction to
the topic of ultrasonic guided waves is provided. The chapter begins with
a description of ultrasonic bulk waves in unbound media. The main
equations describing harmonic motion associated to the propagation of
longitudinal and shear waves are provided. The attention is then focused
on the propagation of ultrasonic waves in bounded media, beginning from
the simple case of isotropic plates and giving a survey of the analytical
and numerical techniques typically adopted for the description of guided
wave motion.
Chapter 3 is dedicated to the study of ultrasonic guided waves in
damped waveguides of arbitrary cross section by the SAFE method.
Dispersion properties are extracted for different waveguides such as
isotropic and composite plates, layered coated cylinders, cylindrical bars
and railroad tracks. The formulation is provided in both Cartesian and
cylindrical coordinates and the expressions of the forced solutions are
derived. A number of examples are discussed in the following chapters of
the thesis.
Chapter 4 shows the first application of guided waves to the
interrogation of the adhesive bond between a composite spar and a

composite plate. The models analyzed are representative of the wing skin
to spar joint of Unmanned Aerial Vehicles. The numerical and
experimental results predict and demonstrate the applicability of guided
waves to bond state monitoring.
Chapter 5 and 6 present the second application to defect detection
in railroad tracks. A commercial finite element code (ABAQUS

5
EXPLICIT) is first used to predict the reflection phenomena induced by
transverse defects in the head of the railroad tracks. Numerical and
experimental work is shown, leading to the development of a rail
inspection prototype which uses a non-contact technique (laser/air-coupled)
of guided wave testing.
Chapter 7 describes the use of ultrasonic guided waves for the
monitoring of multi-wire steel strands typically used in suspended bridges
and in prestressed concrete structures. The goal is the development of a

method able to monitor applied loads and defect notch-like defects in these
components. Wave features sensitive to the applied load are successfully
isolated both numerically and experimentally. Defects are also detected by
reflection measurements aided by statistical pattern recognition algorithms.
Chapter 8 presents the conclusion of the dissertation and
recommendations for future areas of research.

2 INTRODUCTION TO ULTRASONIC GUIDED
WAVES
2.1 Introduction
The term guided wave is used to describe waves that require
boundaries for their existence. Guided waves can travel on the surface of
semi-infinite solids (Rayleigh waves), on the interface between two different
media (Stoneley waves), along plates or layers of plates, in generic cross
section beams, axial symmetric rods and cylinders. All the described
structural components are commonly referred to as waveguides. They are
uniform in one direction, along the longitudinal axis of the waveguide,
therefore their cross-section has the same physical and geometric properties
at all points along the waveguide’s axis. Ultrasonic guided waves are
generated by the constructive interference of longitudinal and shear bulk
waves (Figure 2.1). The bulk stress waves, generated by a generic
transducer, interact with the boundaries of the waveguide. Multiple
reflections and mode conversions take place until their superposition form
wave packets i.e. ultrasonic guided waves.
Ultrasonic guided waves provide a highly efficient method for the
non-destructive evaluation (NDE) and the structural health monitoring
(SHM) of solids with finite dimensions. Principal advantages in using
guided ultrasonic waves (GUW) in non-destructive evaluation and
structural health monitoring are: 1) long range inspection since guided
waves propagate for long distances in waveguide like structures, 2)
6
7
complete coverage of the waveguide cross-section, and 3) increased
sensitivity to small defects when compared to global vibrations.
Drawbacks of guided waves are the existence of multiple modes
(more than one propagating mode is generally excited by real transducers)
and their dispersive nature that is represented by frequency-dependent
velocities and frequency-dependent attenuation. Consequently, guided wave
signals are very complicated.
In order to avoid errors in the interpretation of the guided wave
propagation phenomena and to fully exploit their potential, an accurate
knowledge of their dispersive properties is crucial. Phase velocities and
group/energy velocities are important for mode identification. Similarly, the
knowledge of the mode attenuation helps maximizing the inspection range
by exploiting modes associated to minimum energy attenuation. Finally, an
accurate prediction of the modeshapes can be exploited to selectively excite
and detect only specific modes that are sensitive to critical defects of the
structural component.
The groundwork for studying guided waves in cylindrical and flat
waveguides was laid at the end of the 19th century. Pochhammer (1876)
and later Chree (1889) developed the solution of mechanical wave equation
for the case of cylindrical symmetry, however, owing to its complexity,
detailed calculation of the roots did not appear until the middle of the
20th century. In the same period, Gazis (1959) explored the propagation
phenomena for hollow, single layer, elastic circular cylinders in vacuum.
Papers on this subject are numerous. Guided waves in cylindrical elastic
waveguides have been studied theoretically, numerically and also

8
experimentally. For instance Zemanek (1972) was one of the first authors
to present a complete analytical and experimental study.
The basis for the comprehension of guided waves in flat layered
waveguides were principally given by Rayleigh (1887) and Lamb (1917).
Lord Rayleigh derived the equation for waves traveling along the free
surface of a semi-infinite elastic half space. Its derivation yields a third
order expression whose roots determine the velocity of the propagating
surface wave. Stoneley (1924) carried out a generalization of the above
single interface problem. He performed a study of interface waves
propagating without leakage that exist at the boundary between two solid
half spaces. The ranges of existence of free wave solutions (in which the
wave propagates indefinitely without loss of energy) were explored by
Scholte (1947). Pilant (1972) extended the study of Stoneley and examined
the leaky wave solutions representing waves that attenuate as they travel.
Love (1911) and Lamb (1917) added another interface to the
problem studied by Rayleigh and introduced the notion of flat layer of
finite thickness. The derivation by Lamb (1917) consists of two distinct
expressions (Rayleigh-Lamb equations) which roots represent symmetric and
antisymmetric plate modes (see section 2.3). A plot of the roots in the
frequency domain gives the well known Lamb wave dispersion curves.
Love (1911) showed that transverse modes were also possible in a
half-space covered by a layer of finite thickness and different elastic
properties. His modes involve shearing motion in the plane of the layer.
Details on the calculation and theory of guided waves in isotropic media
can be found in classical text books as Viktorov (1967), Auld (1973),
Achenbach (1984), Graff (1991) and Rose (1999). In order to provide a

9
brief introduction to the topic of guided waves, the essential equations
describing guided plate waves (Lamb waves) are reported in the next
section.
Figure 2.1: Guided plate waves in an isotropic homogeneous plate.
2.2 Equations of motion in Unbounded Medium
In the absence of body forces, the equation of equilibrium for an
unbounded medium, expressed in the Cartesian coordinate system (x,y,z),
can be written as (Achenbach, 1984):

10
∂σ x ∂σxy ∂σxz ∂2ux

+ + = ρ
∂x ∂y ∂z ∂t2
∂σxy ∂σ y ∂σyz ∂ 2uy
+ + = ρ (2.1)
∂x ∂y ∂z ∂t2
∂σxz ∂σyz
∂ σz ∂ 2uz
+ + = ρ
∂x ∂y ∂z ∂t2
where ux, uy and uz are the components of the displacement vector u and
ρ represents the material density. Vector and indicial notation expressions
for the same equation of motion are here summarized:
∂2u ∂ 2ui
∇σ = ρ σij, j = ρ , i, j = 1, 2, 3 (2.2)
∂t2 ∂t2
∂ ∂ ∂
where ∇ = x +y +z is the vector operator “nabla” and
∂x ∂y ∂z
⎡ σx σxy σxz ⎤ ⎡σ11 σ12 σ13 ⎤

⎢ ⎥ ⎢ ⎥ (2.3)
σ = ⎢⎢σxy σy ⎥
σyz ⎥ = ⎢σ12 σ22 σ23 ⎥
⎢ ⎥
⎢σ σyz σz ⎥⎥⎦ ⎢σ13 σ23 σ33 ⎥
⎢⎣ xz ⎣ ⎦
represents the symmetric stress tensor with components σij in indicial
notation. In the second expression of Eq. (2.2) the summation convention
for repeated subscripts applies.
The generalized Hook’s constitutive law relates the stress and the
strain tensor components as:
σ = C: ε σij = Cijklεkl , i, j, k, l = 1, 2, 3 (2.4)
where the elastic fourth order stiffness tensor C with 81 independent
components Cijkl , possesses a number of symmetries and allows to write
the constitutive laws in the simplified notation named after Woldermar
Voigt:
11
⎡ σx ⎤ ⎡ c11 c12 c13 c14 c15 c16 ⎤ ⎡ εx ⎤

⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ σy ⎥ ⎢ c22 c23 c24 c25 c26 ⎥ ⎢ εy ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ σz ⎥ ⎢ c33 c34 c35 c36 ⎥⎥ ⎢ εz ⎥
⎢ ⎥ = ⎢ ⎢ ⎥ (2.5)
⎢σ ⎥ ⎢ c44 c45 c46 ⎥⎥ ⎢ε ⎥
⎢ yz ⎥ ⎢ ⎢ yz ⎥
⎢σ ⎥ ⎢ c55 c56 ⎥⎥ ⎢ε ⎥
⎢ xz ⎥ ⎢ ⎢ xz ⎥
⎢σ ⎥ ⎢sym ⎢ε ⎥
⎢⎣ xy ⎥⎦ ⎢⎣ c66 ⎥⎥⎦ ⎢⎣ xy ⎥⎦
The symmetric stress and strain tensors are conveniently represented as 6
components vectors in Voigt notation. The strain components, within the
restrictions of the linearized theory, are typically expressed as follows:
⎡ ∂ux ⎤
⎢ ⎥
⎢ ∂x ⎥
⎢ ∂ uy ⎥
⎢ ⎥
⎡ εx ⎤ ⎡ ε11 ⎤ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ∂y ⎥
⎢ εy ⎥ ⎢ ε22 ⎥ ⎢ ∂ uz ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ εz ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ = ⎢ ε33 ⎥ = ⎢ ∂z
⎥
⎢ ⎥ ⎢ ⎥ ⎢1 ⎛ ∂u ⎞
∂uz ⎟⎥
⎢ εyz ⎥ ⎢ ε23 ⎥ ⎢ ⎜⎜⎜
y
+ ⎟⎥
⎢ε ⎥ ⎢ε ⎥ ⎢ 2 ⎝ ∂z ∂y ⎠⎟⎥ (2.6)
⎢ xz ⎥ ⎢ 13 ⎥ ⎢ 1 ⎛ ∂u ⎥
⎢ε ⎥ ⎢ε ⎥ ⎢ ⎜ x + ∂uz ⎞⎟⎥
⎢⎣ xy ⎥⎦ ⎢⎣ 12 ⎥⎦ ⎢ ⎜ ⎟⎥
⎢ 2 ⎝ ∂z ∂x ⎠⎟⎥
⎢ 1 ⎛ ∂u ∂uy ⎟⎞⎥
⎢ ⎜⎜ x + ⎟⎥
⎢ 2 ⎜⎝ ∂y ∂x ⎟⎠⎦⎥⎥
⎢⎣
1 1 ⎡∇u + (∇u)T ⎤
εij = (ui, j +uj,i ) , ε = ⎢⎣ ⎥⎦
2 2
An elastically isotropic material has no preferred directions and the elastic
constants are independent of the orientation of the Cartesian coordinate.
As a consequence, for elastic isotropic materials, the 21 independent
stiffness constants cij of Eq. (2.5) reduce to only two material constants,
for example the Young’s modulus (E) and Poisson’s ratio (ν) or
12
equivalently the two Lame¶ constants λ and µ. Hooke’s law then simplifies
to:
⎡σ ⎤ ⎡λ + 2µ λ λ ⎤ ⎡ εx ⎤
⎢ x⎥ ⎢ ⎥ ⎢ ⎥
⎢ σy ⎥ ⎢ λ λ + 2µ λ ⎥ ⎢ εy ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ σz ⎥ ⎢ λ λ λ + 2µ ⎥ ⎢ εz ⎥
⎢ ⎥ = ⎢ ⎥ ⎢ ⎥
⎢σ ⎥ ⎢ 2µ ⎥ ⎢ε ⎥
⎢ yz ⎥ ⎢ ⎥ ⎢ yz ⎥
⎢σ ⎥ ⎢ 2µ ⎥ ⎢ε ⎥
⎢ xz ⎥ ⎢ ⎥ ⎢ xz ⎥
⎢σ ⎥ ⎢ ⎢ε ⎥
⎢⎣ xy ⎥⎦ ⎣⎢ 2µ⎥⎥⎦ ⎢⎣ xy ⎥⎦
(2.7)
1 if i = j
σij = λδij εkk + 2µεij , δij =
0 if i ≠ j
σ = λ tr (ε) I + 2µε, tr (ε) = εkk = ε11 + ε22 + ε33
Introducing the strain displacement relations of Eq. (2.6) into the
constitutive law (Eq. (2.7)) provides the following expression
σ = λ I∇iu + µ ⎡⎢∇u + (∇u ) ⎤⎥ , σij = λδijuk,k + µ (ui, j +uj,i )

T
(2.8)
⎣ ⎦
Finally, the Navier’s displacement equation of motion, assuming the
absence of body forces, and for an isotropic elastic unbounded medium is
obtained substituting the expression of the stress tensor (2.8) in Eq. (2.2)
as
∂ 2u ∂2ui
µ∇ u + (λ + µ ) ∇∇iu = ρ 2 ,
2
(λ + µ) uj, ji + µui, jj = ρ (2.9)
∂t ∂t2
The system of equations just obtained, couples the three displacement
components. For this reason, a common approach is to express the

13
displacement vector u in terms of the derivatives of a scalar potential ϕ
and a vector potential ψ =[ ψx ψy ψz]T that satisfy uncoupled differential
equations. The displacement vector decomposed as follows
u = ∇ϕ + ∇ × ψ (2.10)
is finally substituted in the vector form of Eq. (2.9). In force of the
identities ∇i∇ϕ = ∇2ϕ , ∇2 (∇ϕ ) = ∇ (∇2ϕ ) and ∇2 i∇ × ψ = 0 , the
displacement equations of motion become
⎡ ∂ 2ϕ ⎤ ⎡ 2 ∂2ψ ⎤
∇ ⎢(λ + 2µ) ∇2ϕ − ρ 2 ⎥ + ∇ × ⎢ µ∇ − ρ ⎥ = 0 (2.11)
⎢ ∂t ⎥⎦ ⎢ ∂ t2 ⎥
⎣ ⎣ ⎦
The identity derived is satisfied if either of the two terms on the l.h.s. in
Eq. (2.11) vanish providing the following uncoupled wave equations
1 ∂2ϕ
∇2ϕ = (2.12)
cL2 ∂t2
and
1 ∂2 ψ
∇2 ψ = (2.13)
cT 2 ∂t2
where
λ + 2µ µ
cL2 = cT 2 = (2.14)
ρ ρ
It can be shown that harmonic potential functions of the form
ϕ = Φei (kL i x − ω t), ψ = Ψei (kT i x − ω t) (2.15)
satisfy the decoupled equations. The exponential terms, which are wholly
imaginary, describe the harmonic propagation of the waves, in space and

14
time. The wavenumber vector k, represents the spatial distribution of the
wave. The homogeneous waves propagate in the direction of the
wavevector k with a spatial frequency or wavelength λ=2π/|k| and a
temporal circular frequency ω=2πf. Substituting Eqs. (2.15) into Eqs.
(2.12)-(2.13), yields
2 ω2 2 ω2
kL = , kT =
cL2 cT 2 (2.16)
Thus, from (2.15) it can be seen that two types of homogeneous plane
wave may travel through the medium in any direction:
• dilatational waves with longitudinal (pressure) wavespeed cL
• transverse (shear) waves propagating with speed cT. These (bulk

waves) are the eigensolutions for the equation of motion in an
infinite elastic isotropic medium.
Waves will propagate infinitely with their velocities, without changing

direction or amplitude.
λ
λ
z T
t
y
x
λ spatial period wavelength T temporal period
χ=1/λ spatial frequency F=1/T temporal frequency
k=|k|=2π/λ spatial frequency ω=2πf ang. temporal frequency

wavenumber
Figure 2.2: Bulk wave in a 3D space

15
Vacuum
Plate x
2h
Vacuum
Figure 2.3: Homogeneous elastic layer of infinite in-plane dimensions.
2.3 Guided plate waves: Lamb waves
When harmonic waves propagate in bounded media, reflections occur

along the boundaries. In a homogeneous elastic layer harmonic waves can
propagate by being reflected back and forth between the two plane
boundary surfaces (see Figure 2.1). In a steady state situation, which is
assumed for harmonic waves, the systems of incident and reflected waves
constructively interfere, forming a standing wave across the thickness of
the layer and the propagation is essentially in the direction of the layer.
This phenomenon motivates the term waveguide for the layer and for all
the extended bodies with a cross section of finite dimensions (Achenbach,
1984). The principle of constructive interference can be used to analyze the
time harmonic wave motion in plane strain for an elastic layer (Tolstoy
and Usdin, 1953). However, it is more straightforward to introduce

expressions for the field variables representing a standing wave in the
thickness y-direction of the layer and traveling wave in the x-direction.

16
Let us consider the components of the displacement field

decomposed through the potential functions as in Eq. (2.10)
∂ϕ ∂ψz ∂ψy
ux = + −
∂x ∂y ∂z
∂ϕ ∂ψz ∂ψx
uy = − +
∂y ∂x ∂z (2.17)
∂ϕ ∂ψy ∂ψx
uz = + −
∂z ∂x ∂y
Let us assume harmonic the wave motion in a homogeneous elastic

layer of thickness 2h (see Figure 2.3). Assuming a plane strain motion in
the x-y plane, the following constrains apply
∂
uz = 0, ( )=0 (2.18)
∂z
and the displacement components can be summarized as
∂ϕ ∂ψz
ux = u = + (2.19a)
∂x ∂y
∂ϕ ∂ψz
uy = v = − (2.19b)
∂y ∂x
Furthermore, from Eq. (2.19a) and (2.13) we obtain the two following
partial differential equations:
∂ 2ϕ ∂ 2ϕ 1 ∂2ϕ
+ = (2.20a)
∂x2 ∂y 2 cL2 ∂t2
∂ 2ψz ∂ 2ψz 1 ∂ 2ψz
+ = (2.20b)
∂x2 ∂y 2 cT 2 ∂t2
To investigate the harmonic wave motion in the elastic layer, the following
solutions can be considered
17
ϕ = Φ (y) ei (kx − ω t) (2.21a)
ψz = Ψ (y) ei (kx − ω t) (2.21b)
where Φ (y) and Ψ (y) are functions of the position along the thickness
direction and represent standing waves while the exponential term

exp[i(kx-ωt)] describes an harmonic wave propagating in the x direction
with a wave velocity equal to c=ω/k. The terms ω and k are the
angular temporal frequency of the wave and the wavenumber. The
angular temporal frequency is proportional to the temporal frequency f
(ω=2πf) while the wave number is inversely proportional to the
wavelength λ (k=2π/λ).
Substitution of Eqs. (2.21a) and (2.21b) into Eqs. (2.19a) and

(2.19b) respectively, the solutions of the resulting equations can be
obtained as:
Φ (y) = A1 sin ( py) + A2 cos ( py)

(2.22)
Ψ (y) = B1 sin (qy) + B2 cos (qy)
where
w2 w2
p2 = − k2 , q2 = − k2 (2.23)
cL2 cT 2
and A1, A2, B1 and B2 are the wave amplitudes determined from the
boundary conditions.
Introducing the harmonic solutions (Eqs. (2.21a)-(2.21b) in the

expressions of the displacements (Eqs. (2.19a)-(2.19b)) we then obtain
18
ux = ⎡⎣(A2ik cos py + B1q sin qy) + (A1ik sin py − B2q sin qy)⎤⎦ ei (kx − ω t)
uy = ⎡⎣− (A2 sin py + B1ik sin qy) + (A1p cos py − B2ik sin qy)⎤⎦ ei (kx − ω t) (2.24)
symmetric motion antisymmetric motion
It should be noted that the displacement components ux and uy were

grouped in two parts, the first corresponding to symmetric motion (w.r.t.
the longitudinal x-axis) and the second to antisymmetric motion.
Analogously, the potential functions and the stress components can be

decomposed in two terms as:
Φ (y) = A2 cos ( py) ⎤
⎥
⎥
⎥
⎥
Ψ (y) = B1 sin (qy) ⎥
⎥
⎥
⎥ Symmetric
τ xy = µ ⎢−2ikpA2 sin (py) + (k − q ) B1 sin (qy)⎥ ⎥⎥ Motion
⎡ 2 2 ⎤ (2.25)
⎣ ⎦
⎥
⎥
⎥
σy = −λ (k2 + p2 ) A2 cos (py) + ⎥
⎥
⎥
−2µ ⎡⎣ p A2 cos (py) + ikqB1 cos (qy)⎤⎦
2
⎥⎥
⎦
Φ (y) = A1 sin ( py) ⎤

⎥
⎥
⎥
Ψ (y) = B2 cos (qy) ⎥
⎥
⎥
⎥
⎥ Antisymmetric (2.26)
τ xy = µ ⎢2ikpA1 cos (py) + (k − q ) B2 cos (qy)⎥ ⎥⎥
⎡ 2 2 ⎤
Motion
⎣ ⎦
⎥
⎥
⎥
σy = −λ (k2 + p2 ) A1 sin (py) + ⎥
⎥
⎥
−2µ ⎡⎣ p A1 sin (py) − ikqB2 sin (qy)⎤⎦
2
⎥⎥
⎦
19
The integration constants A1, A2, B1 and B2 can be finally found

applying the stress free boundary conditions at the upper (z= h) and
lower (z={h) surfaces of the plate (see Figure 2.3):
τ xz = σz ≡ 0 (2.27)
From the last two identities of Eqs. (2.25) the respect of the boundary
conditions given in Eq. (2.27) yield a system of two homogeneous
equations for the constants A2 and B1. Similarly, for the antisymmetric
motion, two homogeneous equations for the constants A1 and B2 are
obtained. Since the systems are homogeneous, non trivial solutions are
carried out by imposing the determinant of the coefficients equal to zero.

Symmetric and antisymmetric mode equations are finally obtained as:
(k2 − p2 ) sin (qh) 2µikq cos (qh) (2.28)

= −
2ikp sin (ph) (λk 2
+ λ p + 2µ p
2 2
) cos (ph)
(k2 − q2 ) cos (qh) 2µikq sin (qh) (2.29)
− =
2ikp sin (ph) (λk 2
+ λ p2 + 2µ p2 ) sin (ph)
Equations (2.28) and (2.29) can be rewritten in the more compact form
known as the Rayleigh-Lamb equation:
±1
tan(ph) ⎡ 4k2 pq ⎤ (2.30)
= −⎢ 2 ⎥
tan(qh) ⎢ (k − q2 )2 ⎥
⎣ ⎦
where +1 corresponds to symmetric (S) motion as -1 to antisymmetric (A)
motion. Equations (2.30) accept a number of eigenvalues, k0S , k1S , k2S …,
and k0A , k1A , k2A ,…respectively. Each eigenvalue is associated to a set of
eigencoefficients that are A2, B1 for the symmetric case and A1, B2 for
20
the antisymmetric case. Substitution of these coefficients into Eq. (2.24)

yields the corresponding Lamb mode shapes. The symmetric modes are
designated S0, S1, S2,… while the antisymmetric are designated A0, A1,
A2,...
Since the coefficients p and q in Equations (6) and (7) depend on
the angular frequency ω, the eigenvalues kiS and kiA will change with the
excitation frequency. The curves obtained plotting the eigensoultions as a

function of the frequency are frequently referred to as wavenumber
dispersion curves. An example of such maps is shown in Figure 2.4 where

the Lamb mode wavenumbers are represented as a function of the
frequency half-thickness product for an aluminum plate. As a consequence,
the curves can be used for aluminum plates with different thickness scaling
the frequency. For example, highlighted are the first two symmetric and
antisymmetric mode wavenumbers at 2 MHz for a plate h=1mm. The
same wavenumbers are the eigensolutions for an aluminum plate where
h=2mm at 1 MHz.
The corresponding wavespeeds, given by ci = ω / ki , change with
frequency as well. The change of wavespeed with frequency produces wave
dispersion. Lamb waves are highly dispersive, and their speed depends
on the product fh between the frequency, f, and the plate half-thickness,
h. At a given frequency thickness product, for each solution of the
Rayleigh–Lamb equation, one finds a corresponding Lamb wavespeed and a
corresponding Lamb wave mode. The plot of the Lamb wavespeeds against
the fh product gives the wavespeed (phase velocity) dispersion curves
(Figure 2.5).
21
Figure 2.6 and Figure 2.7 present the displacement fields across the
thickness for the two Lamb modes A0 and S0 for fh=2MHzmm. The two
modes are called fundamental modes and are the only modes that exist
and are propagative in the entire frequency range. Modes with index larger
or equal to 1 (for example A1 A2.. or S0 S1…) start to propagate at a
specific frequency, that depend on the mode, called “cut-off frequency” and
can be estimated on the horizontal axis of the wavenumber dispersion

curve (see Figure 2.4). The cut-off frequency, in fact, corresponds to a
zero value of the wavenumber ki=2π/λi. As a consequence, every mode at
its cut-off frequency has infinite wavelength. For the same reason, the
fundamental modes have infinite wavelength at zero frequency that can be

considered their cut-off frequency.
Figure 2.8 and Figure 2.9 show the modeshapes of the first four
propagating modes (A0, A1, S0 and S1) for different frequency values. At
low frequencies, the symmetric S0 Lamb wave mode (see Figure 2.9) is
almost straight across the thickness and resembles the displacement field of
the simple axial wave. Similarly, the antisymmetric A0 Lamb wave mode
resembles the displacement field of the simple flexural wave at low
frequencies. At higher frequencies, however, modeshape curvature increases.

This explains why wave motion can be represented accurately with
classical plate theories, where the perpendicular to the plate plane is
assumed to remain straight during the deformation, only at low
frequencies. In fact, the wave motion of a line perpendicular to the plate
mid-plane, that is given by a combination of modeshapes with large

curvature at high frequencies, cannot be approximated with a simple
rotation and translation of the segment.

22
Typically in NDE, ultrasonic waves are generated by transducers in

the form of wave packets with a finite frequency content. The phase
velocity, that is the speed of propagation of each individual wave crests of
the excited waveform, is generally different from the speed of the wave
packet as a whole. The speed at which the wave packet (or wave
envelope) propagates is the one measured experimentally and is called
group velocity. Group velocity dispersion curves are needed to recognize

and isolate modes, observing different arrival times in the signals obtained
through experimental measures.
14000
12000
10000
k0Anti
Wavenumber, k
8000 k0Symm
[1/m]
6000
k1Anti
k1Symm
4000
A0 S0 S1
2000 A1
0
0 0.5 1 1.5 2 2.5 3
f h [MHz mm]
Figure 2.4: Wavenumber dispersion curves for Lamb waves in an

aluminum plate (h=half-thickness of the plate, cL=6370 m/sec, cT=3160
m/sec, ρ=2770kg/m3). Curves can be used for plates with different
thickness. Highlighted are the first two symmetric and antisymmetric mode
wavenumbers at 2 MHz for h=1mm (or for example at 1 MHz for
h=2mm).
23
2.5
S1
Phase Velocity c/cT
A1
1.5 S0
1
A0
0.5
0
0 0.5 1 1.5 2 2.5 3
f h [MHz mm]
Figure 2.5: (Phase velocity c) dispersion curves for Lamb Waves in an

aluminum plate (cT=shear wavespeed)
u
x
u
y
λ=2π/k0Anti
Figure 2.6: Displacement field of the antisymmetric mode A0 at the

frequency f=2MHz for an aluminum plate with thickness 2h=1mm.
24
u
x
u
y
λ=2π/k0Symm
Figure 2.7: Displacement field of the symmetric mode S0 at the frequency

f=2MHz for an aluminum plate with thickness 2h=1mm. The
displacement is symmetric with respect to the x-axis.
y y y y y
u u u
y y y x x
x x x
u
x
u u u u u
x x x y y u
x
A0 0.5MHzmm A0 1MHzmm A0 1.5MHzmm A1 1.5MHzmm A1 2MHzmm
Figure 2.8: First two symmetric mode shapes (A0, A1) at various products
of frequency and half thickness showing how modeshape curvature increases
with frequency and mode number
25
y y y y y
u u
x y
x x x x x
u
u y
y u
y
u u u
x x x
u u
x y
S 0 1.5MHzmm S0 2MHzmm S 0 1.5MHzmm S1 1.5MHzmm S 1 2MHzmm
Figure 2.9: First two symmetric mode shapes (S0, S1) at various products
of frequency and half thickness showing how modeshape curvature increases
with frequency and mode number.
Also the group velocity depends on the frequency and it provides the
information of the speed at which the mode carries the energy. The
mathematical definition is cg=∂ω/∂k and can be obtained, for example, by
numerically differentiating the wavenumber dispersion curves of each

propagating mode. For the aluminum plate here analyzed, the group
velocity of the first six propagative modes are shown in Figure 2.10.
Another family of waves that can exist in an isotropic single layer

of infinite in-plane dimensions is formed by the horizontally polarized shear
waves. Not treated here, these shear waves induce a particle motion in the
z-direction and are the result of the constructive interference of incident
and reflected shear horizontal bulk waves. In general, when guided waves
26
propagate in plates constituted of anisotropic materials, this motion is

coupled with the motion introduced by Lamb waves. Guided plate modes
propagating in the x-direction, as a consequence, introduce motion in all
the three directions while in isotropic plates the displacement is either

polarized in x-y the plane (see Figure 2.3) or in the z-direction.
2.5
2
Group Velocity cg/cT
S0
1.5 S1
A1
A0
1
0.5
0
0 0.5 1 1.5 2 2.5 3
f h [MHz mm]
Figure 2.10: Group velocity cg dispersion curves for Lamb Waves in an

aluminum plate (cT=shear wavespeed)
27
2.4 Analytical solutions of Guided Waves for Multilayered systems.
In NDE, the need for modeling of wave propagation in media with

arbitrary numbers of flat and cylindrical layers exists. Guided waves are in
fact increasingly used for the inspection of composite plates and cylindrical
layered structural components. The first derivation for wave propagation in
media composed by multiple layers was obtained by Thompson (1950).
Thompson (1950) introduced a transfer matrix method where, for each
layer, the displacements and stresses at the bottom are related to those at
the top through a field (or transfer matrix). The matrices for any number
of layers could be multiplied and modal or response solutions could then
be found by application of the appropriate boundary conditions. Few years

later Haskell (1953) enhanced the technique correcting a small error in the
derivation. The Thompson-Haskell method is affected, however, by a loss
in precision when layers of large thickness are considered and high
frequencies are taken into account (the so called frequency-thickness
problem).
The availability of digital computers has supported the increase in
investigations into the modeling of wave propagation in multilayered
media. In this contest, of particular interest is the publication of Knopoff
(1964) that introduced a global matrix for the full system instead of
transfer matrices. With the contribution of Schmidt and Jensen (1985) and
Pialucha (1992), the “Global Matrix Method” was improved eradicating the
loss of precision which affect the “Transfer Matrix Method”. The methods
were developed primarily for seismological applications to study surface

waves in stratified rock media. Many studies dealt with the dynamic
28
forced solutions for layered half spaces (Luco and Apsel, 1983; Apsel and
Luco, 1983) also in the presence of moving sources (Barros and Luco,
1994; 1995).
The majority of the mentioned studies have been confined to
systems of equations whose modal solutions describe free wave propagation

only while attenuation of ultrasonic guided waves is rarely accounted for.
Attenuation describes the decrease of ultrasonic wave amplitudes i.e. the
decrease of the wave energy. Different mechanisms are responsible for
attenuation. The amplitude reduction of ultrasonic waves is typically
associated to material absorption, scattering and geometrical energy
spreading (Shull, 2002; Krautkramer, 1990). Material absorption of

ultrasonic stress waves is responsible for a conversion of the wave energy
in heat. As a consequence, part of the energy of motion is no longer
available to propagate the wave. Scattering is caused by inhomogeneities
and variations of the material structure. When traveling ultrasonic waves

encounter material variations (such as inclusions and phase changes in
metals or discontinuities due to different constituent materials in
composites) they are scattered (reflected, refracted and mode converted).
When scattering occurs, part of the energy from the original wave is sent
randomly into different directions within the material. The last mechanism
responsible of wave attenuation is the so called geometric attenuation. Due

to the finite extension of the ultrasonic sources, true plane waves cannot
be excited. Waves either converge or diverge and the wave amplitude can
geometrically amplify or attenuate. A simple example of geometrical
attenuation can be provided by a pebble dropped in a lake producing
circular divergent waves. Each wave has the same energy if absorption and
29
scattering are neglected but, as the wave diameter increases, the energy is
spread out over a larger area and the displacement decreases. Another
example can be provided by the energy leakage of ultrasonic guided waves
travelling in waveguides surrounded by lossy materials. This, for example,
is the main dissipation mechanism encountered in guided waves travelling

in steel bars embedded in concrete.
The described attenuation mechanisms are the main sources of
damping and energy absorption of ultrasonic guided waves. In the current
dissertation the first two mechanisms will be considered explicitly while
geometric attenuation will be studied elsewhere. Another attenuation
mechanism might be represented by the friction in between different

waveguides in contact. Chapter 7, for example, studies the wave
propagation phenomena in steel strands formed by seven independent wires
in contact. Friction between wires changes as a function of the axial load
imposed to the strand. Friction has been discarded as a damping

mechanism in the wave propagation analysis. In fact, due to the small
displacements induced by ultrasonic high frequency guided waves travelling
in the strand, damping induced by friction is negligible compared to other
sources of absorption as scattering. It should be noted that friction would

constitute a major source of damping for low frequency vibrations where
the relatively large motion activates friction in between different structural

elements and is the major cause of structural damping.
Attenuated guided modes, were studied for the first time by Coquin
(1964). These waves leak energy into the surrounding media while
propagate and require a complex wavenumber to be considered in the
solution. Coquin (1964) developed an approximate method to extract the
30
roots of the Rayleigh-Lamb frequency equation generalized for an infinite

viscoelastic isotropic plate. Substituting real frequencies in the generalized
Rayleigh-Lamb equation gives complex wavenumbers, the imaginary part
representing the attenuation of the modes i.e. the decay of propagating
waves along the layer system. Complex wavenumbers are a consequence of

substituting in the Rayleigh-Lamb equation the elastic constants with
complex functions of frequency. Hosten (1991) joined the approach of
Coquin and the Transfer matrix method to study the behavior of bulk
heterogeneous plane waves in viscoelastic stratified media.
Recently, the interest in modeling guided wave propagation
phenomena in anisotropic media has involved different researchers. In this

case the analytical formulations are more complex due to the coupling
between the wave propagation in different directions that is absent in
isotropic media. Models based on the transfer matrix method for
calculating dispersion curves in arbitrarily in-plane oriented orthotropic

plates can be found for example in Nayfeh and Chimenti (1989) and
Nayfeh (1991). Dispersion properties were extracted also for layered
piezoelectric composite rods by Nayfeh et al. (2000). Hosten and Castaings
(1993) presented an adaptation of the Thompson-Haskell method to

introduce anisotropic attenuation into the formulation of the transfer
matrix.
Overall, the guided waves in plate, bar or pipe structures are
described by the dispersion and attenuation curves which show how the
wavenumbers and velocities of the waves vary with the frequency. All
these structures are well addressed in the literature. The papers mentioned
are only a small sample of the articles that treat the modeling of guided
31
waves. They all present analytical solutions based on the “superposition of
partial bulk waves method (SPBW)”.
Many excellent publications that describe the SPBW approach have

been done by the Nondestructive testing NDT group, directed by Prof.
Cawley, at Imperial College, London. Among the numerous publications

and thesis on this topic, here for obvious reasons just a few will be
referenced. An excellent review on the “Transfer Matrix method” and
“Global Matrix method” for flat systems, is given by Lowe (1993, 1995).
The model for wave propagation in multilayered cylindrical structures,

based on the work by Gazis (1959), can be found in the PhD thesis of
Pavlakovic (1998). The amount of work done by the NDT group of the
Imperial College to describe the wave propagation phenomena resulted in a
commercial software whose user’s manual is rich in information and
references on the topic (Pavlakovic et al., 1997; 2003).
2.5 Numerical solutions of guided waves in Multilayered systems.
As alternative to the “exact” methods, based on the superposition of
bulk waves, different numerical techniques have been presented. When the
cross section geometry of the waveguide is generic, the last approaches
represent the only possible way to study the wave propagation phenomena.
Numerical methods are frequently based on finite element methods (FE).
As other numerical methods commonly used, for high frequency analysis,

the finite element method becomes very costly in terms of algebraic
32
manipulations. FE models of the whole structure become impractically

large due to the increasing refinement of the mesh required. Furthermore
the physical interpretations of computational results from these methods,
are often quite difficult.
A first alternative approach is based on the theory of wave

propagation in periodic structures. A periodic structure consists
fundamentally of a number of identical structural components (“periodic
elements”) which are joined together end-to-end and/or side-by-side to
form the whole structure. From Floquet’s principle, the wave propagation
in the whole structure can be studied considering only a single repetitive

substructure. A set of periodic boundary conditions, involving a
propagation constant, has to be imposed to the substructure. A review of
the topic was performed by Mead (1996). The introduction of the finite
element method to discretize the substructure allows to study waveguides

with arbitrary geometry. The approach can be implemented using standard
finite element packages. Stiffness, mass and damping matrices of the

substructure, given by the commercial FE code, can be postprocessed
imposing the abovementioned boundary conditions. The solution is obtained

by solving an eigenvalue problem. A three dimensional discretization has to
be initially performed but only a small “periodic” portion of the original
structure is investigated, dramatically reducing the number of degrees of

freedom (Mace, 2005). This approach is frequently referred to as
Waveguide-Finite Element method (WFE) and it has been used to model

plate systems (Mace, 2005), railway tracks (Gry and Gontier, 1997) and
car tyres (Nilsson, 2004). The approach requires a discretization also in the
longitudinal direction of the waveguide therefore the precision is somewhat
33
less than the accuracy obtained with the Semi Analytical Finite Element
(SAFE) method discussed in the following.
SAFE methods have emerged for modeling the guided wave
propagation numerically as an alternative to the “exact” methods based on
the superposition of bulk waves – SPBW and pure finite element methods.
Motivations for hybrid numerical-analytical methods include the necessity

for modeling a large number of layers such as composite laminates and
that of modeling waveguides with arbitrary cross-section for which exact

solutions do not generally exist. In addition, when complex wavenumbers
are part of the solution such as in the case of leaky and/or damped
waveguides, the exact SPBW methods require iterative bi-dimensional root
searching algorithms that may miss some of the solutions (Lowe, 1993).
The general SAFE approach for extracting dispersive solutions uses
a finite element discretization of the cross-section of the waveguide alone.

The displacements along the wave propagation direction are conveniently
described in an analytical fashion as harmonic exponential functions. Thus
only a bi-dimensional discretization of the cross-section is needed, with

considerable computational savings compared to a three-dimensional
discretization of the entire waveguide. The SAFE solutions are obtained in
a stable manner from an eigenvalue problem, and thus do not require the
root searching algorithms used in SPBW approaches. In addition, since
polynomial approximation of the displacement field along the waveguide is

avoided (as an opposite to WFE approach), the method is applicable to
predicting waves with very short wavelengths, where a traditional three-

dimensional approximation may fail. It should be noted that deformations
caused by ultrasonic wave propagation are in general considerably small
34
and linear relationships between the components of stress and strain are
applicable. Consequently the study presented in this dissertation will
assume small deformation and linear elastic behavior to predict wave
propagation in the structural components analyzed.
The next chapter will be devoted to the use of the SAFE to study
guided wave dispersion properties in dissipative waveguides of arbitrary
cross section.
3 MODELING WAVE PROPAGATION IN DAMPED
WAVEGUIDES OF ARBITRARY CROSS-SECTION
3.1 Introduction
It was discussed earlier that guided ultrasonic waves provide a
highly efficient method for the non-destructive evaluation (NDE) and the
structural health monitoring (SHM) of solids with finite dimensions. The
advantages of guided waves can be fully exploited only once the

complexities of guided wave propagation (multimode, dispersion,
attenuation) are unveiled and managed for the given test structure. For
example, the knowledge of the wave velocity is important for mode
identification. Similarly, the knowledge of those mode-frequency
combinations propagating with minimum attenuation losses helps
maximizing the inspection coverage.
Analytical methods based on the superposition of bulk waves
(SPBW) (Lowe, 1995), are well-established algorithms for guided wave
features extraction in plane and axisymmetric problems. In these methods,
the dispersive equations of motion are formulated via constructive
interference of partial bulk waves in the respect of the waveguide
boundary conditions. For damped and/or leaky waves, the solutions are
obtained by looking for those combinations of frequency and complex
wavenumber that satisfy the dispersive equation. The complex roots of the
dispersive equation can be quite difficult to find, due to the solution
change over many orders of magnitude between roots. Routines that rely
on the slope (Muller, 1956) or on the local minima (Lowe, 1995) of the
35
36
characteristic function are commonly used. These procedures are not
straightforward, especially for waveguide with a large number of layers,
where the order of the dispersive characteristic equation outgrows and the
routines results to be computationally inefficient. Searching routines may
miss some roots and a complete solution spectrum is complicated to
obtain. In addition, the degree of geometrical and material complexity
always involves difficulties in the development of analytical equation of
motions. For example for composite multilayer structures, it is, if not
impossible, extremely difficult to achieve an analytical formulation.
A comprehensive literature review on analytical wave propagation
methods in plate and cylindrical systems can be found in (Lowe, 1995)
and (Soldatos, 1994), respectively.
Instead of solving the differential equations of motion, as it is for
SPBW methods, a variational formulation can be used in order to describe
the problem and calculate the dispersion relations. Built on this concept,
semi-analytical approaches, such us Semi-Analytical Finite Element (SAFE)
method, are powerful dedicated algorithms for wave propagation in straight
waveguide that depicts advantages respect to others methods.
The general SAFE approach for extracting dispersive solutions uses
a finite element discretization of the cross-section of the waveguide alone.
The displacements along the wave propagation direction are conveniently
described in an analytical fashion as harmonic exponential functions. Thus
only a bi-dimensional discretization of the cross-section is needed. A SAFE
method for waveguides of arbitrary cross-section was demonstrated for the
first time in 1973 (Lagasse, 1973; Aalami, 1973). In these works dispersive
solutions were obtained for the propagative modes only (i.e. real
37
wavenumbers only). The same technique was used a decade later (Huang
and Dong, 1984) to calculate both propagative modes and nonpropagative,
evanescent modes (complex wavenumbers) for anisotropic cylinders. While
the evanescent modes do not transport any energy along the structure,
they are important from a theoretical viewpoint to satisfy the boundary
conditions. More recently, SAFE methods confined to obtaining the
propagative solutions were applied to thin-walled waveguides (Gavrić, 1994),
railroad tracks (Gavrić, 1995) and wedges (Hladky-Hennion, 1996). An
approximation of the method was also implemented in a standard finite
element package by imposing a cyclic axial symmetry condition (Wilcox et
al., 2002). Other versions of the general SAFE method, again for the
propagative modes, were applied to non-homogenous anisotropic beams
(Volovoi et al., 1998), rods and rails (Hayashi et al., 2003). Both
propagative and evanescent modes in twisted waveguides were studied with
the SAFE method by Onipede and Dong (1996). Reflection phenomena
from the end of a waveguide were studied by Taweel et al. (2000).
Finnveden (2004) investigated the modes in built-up thin-walled structures,
including a channel beam and a plate in a wind tunnel. In this work an
interesting formulation was presented for obtaining the group velocity
values from the individual solutions of the SAFE eigenproblem. This is
advantageous compared to the incremental calculations that are required in
the conventional derivation of the group velocity defined as ∂ω/∂k (ω is
the frequency and k is the wavenumber). Laminated composite waveguides
were studied by SAFE methods for the first time by Dong and Huang
(1985) and, subsequently by Mukdadi et al. (2002) and Mukdadi and
Datta (2003) for laminated plates of both finite and infinite widths. The
38
focus of previous SAFE works was obtaining propagative and evanescent
modes in undamped waveguides. A need exists to extend this technique to
account for material damping. One recent work (Shorter, 2004)
demonstrates a SAFE application to damped, viscoelastic composite
laminates. In this reference a damping loss factor was estimated indirectly
from the power dissipated by the wave. However, the formulation proposed
by Shorter (2004) still does not allow for the calculation of the true wave
attenuation since the governing stiffness matrix was assumed real.
The present study extends the SAFE method for modeling dispersive
solutions in waveguides of arbitrary cross-sections by accounting for
material damping. This extension is particularly relevant for NDE/SHM
applications on high-loss materials such as viscoelastic fiber-reinforced
polymer composites. When accounting for damping, the exact energy
velocity, rather than the conventional group velocity, is calculated along
with the frequency-dependent attenuation of the modes. Various examples
are shown, including isotropic plates, composite laminates, composite-to-
composite adhesive joints and railroad tracks.
3.2 Viscoelastic models for wave propagation
This section reviews the linear viscoelastic models that were used in
the SAFE formulation proposed in the present work. As well known, for
time harmonic motion e-iwt, linear viscoelasticity can be modeled by
allowing complex components in the material’s stiffness matrix
C = C′ − iC′′ (3.1)
39
where C′ contains the storage moduli and C′′ contains the loss moduli.
In practice, the matrix C can be expressed as a combination of the
elastic stiffness tensor, C , and the viscosity tensor, η :
⎡ C11 C12 C13 C14 C15 C16 ⎤

⎢ ⎥
⎢ C22 C23 C24 C25 C26 ⎥
⎢ ⎥
⎢ C33 C34 C35 C36 ⎥⎥
C = ⎢⎢ ,
⎢ C44 C45 C46 ⎥⎥
⎢ C55 C56 ⎥⎥
⎢
⎢Sym. C66 ⎥⎥⎦
⎢⎣
(3.2)
⎡ η11 η12 η13 η14 η15 η16 ⎤
⎢ ⎥
⎢ η22 η23 η24 η25 η26 ⎥
⎢ ⎥
⎢ η33 η34 η35 η36 ⎥⎥
η = ⎢⎢
⎢ η44 η45 η46 ⎥⎥
⎢ η55 η56 ⎥⎥
⎢
⎢Sym. η66 ⎥⎥⎦
⎢⎣
The coefficients of the viscosity tensor are typically measured at a single
frequency value, f (characterization frequency).
The Kelvin-Voigt model and the hysteretic model, both well-
established in ultrasonic NDE, were considered in this study to represent
material damping. In the Kelvin-Voigt model (Neau, 2003; Rose, 1999) the
imaginary component of the stiffness matrix in Eq. (3.26) is frequency
dependent as C′′ = ωη . The complex stiffness coefficients at a generic
frequency, f , can be obtained by opportunely scaling the viscoelastic
tensor coefficients that are given at the characterization frequency, f :

40
ω f
C = C′ − i η = C′ − i η (3.3)
ω f
In the hysteretic model (Neau, 2003) the complex component of the
stiffness matrix is independent of frequency, thus:
C = C′ − iη (3.4)
As a consequence, the hysteretic stiffness matrix has to be determined
only once for the entire frequency range examined.
The wave attenuation, defined as the loss per unit distance
travelled, is commonly modelled as proportional to the frequency times the
imaginary part of the stiffness matrix C′′ (Neau, 2003). From Eqs. (3.3)-
(3.4), the attenuation is a quadratic function of the frequency in the case
of the Kelvin-Voigt model, and a linear function of the frequency in the
case of the hysteretic model. It is also evident that both models predict
the same attenuation at the characterization frequency f . The difference
between the models becomes increasingly significant as the working
frequency differs from the characterization frequency, with the Kelvin-Voigt
model resulting in a smaller attenuation than the hysteretic model below
f , and in a larger attenuation above f .
3.3 SAFE mathematical framework
3.3.1 Problem definition
The mathematical model is presented here for the case of a
waveguide immersed in vacuum, as shown in Figure 3.1(a) and Figure

41
3.1(c). This figure refers to an infinitely wide plate; however, the
formulation is extendable to arbitrary cross-sections. The wave propagates
along direction x with wavenumber k and frequency ω. The cross-section
lies in the y-z plane. The waveguide can generally be composed of
anisotropic viscoelastic materials. The harmonic displacement, stress and
strain field components at each point of the waveguide are expressed by:
u = ⎡⎣ux uy uz ⎤⎦
T
(3.5)
σ = ⎡⎣σx σy σz σyz σxz σxy ⎤⎦

T
(3.6)
ε = ⎡⎣ εx εy εz γyz γxz γ xy ⎤⎦
T
(3.7)
The constitutive relations at a point are given by σ = Cε , where C is
generally complex as defined in Eq. (3.2). The compatibility equations can
be written in matrix form as:
⎡ ∂ ∂ ∂⎤
ε = ⎢L x + Ly + Lz ⎥u
⎢⎣ ∂x ∂y ∂z ⎥⎦ (3.8)
where
⎡1 0 0⎤ ⎡0 0 0⎤ ⎡0 0 0⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢0 0 0⎥ ⎢0 1 0⎥ ⎢0 0 0⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢0 0 0⎥⎥ ⎢0 0 0⎥⎥ ⎢0 0 1⎥⎥
Lx = ⎢⎢ , Ly = ⎢⎢ , Lz = ⎢⎢ (3.9)
⎢0 0 0⎥⎥ ⎢0 0 1⎥⎥ ⎢0 1 0⎥⎥
⎢0 0 1⎥⎥ ⎢0 0 0⎥⎥ ⎢1 0 0⎥⎥
⎢ ⎢ ⎢
⎢0 1 0⎥⎥⎦ ⎢1 0 0⎥⎥⎦ ⎢0 0 0⎥⎥⎦
⎢⎣ ⎢⎣ ⎢⎣
42
(a) (b)
(c) (d)
Figure 3.1: (a) Waveguide of cross-section Ω and (b) generic element of

surface Ωe. (c) Discretization of an infinite plate and (d) degrees of
freedom of the mono-dimensional three-node element.
3.3.2 Equations of motion
Equations of motion for the cross section are formulated by inserting the
kinetic and potential energies into Hamilton’s equation. In general, the
nonconservative form of Hamilton’s principle should be used to account for
dissipation. However, the following analysis adopts a simplified approach
that assumes a conservative waveguide; the resulting imaginary cross-

43
sectional strain energy distribution is used to estimate the power dissipated
by the section via imaginary wavenumbers. The assumption is valid if the
cross-sectional strain energy distribution of a propagating wave is not
significantly modified by increasing levels of damping (Shorter, 2004).
The variation of the Hamiltonian of the waveguide, which vanishes
at all material points, is:

t2
δH = ∫ δ (Φ − K ) dt = 0 (3.10)
t1
where Φ is the strain energy and K is the kinetic energy. The strain
energy is given by:
1
Φ =
2 ∫
T
ε C ε dV (3.11)
V
where the upper script T means a transpose vector and V is the volume.
The result of this equation is complex: the real component represents the
elastic energy, while the imaginary component represents the dissipated
energy.
The kinetic energy is given by:
1
∫
T
K = u ρ udV (3.12)
2 V
where ρ is the mass density and the dot represents a time derivative. By
integrating by parts the kinetic term, Eq. (3.10) can be written as:
t2
⎡ ⎤
⎢ δ (εT ) C ε dV + δ (uT ) ρ udV ⎥⎥ dt = 0
∫ ⎢∫ ∫ (3.13)
t1 ⎣⎢ V V ⎦⎥
44
The displacement field is assumed harmonic along the propagation
direction, x, and spatial functions are used to describe its amplitude in
the cross-sectional plane y-z:
⎡ux (x, y, z, t)⎤ ⎡ux (y, z)⎤

⎢ ⎥ ⎢ ⎥
u(x, y, z, t) = ⎢uy (x, y, z, t)⎥ = ⎢uy (y, z)⎥ ei(ξ x − ω t) (3.14)
⎢ ⎥ ⎢ ⎥
⎢⎢u (x, y, z, t)⎥⎥ ⎢⎢u (y, z)⎥⎥
⎣ z ⎦ ⎣ z ⎦
where i=sqrt(-1) is the imaginary unit.
3.3.3 Semi Analytical Finite element method
The waveguide’s cross-sectional domain, Ω, can be represented by a
system of finite elements with domain Ωe. Mono-dimensional and bi-
dimensional elements were considered in the examples that follow. When
mono-dimensional elements were used, an original routine was adopted to
discretize Ω. Matlab’s “pdetool” and the “GID” software were used for the
discretization by bi-dimensional elements (Matlab User’s Guide, 2000).
The discretized version of the displacement expressions in Eq. (3.14)
over the element domain can be written in terms of the shape functions,
Nk(y,z), and the nodal unknown displacements, (Uxk , Uyk , Uzk ) in the
x, y and z directions (Figure 3.1(b), Figure 3.1(d)):
45
⎡ n ⎤
( e)
⎢ ∑ N j (y, z)Uxj ⎥
⎢ j =1 ⎥
⎢ ⎥
⎢ n ⎥
u (x, y, z, t) = ⎢ ∑ N j (y, z)Uyj ⎥⎥
(e) ⎢ ei(ξ x − ω t) = N(y, z)q (e)ei(ξ x − ω t) (3.15)
⎢ j =1 ⎥
⎢ n ⎥
⎢ ⎥
⎢ ∑ N j (y, z)Uzj ⎥
⎢⎣⎢ j =1 ⎥⎦⎥
where:
⎡N1 N2 Nn ⎤
⎢ ⎥
N(y, z) = ⎢ N1 N2 Nn ⎥ (3.16)
⎢ ⎥
⎢⎢ N1 N2 Nn ⎥⎥⎦
⎣
q (e) = ⎡⎣Ux1 Uy1 Uz1 Ux2 Uy2 Uz2 Uxn Uyn Uzn ⎤⎦
T
(3.17)
and n denotes the number of nodes per element. The strain vector in the
element can be represented as a function of the nodal displacements:
⎡ ∂ ∂ ∂⎤
ε(e) = ⎢Lx + Ly + Lz ⎥ N(y, z)q (e)ei(kx − ω t)
⎢⎣ ∂x ∂y ∂z ⎥⎦ (3.18)
= (B1 + ikB2 ) q (e)ei(kx − ω t)
where L is given by Eq. (3.18), B1 = LyN, y + LzN, z , B2 = LxN , and
N,y and N,x are the derivatives of the shape function matrix with respect
to the y and z directions, respectively.

46
Indicating by nel the total number of cross-sectional elements, the discrete
form of the Hamilton formulation of Eq. (3.13) becomes:
t2
⎪⎧⎪ nel ⎡ ⎤ ⎪⎫
⎪
⎪⎨ ⎢ δ (ε(e)T ) C ε(e) dV + ) ρe u dVe ⎥⎥⎪⎬⎪dt = 0
∫ ⎪⎪e∪ ⎢∫ ∫ δ (u
(e)T (e )
e e (3.19)
=1 ⎢ V ⎥⎦ ⎪⎭⎪
t1 ⎩⎪ ⎣ e Ve
where Ce and ρe are the element’s complex stiffness matrix and density,
respectively.
The substitution of Eq. (3.18) into the strain energy term in Eq.
(3.19), followed by algebraic manipulations, yields:
∫ δ (ε ) C ε(e) dVe =
(e)T
e
Ve
∫∫
Ωe x
(
δ q (e)T (B1T − ikB2T ) ⎡⎣ei(kx − ω t) ⎤⎦
T
) C (B e 1 + ikB2 ) q (e)ei(kx − ω t) dxdΩe
(3.20)
∫ δ ⎡⎢q (e)T (B1T − ikB2T )⎤⎥ Ce (B1 + ikB2 ) q (e) dΩe =
⎣ ⎦
Ωe
δ q (e)T ∫ ⎡⎢B1T CeB1 − ikB2T CeB1 + ikB1T CeB2 + k2B2T CeB2 ⎤⎥dΩeq (e)
⎣ ⎦
Ωe
where iT=-i . Thus the element stiffness matrix can be calculated by
integrating over the cross-sectional domain Ωe only, since the integration
over x reduces to a unity factor due to the complex conjugate terms
exp[±i(kx-ωt)]. For viscoelastic materials, the strain energy defined by Eq.
(3.20) consists of a real component, describing the time-averaged elastic

energy in the section, and an imaginary component, related to the time-
averaged power dissipated by the section.

47
As for the element kinetic energy contribution in Eq. (3.19), by

using the displacement expressions of Eq. (3.15) and simplifying the
harmonic terms exp[±i(kx-ωt)], the following can be written:
(e ) ( e)
∫ δ (u ) ρ u dVe = ∫ ∫ δ (u ) ρ u dxdΩe
(e)T (e)T
e e
Ve Ωe x
(3.21)
= −ω δ q ∫ N ρeNdΩeq
2 (e)T T ( e)
Ωe
Substituting Eqs. (3.20)-(3.21) into Eq. (3.19) yields:

t2
⎪⎧⎪ nel ⎫
(e) ⎪
⎪⎬dt = 0
∫
(e)T ⎡ (e ) ⎤
⎨ ∪ δq + + − ω
(e ) (e ) 2 (e ) 2
⎣ k ik k k k m ⎦ q (3.22)
⎪⎩⎪e =1 1 2 3
⎪⎭⎪
t1
where
⎡B T C B ⎤dΩ
k1(e) = ∫ ⎣⎢ 1 e 1 ⎦⎥ e
Ωe
⎡ T ⎤
k2(e) = ∫ ⎢⎣B1 CeB2 − B2 CeB1 ⎥⎦dΩe
T
Ωe
(3.23)
⎡B T C B ⎤dΩ
k 3(e) = ∫ ⎢⎣ 2 e 2 ⎥⎦ e
Ωe
m( e ) = ∫ NT ρeNdΩe
Ωe
Applying to Eq. (3.22) standard finite element assembling procedures:

t2
∫ {δU ⎡K1 + ikK2 + k2K3 − ω 2M⎤ U}dt = 0

T
⎣ ⎦ (3.24)
t1
where U is the global vector of unknown nodal displacements, and:

nel nel nel nel
K1 = ∪
e =1
k1(e) , K2 = ∪
e =1
k2(e) , K3 = ∪
e =1
k 3(e) , M = ∪m
e =1
(e )
(3.25)
48
Due to the arbitrariness of δU, the following homogeneous general wave
equation is finally obtained:
⎡K1 + ikK2 + k2K3 − ω 2M⎤ U = 0 (3.26)

⎣ ⎦M
where the subscript M is the number of total degrees of freedom of the

system.
The stiffness matrices K1 and K3 in Eq. (3.26) are symmetric, while
K2 is skew symmetric when undamped motion is considered. For damped
motion, the Ki are all generally complex. The "mass" matrix, M, is real
symmetric and positive definite regardless of the type of motion
(undamped or damped). K1 is related to the strain-transformation matrix
B1 that pertains to generalized planar deformations, and thus describes the
generalized plane strain behavior or cross-sectional warpage. K3 models the
out-of-plane deformation behavior since it depends on the matrix B2. K2

contains both B1 and B2 and thus couples the cross-sectional warpage to
the out-of-plane deformations.
Without loss of generality, an M×M transformation diagonal matrix
T is introduced to eliminate the imaginary unit in Eq. (3.26). The

elements of T corresponding to the uy and uz displacement components
are equal to 1, while those corresponding to ux are equal to the imaginary
unit:
49
⎡i ⎤
⎢ ⎥
⎢ 1 ⎥
⎢ ⎥
⎢ 1 ⎥
⎢ ⎥
⎢
T = ⎢ ⎥ (3.27)
⎥
⎢ i ⎥
⎢ ⎥
⎢
⎢ 1 ⎥⎥
⎢ ⎥
⎣⎢ 1⎦⎥
This matrix has the properties TT=T* and T*T= TT*=I, where I is the
identity matrix. The terms in Eq. (3.26) are pre-multiplied by TT and
post-multiplied by T. This manipulation does not alter the matrices K1, K3
and M since they do not mix ux with uy or uz:
TTK1T = K1 TTK3T = K3 TTMT = M (3.28)
The matrix K2, instead, mixes ux with uy and uz but it does not mix uy
and uz with each other. It follows that:
TTK2T = −iK2 (3.29)
where K2 is a symmetric matrix for undamped motion. The introduction
of the matrix T is equivalent to multiplying ux by the imaginary unit to
force the quadrature with uy and uz as done in previous works (Gavrić,
1994; 1995).
The final form of the eigenvalue problem in Eq. (3.26) is:
⎡K1 + ξ K2 + ξ 2K3 − ω 2M⎤ U = 0 (3.30)

⎣⎢ ⎦⎥ M
where U is a new nodal displacement vector. Nontrivial solutions can be

found by solving a twin-parameter generalized eigenproblem in k and ω.
50
The frequency ω is a real positive quantity. The wavenumber k can be

either real or complex and can have both positive and negative signs.
3.3.4 Solutions for undamped media
For lossless materials the stress-strain relation is governed by a real

stiffness matrix C. The use of the operator T is particularly useful in this
case since it simplifies Eq. (3.30) to a real and symmetric system. By
assigning real values to k, Eq. (3.30) can be solved as a standard
eigenvalue problem in ω(k). All the solutions for this case correspond to
propagative waves. Thus if the dimension of the system is equal to M, for
each wavenumber km, M propagating modes (km, ωm) are found along
with the Um cross sectional wavestructure or mode shape.
If the full complex spectrum for both propagative and evanescent
modes is of interest, the unknown complex wavenumbers k(ω) must be
obtained for a given frequency ω, solving Eq. (3.30) as a second-order
polynomial eigenvalue problem. The resulting complex wavenumbers
k=kRe+ikIm are used to describe the velocity of the traveling waves

through their real part, kRe, and their amplitude decay through the
imaginary part, kIm. A classic technique to solve the eigenvalue problem
k(ω) consists of recasting Eq. (3.30) to a first-order eigensystem by

doubling its algebraic size:
[A − kB]2M Q = 0 (3.31)
where:
51
⎡ 0 K1 − ω 2M⎤⎥
⎢
A = ⎢ (3.32)
⎥
⎣⎢K1 − ω M
2
K2 ⎦⎥
⎡K1 − ω 2M 0 ⎤⎥ ⎡U⎤
B = ⎢ , Q = ⎢⎢ ⎥⎥ (3.33)
⎢ 0 −K3 ⎥⎦
⎣ ⎣⎢kU⎦⎥
A and B are real symmetric matrices. From Eq. (3.31), at each frequency
ω, 2M eigenvalues km and, consequently, 2M eigenvectors are obtained.
The eigenvectors are the M forward and the corresponding M backward
modes. The eigenvalues occur as pairs of real numbers (±kRe), representing
propagative waves in the ±x-directions, as pairs of complex conjugate
numbers (±kRe±ikIm), representing evanescent waves decaying in the ±x-
directions, or as pairs of purely imaginary numbers (±ikIm), representing
the nonoscillating evanescent waves in the ± x directions. The phase
velocity can be then evaluated by cph=ω/kreal and the attenuation, in
Nepers per meter, by kIm.
It should be pointed out that if only the propagative modes in

undamped waveguides are of interest, Eq. (3.30) remains the preferred
formulation, since it gives a stable numerical problem and it can be
evaluated roughly one hundred times faster that the linearized version in
Eq. (3.31). In the undamped case, the waves which are nonoscillating
evanescent at low frequencies (i.e. purely imaginary wavenumbers) become

propagative (i.e. purely real wavenumbers) above their cut-off frequencies,
ωc. The cut-off frequencies can be computed by letting k=0 in Eq. (3.30)
and solving the eigenvalue problem:

52
⎡K1 − ωc2M⎤ U = 0 (3.34)

⎣ ⎦M
3.3.5 Solutions for damped media
When material damping is considered, the stiffness matrix C is
complex according to Eq. (3.1) and, consequently, the matrices in Eqs.

(3.30) and (3.31) are also complex. In this case the eigenvalue problem
can only be solved for a given frequency in the k(ω) manner from Eq.
(3.31). Since A and B are now complex, 2M complex eigenvalues km and,
consequently, 2M complex eigenvectors are obtained for each input

frequency ω. In this case there is no analytical distinction between
propagative and evanescent modes due to the fact that both types are
now represented by complex wavenumbers.

If a Kelvin-Voigt model is used, C needs to be scaled in
accordance with Eq. (3.3) considering the characterization frequency f .
Thus this matrix must be updated at each iteration over the frequency
domain of interest. In the case of a hysteretic model, instead, C can
simply be set at the beginning of the simulation for the entire frequency
domain.
3.3.6 Group and energy velocity
In order to compute the group velocity by the conventional manner,

the derivatives of the frequency-wavenumber dispersion relations must be
calculated based on the differences of the values for adjacent points of the
same mode, A and B, i.e. cg = ∂ω ∂k (ωB − ωA ) / (kB − kA ) . This
53
implies that the accuracy of the velocity solution is sensitive to that of

the (k, ω) solutions. Also, in this case the (k, ω) solutions must be
categorized for the different modes (mode tracking). Tracking the modes is
not straightforward when one mode approaches another. One technique to
track the same mode consists of monitoring the cross-sectional mode
shapes in proximity of the overlap between two modes (Pavlakovic, 1997).
A method that avoids the necessity for tracking the modes was proposed
by Hayashi et al. (2003); however, in this work the group velocity
accuracy remains dependent on the resolution of the frequency steps.
The necessity for mode tracking, as well as the dependency on the
frequency step resolution, can be avoided by calculating the group velocity
directly at each (k, ω) solution point without any contribution from
adjacent points. This approach, recently proposed by Finnvedent (2004)
and Han et al. (2002), was used for the results presented here relative to
undamped waveguides. The procedure starts by evaluating the derivative
of Eq. (3.30) with respect to the wavenumber:
∂ ⎡
∂k
( ⎣ K(k) − ω 2M⎤⎦ UR ) = 0 (3.35)
where K(k) = K1 + kK2 + k2K3 and UR represents the right eigenvector.
Pre-multiplying Eq. (3.34) by the transpose of the left eigenvector, ULT
⎡∂ ∂ω ⎤
U LT ⎢ K(k) − 2ω M⎥ U R = 0 (3.36)
⎢⎣ ∂k ∂k ⎥⎦
Since ∂ω ∂k is a scalar, the group velocity can be now written as:

54
∂ω ULT (K2 + 2kK3 ) UR

cg = = (3.37)
∂k 2ω ULT MUR
From this relation the group velocity can be evaluated for each individual
solution (ω, k) of the dispersion relations at a time independently of any
adjacent solution.
As reported by Auld (1990), the group velocity definition is not
valid in damped waveguides. In this case the wavenumber become complex
and the differentiation cg = ∂ω ∂k is no longer possible. If the

differentiation is made with respect to the real part of the complex
wavenumber, then the group velocity calculation yields non-physical
solutions such as infinite velocities at some locations of the dispersion

curves. The energy velocity, Ve, is the appropriate property for damped
media. The definition of the energy velocity can be found in classical

textbooks Auld (1990):
1
Ω ∫ Ω
P ⋅ x dΩ
Ve = (3.38)
1
T ∫
T
1
Ω
(∫ Ω
)
etotdΩ dt
where x is the unit vector along the wave propagation direction, etot is
the total energy density (kinetic and potential), and P represents the time
averaged Poynting vector (real part only). The time averaged Poynting
vector can be calculated from:
1
P = − Re (σu *) (3.39)
2
55
where σ is the classical 3×3 stress tensor, and u * is the complex

conjugate of the particle velocity vector. The numerator in Eq. (3.38) is
the average power flow carried by a mode in the wave propagation
direction over a unit period of time.
The denominator in Eq. (3.38) can be evaluated by introducing the
expressions of the time averaged energy for the kinetic component, ek t ,
and the potential component, ep t , following the formulation by Neau

(2003):
ω2
ek t
= ρ uT u (3.40)
4
1 T
ep = ε C′ ε (3.41)
t
4
where the constants 1/4 result from the time integration over the period
T. Eqs. (3.40) and (3.41) can be evaluated once the element nodal
displacements are calculated from the eigenvalue problem in Eq. (3.31),

and the displacement and strain fields are then reconstructed from Eqs.
(3.15) and (3.18), respectively.
3.4 Results
3.4.1 Plate systems
The general plate system consists of an arbitrary number n of
orthotropic layers stacked along the z direction (Figure 3.1 (a)). The
origin of the reference Cartesian system (x, y, z) is located at the top of
the layered plate and each layer lies parallel to the x-y plane. The plates
56
considered in this study have an infinite length in the width direction, y.
Thus the two-dimensional cross-section that needs to be interpolated by

finite elements reduces to a single line through the plate thickness (Figure
3.1). Mono-dimensional quadratic elements were used for the line
discretization.
In general, each element can have three degrees of freedom (d.o.f.)
per node, associated to the displacements ux, uy and uz. For isotropic
plates, the Lamb modes polarized in the x-z plane are de-coupled from
the shear horizontal (SH) modes that are, instead, polarized in the x-y
plane. The de-coupling holds for orthotropic plates when the wave
propagation direction is along a direction of principal material symmetry.
Consequently, in these cases the number of d.o.f. of each analysis can be

reduced by solving for the Lamb modes and for the SH modes separately
(thus considering only ux and uz for the former modes, and only uy for
the latter modes). For orthotropic plates with an arbitrary wave

propagation direction or for laminated composite plates, the Lamb and the
SH modes are coupled and thus these solutions must be found

simultaneously.
The mesh refinement used in all of the examples that follow was
determined after convergence studies for the dispersive solutions in the

frequency range of interest. Clearly, frequencies higher that those examined
would require more refined meshes.
57
3.4.1.1 Viscoelastic isotropic plate
The first system examined is a viscoelastic isotropic High

Performance PolyEthylene (HPPE) plate in vacuum. This plastic material
has a relatively high damping. This example was chosen because it was
fully studied by Bernard et al. (1999; 2001) by using the software
DISPERSE that is based on a SPBW method. The physical and geometric
characteristic of the HPPE plate are the same as those used by Bernard
et al. (1999; 2001): density ρ=953 kg/m3, thickness h=12.7mm,

longitudinal bulk velocity cL = 2344m/s, shear bulk velocity cT = 953 m/s,
longitudinal bulk wave attenuation κL = 0.055 Np/wavelength and shear
bulk wave attenuation κT = 0.286 Np/wavelength.
For the SAFE modeling, the complex bulk velocities for the
viscoelastic material must be first calculated as follows:
−1
⎛ ⎞
cL, T = cL, T ⎜⎜1 + i κL, T ⎟⎟ (3.42)
⎜⎝ 2π ⎠⎟⎟
The complex Young's modulus, E , and Poisson's ratio, ν , can be
obtained as:
⎛ 3c 2 − 4cT 2 ⎞⎟ 1 ⎛⎜ cL2 − 2cT 2 ⎞⎟

E = ρcT 2 ⎜⎜ L2 ⎟ ν = ⎜ ⎟ (3.43)
⎜⎝ cL − cT 2 ⎠⎟⎟ 2 ⎝⎜ cL2 − cT 2 ⎠⎟⎟
Alternatively the complex Lame’ constants can be calculated as:
Eν E
λ = µ = (3.44)
(1 + ν )(1 − 2ν ) 2(1 + ν )
Finally, the complex viscoelastic stiffness matrix is given by:

58
⎡λ + 2µ λ λ ⎤
⎢ ⎥
⎢ λ λ + 2µ λ ⎥
⎢ ⎥
⎢ λ λ λ + 2µ ⎥
C = ⎢⎢ ⎥
⎥ (3.45)
⎢ µ ⎥
⎢ ⎥
⎢ µ ⎥
⎢ ⎥
⎢⎣ µ ⎥⎦
In this case the viscoelastic stiffness matrix is based on the complex bulk
wave velocities that are kept constant throughout the frequency range
examined. Consequently C is independent of frequency and needs to be
defined only once at the beginning of the analysis. This procedure is

equivalent to an assumption of a hysteretic viscoelastic model.
Forty, quadratic mono-dimensional elements, Figure 3.1(b), were

used for the SAFE discretization. For the Lamb wave solutions, only the
ux and uz d.o.f. were used. These corresponded to 162 of the total

[nel×(nn-1)+1] ×ndof = 243 d.o.f. In the previous equation nel is the
number of finite elements, nn=3 is the number of nodes per element and
ndof is the number of d.o.f. per node. The resulting Lamb wave solutions
are shown in Figure 3.2(a)-(d). The energy velocity values, Figure 3.2 (b),
were obtained from Eq. (3.37). The attenuation values are shown up to
500 Np/m in Figure 3.2(c) and up to 3500 Np/m in Figure 3.2 (d). The
frequency range presented is coincident with the one considered by

Bernard et al. (1999; 2001). In these references, however, some of the
solutions of the attenuation curves are missing resulting in interrupted or

discontinuous branches. This is a consequence of the difficulty of the
searching algorithm based on the SPBW method to converge. The SAFE
results in Figure 3.2 show no missing roots.
59
(a) (b) cLL
Energy Velocity [m/sec]

Phase Velocity [m/sec]
8000 2000
6000 1500
4000 1000
cLL
2000 500
0 0
0 100 200 300 400 500 0 100 200 300 400 500
Frequency [kHz] Frequency [kHz]
(c) (d)
3000
400
Attenuation [Np/m]
Attenuation [Np/m]
2500
300 2000
200 1500
1000
100
500
0 0
0 100 200 300 400 500 0 100 200 300 400 500
Figure 3.2: Dispersion results (Lamb modes) for a 12.7mm thick,

viscoelastic HPPE plate in vacuum: (a) phase velocity, (b) energy velocity,
(c) attenuation below 500 Np/m, (d) attenuation below 3500 Np/m. _____,
Low-attenuation symmetric mode m.
60
10000
10000
1500
1500
9000 (a) (b)

8000
8000
Energy velocity [m/s]

Phase velocity [m/s]
7000
1000
1000 cT

6000
6000
5000
4000
4000
3000 500500
2000
2000
1000
cT
0 00 50 100 150 200 250 300 350 400 450 500 0 00 50 100 150 200 250 300 350 400 450 500
0 100 200 300
Frequency [kHz]
400 500 0 100 200 300
Frequency [kHz] 400 500
Frequency [KHz] Frequency [KHz]
5000 3500
3500
0
(c) (d)
3000
3000
400 0
Attenuation [Np/m]
0 2500
2500
Attenuation [Np/m]
3000
Attenuation [Np/m]
2000
2000
2000 1500
1500
0
1000
1000
1000
0
500500
000 50 100 150 200 250 300 350 400 450 500
0 00 50 100 150 200 250 300 350 400 450 500
0 100 200Frequency [kHz]
300 400 500 0 100 200Frequency [kHz]
300 400 500
Figure 3.3: Dispersion results (SH modes) for a 12.7mm thick, viscoelastic
HPPE plate in vacuum: (a) phase velocity, (b) energy velocity, (c)
attenuation up to 500 Np/m, (d) attenuation up to 3500 Np/m.
Compared to an undamped elastic plate where no solutions exist
below the cut-off frequencies, all modes in Figure 3.2 have solutions that
extend to the origin of the frequency axis. This is the result of the real
wavenumber that is now associated to the formerly nonpropagative roots
of the undamped case. Below the undamped cut-off frequencies, the
damped solutions are characterized by large attenuation values and small
energy velocity values. Although these portions have an interesting
theoretical significance, they have little practical use in NDE/SHM. If
61
needed, the “nonpropagative” branches can be easily deleted from the

dispersion curves by thresholding either the attenuation or the energy
velocity values.
Highlighted in Figure 3.2 is the symmetric mode, m, that has the
lowest attenuation above 165 kHz. Because of the low attenuation, this
mode was examined in detail by Bernard et al. (2001). As confirmed in
this reference, both phase and energy velocities for m tend to the bulk
longitudinal velocity as the frequency increases, since the dominant

displacements are along the wave propagation direction.
SAFE solutions for the SH modes are presented in Figure 3.2. In

this case only 81 d.o.f., corresponding to the uy displacements, were used
in the model. SH modes were not presented by Bernard et al. (1999;
2001), and thus no comparison was possible. All SH solutions are found in
the frequency range examined. As expected, the velocities now tend to the
shear bulk wave velocity at high frequencies. As found for the Lamb
modes, the “nonpropagative” SH modes (i.e. below the undamped cut-off
frequencies) have large attenuation values and small energy velocity values.
3.4.1.2 Elastic transversely-isotropic plate
This example examines a 1mm-thick, unidirectional laminate made

of thirty-two, T300/914 carbon-epoxy laminae. The material is considered
elastic (undamped). This case was previously studied by using the
DISPERSE software based on the SPBW method (Pavlakovic and Lowe,
2003). The material density is ρ=1560 kg/m3 and the elastic properties in
the principal directions of material symmetry are given in Table 3.1, where
62
1 is the fiber direction, 2 is the direction perpendicular to the fibers in

the laminate plane, and 3 is the through-thickness direction. The laminate
can thus be treated as transversely isotropic, with five independent elastic
constants remembering that:
C44=0.5×(C33-C12) (3.46)
For waves propagating along a direction x oriented at any angle θ with

respect to the fiber direction 1, the SAFE model simply requires the
rotation of the stiffness matrix of each lamina through:
Cθ = R1CR −2 1 (3.47)
where C is the stiffness matrix in the lamina’s principal directions (that
can generally be complex), and R1 and R2 are the rotation matrices:
⎡ m2 n2 0 0 0 2mn ⎤
⎢ ⎥
⎢ n2 m2 0 0 0 −2mn ⎥
⎢ ⎥
⎢ 0 0 1 0 0 0 ⎥
R1 = ⎢⎢ ⎥
⎥
⎢ 0 0 0 m −n 0 ⎥
⎢ ⎥
⎢ 0 0 0 n m 0 ⎥
⎢ 2⎥
⎢⎣−mn mn 0 0 0 m2 − n ⎥⎦
(3.48)
⎡ m2 n2 0 0 ⎤ 0 mn
⎢ ⎥
⎢ n2 m2 0 0 ⎥ 0 −mn
⎢ ⎥
⎢ 0 0 1 0 ⎥ 0 0
R 2 = ⎢⎢ ⎥
⎥
⎢ 0 0 0 m −n 0 ⎥
⎢ ⎥
⎢ 0 0 0 n m 0 ⎥
⎢ 2 2⎥
⎢⎣−2mn 2mn 0 0 0 m − n ⎥⎦
63
with m=cosθ and n=sinθ. The governing eigenvalue problem is then

solved by using the rotated stiffness matrix in the constitutive relations
σ = Cθ ε .
The results are shown in Figure 3.4 in terms of phase velocity and
group velocity. The latter result was computed from Eq. (3.37). Since an
elastic material is being considered, the group velocity is now appropriate.
Solutions were found for a wave propagation direction oriented at 0deg
(Figure 3.4(a) and (b)), at 45deg (Figure 3.4(c) and (d)) and at 90deg
(Figure 3.4(e) and (f)) with respect to the fiber direction 1. In the 45deg
case, the option of tracing the Lamb and the SH modes separately is no
longer possible due to the coupling effects between them. This option is
instead viable in the 0deg and 90deg propagation directions. For purposes
of comparison, both Lamb and SH modes are shown in all plots of Figure
3.4. That Lamb and SH modes are coupled in the 45deg direction can be
readily seen in Figure 3.4(c) and (d) by noticing that the lowest-order SH0
mode has some degree of dispersion. The same SH0 mode is, instead,
perfectly nondispersive in the 0deg and 90deg plots. It can also be seen
that the velocity values for the Lamb modes tend to decrease with
increasing wave propagation angle as expected, particularly for the
symmetric modes. The results are coincident with those obtained by
Pavlakovic (1997) using the SPBW method for the 0deg and the 45deg
directions; this reference, however, did not report the SH modes in the
0deg direction and the A0 mode in the 45deg direction.

64
3.4.1.3 Viscoelastic orthotropic plate
The next example is a viscoelastic orthotropic plate that was

examined by Neau (2003). The plate is a 3.6mm-thick carbon-epoxy with
density ρ=1560 kg/m3. The elastic (Cij) and viscoelastic (ηij) properties
are given in Table 3.2 in the principal directions of material symmetry.

The hysteretic viscoelastic model was used here.
The results are shown in Figure 3.7, for a propagation direction at
60deg from principal direction 1. Since damping is being considered, the
solutions exist in the entire frequency range. When damping is small, the
existence of solutions below the undamped cut-off frequencies has little

effect on the phase velocity above these frequencies. Nevertheless, branches
of the phase velocity curves that would be distinct in the undamped case
may become connected in the damped case. An example of a branch
connection is shown in the inset of Figure 3.7(a) that zooms into the 800
kHz – 820 kHz range. Similar branch connection phenomena were

observed in damped isotropic plates by Bernard et al. (1999).
Table 3.1: Elastic properties for the T300/914 laminate (elastic constants
in GPa).
C11 C12 C13 C22 C23 C33 C44 C55 C66
143.8 6.2 6.2 13.3 6.5 13.3 3.6 5.7 5.7
65
Table 3.2: Elastic and viscous properties of the orthotropic plate examined
by Neau (2003) and in the present study (elastic constants in GPa,
Viscosities given at 2.0 MHz).
C11 C12 C13 C22 C23 C33 C44 C55 C66
86.60 9.00 6.40 13.50 6.80 14.00 2.72 4.06 4.70
η11 η12 η13 η22 η23 η33 η44 η55 η66

7.50 0.30 0.60 0.60 0.25 0.28 0.10 0.12 0.28
Table 3.3: Elastic properties of the T800/924 lamina (elastic constants in

GPa).
C11 C12 C13 C22 C23 C33 C44 C55 C66
168.4 5.45 5.45 11.3 4.74 11.3 3.28 6.0 6.0
Table 3.4: Elastic and viscous properties of the orthotropic plate examined
by Deschamps (1992) and in the present study (elastic constants in GPa,
Viscosities given at 2.242 MHz).
C11 C12 C13 C22 C23 C33 C44 C55 C66
132 6.9 5.9 12.3 5.5 12.1 3.32 6.21 6.15
η11 η12 η13 η22 η23 η33 η44 η55 η66
0.40 0.001 0.016 0.037 0.021 0.043 0.009 0.015 0.02
3.4.1.4 Elastic composite laminate
A quasi-isotropic composite laminate made of unidirectional
T800/924 graphite-epoxy laminae with a stacking sequence of [±45/0/90]S
was considered next. Each lamina has a thickness of 0.125 mm resulting in

66
θ=0° θ=0°
15000 10000
(a) (b)

8000
Group Velocity [m/sec]

10000
6000
4000
5000
2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
θ=45° θ=45°
15000 10000
(c) (d)
8000

10000
6000
4000
5000
2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
θ=90° θ=90°
15000 10000
(e) (f)
8000
10000
6000
4000
5000
2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Figure 3.4: Dispersion results for a 1mm thick, elastic transversely-isotropic

carbon-epoxy laminate in vacuum: (a) phase velocity for waves propagating
along principal direction 1, (b) group velocity for waves propagating along
principal direction 1, (c) phase velocity for waves propagating at 45deg
from principal direction 1, (d) group velocity for waves propagating at
45deg from principal direction 1, (e) phase velocity for waves propagating
along principal direction 2, (f) group velocity for waves propagating along
principal direction 2. SH0 mode.
67
Symmetric Antisymmetric
10000 10000
(a) (c)
8000 8000

6000 6000
4000 4000
2000 2000
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
200 200
(b) (d)
150 150
Attenuation [Np/m]
Attenuation [Np/m]
100 100
50 50
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
Figure 3.5: Dispersion results for a 3.6mm thick, viscoelastic orthotropic

carbon-epoxy plate in vacuum for waves propagating at 60deg from
principal direction 1: (a) phase velocity for the symmetric modes (inset:
οοοοο undamped case, ⋅⋅⋅⋅⋅⋅⋅ damped case), (b) attenuation for the
symmetric modes, (c) phase velocity for the antisymmetric modes, (d)
attenuation for the antisymmetric modes.
a total laminate thickness of 1 mm. The elastic constants for the material
were taken from Disperse (2003): E11=161 GPa, E22=9.25 GPa, G12=6.0
GPa, υ12=0.34 and υ23=0.41. The material density was ρ = 1500 kg/m³.
The corresponding terms of the stiffness matrix are shown in Table 3.3.
Two, quadratic mono-dimensional finite elements were used to model
each of the eight laminae. Figure 3.5 presents the phase and group
velocity results for waves propagating along the fiber direction of the 0deg
68
lamina. The C matrices of each lamina were opportunely rotated according

to Eq. (3.47) Also shown in Figure 3.5 are the displacement cross-sectional
shapes extracted from the SAFE analysis for modes m1, m2 and m3 at
800 kHz and m4, m5 and m6 at 1.6 MHz.
The reference system is consistent with Figure 3.1(a). It can be

clearly seen that the mode shapes for a laminated composite are generally
quite different from the analogous mode shapes of an isotropic
homogeneous plate. Abrupt changes can occur at the interface between two
laminae. The mode shape predictions can be exploited in NDE/SHM
applications to enhance the sensitivity to a particular structural defect
and/or to minimize the losses due to leakage in the surrounding medium.
For example, an internal defect would be most efficiently detected by a
mode such as m2 in Figure 3.6 with large displacements at the laminate’s
mid-plane. Acoustic losses from leakage into a surrounding fluid medium

would be severe for modes with large out-of-plane displacements at the
laminate’s surface such as m1 in Figure 3.6. Clearly these trends will
depend on the particular frequency used to probe the structure.
3.4.1.5 Viscoelastic orthotropic plate: comparison between hysteretic and
Kelvin-Voigt models
In this section the viscoelastic orthotropic plate studied Deschamps
(1992) is examined. The plate is a unidirectional lamina of 1 mm in
thickness. The material’s properties are shown in Table 3.4. The viscosities
are given at 2.242 MHz.

69
SAFE results were obtained for both the hysteretic and the Kelvin-
Voigt viscoelastic models for the purpose of comparing the two solutions.
Eqs. (3.3) and (3.4) were used to define the complex stiffness matrix in
the two models, where f was set equal to 2.242 MHz in Eq. (3.3). Ten,
quadratic mono-dimensional elements were used for the discretization. The
wave propagation direction coincided with the fiber direction 1. Thus the
Lamb and SH modes could be solved separately.
Figure 3.6 presents the Lamb wave results obtained by using the
hysteretic model in the left column, and by using the Kelvin-Voigt model
in the right column. Figure 3.8 presents the corresponding plots for the
SH modes. It can be seen in both figures that changing the viscoelastic
model has little effect on the phase or the energy velocity results in the
frequency range considered. Appreciable effects are instead seen in the
attenuation plots (Figure 3.7(c) and (f) and Figure 3.8(c) and (f)). In
particular, both models give the same solution at the characterization
frequency of 2.242 MHz as expected. Above and below this frequency, the
Kelvin-Voigt model results in respectively larger and smaller attenuation

values compared to the hysteretic model.
3.4.1.6 Composite-to-composite adhesively-bonded joint
This section presents SAFE results relative to ongoing efforts at

UCSD aimed at the development of an on-board SHM system for
Unmanned Aerial Vehicles (UAVs) based on integrated sensors and
ultrasonic guided waves (Matt et al., 2005). The monitoring is being

targeted to the adhesive bond between the UAV composite wing skin and
70
the tubular composite spar shown in the drawing of Figure 3.9. The spar
runs along the lengthwise direction of the wing.
The wing skin under investigation is a T300/5208 carbon-epoxy
laminate with a stacking sequence [0/±45/0]S and a thickness of 0.133mm
per lamina. The 0deg direction is parallel to the spar lengthwise direction.
The wing skin was modeled by the usual rotation of the stiffness matrix
according to Eq. (3.47). The spar is a cross-ply tubular section made of
T800/924 and having a total wall thickness of 5.235 mm.
In the model, the spar was considered as one equivalent viscoelastic
orthotropic layer without loss of accuracy (Matt et al., 2005). The
adhesive layer had a typical thickness of 0.203 mm. One quadratic element
was used for each lamina of the skin and for the bond layer, whereas five
elements were used for the spar wall. The hysteretic viscoelastic model
was used for each of the components.
The on-board sensor disposition is such that the wave is generated
and detected on the wing skin on either side of the joint. The wave
propagation direction is perpendicular to the spar, along direction x in the
drawing of Figure 3.9. Any degradation in the bond condition can then be
monitored by measuring changes in the strength of the ultrasonic
transmission through the joint.
71
15000 8000
(a) (b)

6000
10000
4000
5000
2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
m1 m2 m3
0 0 0
Ux1
0.2 ux
Thickness [mm]
0.2 0.2
Thickness [mm]
Thickness [mm]
U x2
0.4 uy
0.4 0.4
U x3
0.6
uz
0.6 0.6
0.8 0.8 0.8
1 1 1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
Normalized displacement Normalized displacement Normalized displacement
m4 m5 m6
0 0 0
0.2 0.2
Thickness [mm]
Thickness [mm]
0.2
Thickness [mm]
0.4 0.4 0.4
0.6 0.6 0.6
0.8 0.8 0.8
1 1 1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
Normalized displacement Normalized displacement Normalized displacement
Figure 3.6: Dispersion results for a 1mm thick, elastic [±45/0/90]S

composite laminate in vacuum for waves propagating along principal
direction 1 of the 0deg lamina: (a) phase velocity, (b) group velocity.
Cross-sectional displacement mode shapes shown for: mode m1, mode
m2, mode m3, mode m4, mode m5, mode m6.
72
Hysteretic Kelvin-Voigt
15000 15000
(a) (d)

10000 10000
5000 5000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
10000 10000
(b) (e)

8000 8000
6000 6000
4000 4000
2000 2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
100 100
(c) (f)
80 80
Attenuation [Np/m]
Attenuation [Np/m]
60 60
40 40
20 20
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Figure 3.7: Dispersion results (Lamb modes) for a 1mm thick, viscoelastic
orthotropic plate in vacuum: (a), (b) and (c) case of hysteretic viscoelastic
model; (d), (e) and (f) case of Kelvin-Voigt viscoelastic model.
73
Hysteretic Kelvin-Voigt
10000 10000
(a) (d)
8000 8000

6000 6000
4000 4000
2000 2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
5000 5000
(b) (e)
4000

4000
3000 3000
2000 2000
1000 1000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
100 100
(c) (f)
80 80
Attenuation [Np/m]
Attenuation [Np/m]
60 60
40 40
20 20
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Figure 3.8: Dispersion results (SH modes) for a 1mm thick, viscoelastic
orthotropic plate in vacuum: (a), (b) and (c) case of hysteretic viscoelastic
model; (d), (e) and (f) case of Kelvin-Voigt viscoelastic model.
74
Table 3.5: Elastic and viscous properties for the UAV wing skin-to-spar
joint (elastic constants in GPa. *Neau et al., 2002)
Layer C'11 C'12 C'13 C'22 C'23 C'33 C'44 C'55 C'66
(C''11) (C''12) (C''13) (C''22) (C''23) (C''33) (C''44) (C''55) (C''66)
Wing skin 135 5.70 5.70 14.2 8.51 14.2 2.87 4.55 4.55
lamina (8.23)* (0.65)* (0.60)* (0.34)* (0.25)* (0.65)* (0.24)* (0.28)* (0.25)*
Spar 88.0 5.45 5.09 88.0 5.09 11.3 4.64 4.64 6.00
wall (4.28) (0.65) (0.425) (4.28) (0.425) (0.65) (0.26) (0.26) (0.25)
0.070 0.069 0.069 0.070 0.069 0.070 0.00012 0.00012 0.00012
Disbond (0.035) (0.035) (0.035) (0.035) (0.035) (0.035) (0.00013) (0.00013) (0.00013)
Table 3.6: Bulk ultrasonic velocities and attenuations

of the UAV wing skin-to-spar interface layer.
Layer cL[m/s] cT [m/s] κL [Np/λ] κT[Np/λ]
Regular bond 2410 1210 0.149 0.276
Disbond 241 12.1 1.497 2.763
SAFE dispersion results are presented in Figure 3.9 for the

disbonded skin-to-spar interface, the most extreme bond degradation that
was considered in this study. The specific material properties assumed for
the various layers of the joint are summarized in Table 3.5. Densities were
1530 kg/m3 for the skin and the spar, and 1421 kg/m3 for the adhesive
layer.
Table 3.6 compares the ultrasonic properties assumed for the

disbonded layer to those assumed for the regular adhesive layer. It can be
75
seen that the largest degradation was imposed to the shear wave velocity
to reflect the inability of the disbond to transfer shear stresses.
The weak properties of the disbonded interface essentially de-couple
the dispersive behavior of the wing skin from that of the spar.
Accordingly, the solutions in Figure 3.9 show modes whose energy is
10000
10000 10000
10000
9000
(a) A1,spar 9000
(b)
8000
8000
S0,spar 8000 8000
S0,spar

7000 7000

S0,plate
6000
6000
6000
6000
s0
5000 SH0,plate 5000 sh0 A1,spar
S0,plate
s0
4000 4000
4000 4000
3000
A0,plate sh0 3000 A
SH 0,spar A0,spar SH0,plate 0,spar A0,plate
SH0,spar
2000
2000
2000
2000
1000 1000
a0 a0
000 50 100 150 200 250 300 000 50 100 150 200 250 300
0 50 100 150
Frequency [kHz] 200 250 300 0 50 100 150
Frequency [kHz]
200 250 300
Propagation plane Upper Wing Skin

x
8080
(c) y
7070 z
A0,plate
6060 Lower Wing Skin Lengthwise direction
Spar of wing
Attenuation [Np/m]
50 50
Attenuation [Np/m]
A1,spar a0
4040 A0,spar [0/±45/0] S Skin
30 30
SH0,plate ,sh0
2020 x
SH0,spar
10 10 S0,plate ,s0 Disbond
S0,spar Spar
000 50 100 150 200 250 300 z
0 50 100 150
Frequency [kHz] 200 250 300
Frequency [KHz]
Figure 3.9: Dispersion results for UAV wing skin-to-spar adhesive joint
with a disbonded interface for waves propagating perpendicularly to the
spar lengthwise direction: (a) phase velocity, (b) energy velocity, (c)
attenuation. οοοοο single skin modes.
76
mainly concentrated within the wing skin above the bondline (identified in
the figure by S0,plate, A0,plate, SH0,plate , etc.), and modes whose energy is
mainly concentrated within the spar below the bondline (identified by

S0,spar, A0,spar, SH0,spar, etc.). The former modes closely match those that
would be supported by the wing skin alone (identified by s0, a0, sh0 and
represented by open dots in Figure 3.9). The match between the “skin”
modes of the disbonded joint and the pure single-skin modes becomes
closer as the frequency increases (compare, for example, A0,plate and a0).
One implication is that waves generated on the wing skin outside of the
joint will be transferred very efficiently across the disbonded interface
through one of the “skin” modes. Consequently, the occurrence of a
disbond can be detected by an increased strength of ultrasonic
transmission compared to a regularly-bonded joint.
3.4.2 Arbitrary cross-sections: viscoelastic railroad track
The purpose of this section is to demonstrate the applicability of

the SAFE approach to waveguides of arbitrary cross-sections that cannot
be solved by exact methods. The case treated is that of a railroad track.
Knowledge of the dispersive behavior of guided waves in rails is relevant

for the purpose of train noise reductions at low frequencies, below 6 kHz
(Thompson, 2003), and for long-range NDE defect detection at high
frequencies, up to 50 kHz (Wilcox et al., 2003; Lanza di Scalea and
McNamara, 2003).
The rail considered is a typical 115-lb A.R.E.M.A. section, modeled

as an isotropic material with hysteretic damping, and having the following
properties: ρ=7932 kg/m3, cL = 5960 m/s, cT = 3260 m/s, κL = 0.003

77
Np/wavelength and κT = 0.043 Np/wavelength. The rail cross-section has

a complex geometry with one vertical axis of symmetry. The mesh, shown
in Figure 3.10 and generated by Matlab’s “pdetool,” used 81 nodes for 106
triangular elements with linear interpolation displacement functions.

Compared to the plate systems, a compromise was made by decreasing the
order of the interpolation function with the increased number of degrees of
freedom necessary to move from a one-dimensional discretization to the bi-
dimensional discretization of the rail track.

The dispersion results are shown in Figure 3.10 up to a frequency
of 50 kHz. The complexity of the modes is evident in these plots. Notice
that no prior solutions for the attenuation values are available from the
literature in this frequency range, since previous wave propagation models
of rails did not include damping, with the exception of low-frequency (< 6
kHz) studies (Wu and Thompson, 1999; 1999). Knowledge of the high-
frequency attenuation, however, is important to identify low-loss mode-

frequency combinations that can provide truly long-range defect detection.
A zoom into the low-frequency phase velocity curves is shown in Figure
3.11 along with the first five cross-sectional mode shapes at 5 kHz. It can
be seen that modes m1, m2 and m4 are antisymmetric with respect to
the x-z plane, while modes m3 and m5 are symmetric. It can also be
seen that some of the modes excite preferably a certain portion of the rail,
whether the head or the base. This information can be used in practical
NDE tests to target defects at various locations in the rail section.
Similarly, knowledge of the mode shapes is necessary to design the
appropriate wave excitation/detection approach.
78
10000 6000
(a) (b)
8000 5000

4000
6000
3000
4000
2000
2000
1000
0 0
0 10 20 30 40 50 0 10 20 30 40 50
2
(c)
1.5
Attenuation [Np/m]
0.5
0
0 10 20 30 40 50 y x
Frequency [kHz]
z
Figure 3.10: Dispersion results for a 115-lb A.R.E.M.A., viscoelastic rail for
waves propagating along the rail running direction: (a) phase velocity, (b)
energy velocity, (c) attenuation.
3.4.3 Hollow Cylinder
The specimen is a ASTM – A53 – F steel pipe with an outer

external diameter of 60.2 mm (2-3/8 in) and a wall thickness of 5.08 mm
(0.2 in). This waveguide is characterized by axial symmetry. The

corresponding SAFE mesh of the cross-section is shown in Figure 3.12(a).
The system is considered composed of an isotropic material with hysteretic

damping, and having the following properties: ρ=7932 kg/m3, cL = 5960
m/s, cT = 3260 m/s, κL = 0.003 Np/wavelength and κT = 0.043

79
Np/wavelength. The analysis used 144 three-node, triangular elements

with linear shape functions. The resulting phase velocity, energy velocity
and attenuation curves are shown in Figure 3.12(b), (c) and (d),
respectively.
6000 m1
5000
4000
3000
2000
1000
0
0 2 4 6 8 10
Frequency [kHz]
m2 m3
m4 m5
Figure 3.11: Dispersion results for a 115-lb A.R.E.M.A., viscoelastic rail:

phase velocity for frequencies below 10 kHz, and first five cross-sectional
mode shapes at 5 kHz. mode m1, mode m2, mode m3, mode
m4, mode m5. οοοοο symmetric modes, ……. antisymmetric modes.
80
Modes L(0,2) and F(1,3), highlighted in the plots are the most interesting
because they show the lowest attenuation and they are characterized by
the higher values of energy velocity. Even if dispersion properties of
axialsymmetric waveguides can be predicted using the SAFE method and
discretizing the entire cross section with two dimensional elements, a more
efficient way to predict dispersion features can be performed by exploiting

the axial symmetry.
(a) (b)
10
Phase Velocity (km/sec)

9 F(1,3)
L(0,2)
6
0
0 50 100 150 200 250
Frequency (kHz)
(c) (d)
L(0,2) F(1,3)
6 0
2.0 2
8
Energy Velocity (km/sec)
5 0
1.6 6
Attenuation (Np/m)
4
4 0
1.2 2
3 0 1
0.8 8
2 0
6
1 0 0.4 4 F(1,3)
2
L(0,2)
0 0 0.0 0
0 50 100 150 200 250 0 50 100 150 200 250
Frequency (kHz) Frequency (kHz)
Figure 3.12: (a) Notation and mesh for the SAFE model of the pipe; (b)
phase velocity dispersion curves; (c) energy velocity dispersion curves; (d)
attenuation curves
81
3.5 SAFE for axial symmetric systems
In this paragraph the S.A.F.E. approach is used to model waveguides

defined in cylindrical coordinates. The basis of the method were given by
Nelson et al. (1971). The idea to use an uncoupled procedure assuming
simple polynomial displacement functions in the transverse direction and
continuously differentiable smooth series in the longitudinal direction, was
published first by Cheung (1968) under the name of Finite Strip Method
(FSM). In this work the FSM approach was used to study the vibration
of simply supported plate systems. Based on this philosophy, Nelson and
his coworkers (1971), presented an application of the extended Rayleigh-
Ritz method to study vibrations and wave propagation in laminated
orthotropic cylinders. In this pioneering work, the solution of wave motions

was obtained by representing the displacement vector as the product of
trigonometric functions accounting for axial, circumferential and time
dependences, while the displacement field in the radial direction was
approximated by interpolation functions. The approximate governing

equations lead to an algebraic eigenvalue problem. The same method was
used by many researchers. For example Kohl et al. (1992) focused on
axially symmetric modes and studied the propagation of a disturbance in
axially cylindrical waveguides determining the modal amplitudes as a
function of the boundary conditions. More recently the approach was

adopted to characterize the wave properties in a functionally graded
cylinder considering a linear variation of the material properties in the

thickness direction (Han et al., 2002).
In the following details of this approach are reported. The advantage
of the mixed analytical numerical technique is the possibility to solve a
82
three dimensional problem with a simple monodimensional FEM

interpolation in the radial direction. This leads to an enormous
computational saving and allows distinguishing modes that are
characterized by different circumferential order number.
3.5.1 Formulation
Consider an infinitely long axisymmetric waveguide. In the most

general case, the waveguide is composed of perfectly bonded elastic or
viscoelastic layers each with distinct material properties and thickness. The
outer surface for rods and both the inner and outer surfaces in the case of
hollow cylinders are assumed to be traction free. A reference system of
cylindrical coordinates ( r, θ, z ) is assumed herein. The cross-section is set
in the r − θ plane while the z -axis is assumed fixed at the center of a
generic cross-section and parallel to the waveguide length as shown in

Figure 3.13.
A guided wave with wavenumber k and frequency f, propagates
harmonically along the z-direction, ei(kz − ω t) , where ω = 2πf is the wave
circular frequency, t being the time and i=sqrt(-1) is the imaginary unit.
The energy carried by the wave displaces harmonically the material point;
under the assumption of small deformations, the strain-displacement and

the stress-strain relations are given by:
ε = Lu (3.49)
σ = C*ε (3.50)
83
where ε = [ εrr εθθ εzz εθz εzr εrθ ] is the vector of strains, u = [ur vθ wz ] is
T T
the vector of displacements, where ur, uθ and uz represent the radial,
circumferential and axial displacement, respectively, and
σ = [σrr σθθ σzz σθz σzr σrθ ]

T
is the vector of stresses. In Eq. (3.49) L is
the congruence differential operator in cylindrical coordinates:
⎡∂ 1 ∂ ⎤
T
1 ∂
⎢ 0 0 ⎥
⎢ ∂r r ∂z r ∂θ ⎥
⎢ ⎥
⎢ 1 ∂ ∂ ∂ 1⎥
L = ⎢0 0 0 − ⎥ =
⎢ r ∂θ ∂z ∂r r⎥
⎢ ⎥
⎢0 ∂ 1 ∂ ∂
⎢ 0 0 ⎥⎥
⎣⎢ ∂z r ∂θ ∂r ⎦⎥
⎡0 0 0⎤ ⎡0 0 0⎤ ⎡1 0 0⎤ ⎡0 0 0⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ (3.51)
⎢0 0 0⎥ ⎢0 1 0⎥ ⎢0 0 0⎥ ⎢1 0 0⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢0 0 1⎥⎥ ∂ ⎢0 0 0⎥ 1 ∂ ⎢0 0 0⎥ ∂ ⎢0 0 0⎥⎥ 1
= ⎢⎢ + ⎢ ⎥ + ⎢ ⎥ + ⎢ =
⎢0 1 0⎥⎥ ∂z ⎢ 0 0 1⎥ r ∂θ
⎢ ⎥
⎢0 0 0⎥ ∂r
⎢ ⎥
⎢0 0
⎢ 0⎥⎥ r
⎢1 0 0⎥⎥ ⎢ ⎥ ⎢ ⎥ ⎢0 0 0⎥⎥
⎢ ⎢0 0 0⎥ ⎢ 0 0 1⎥ ⎢
⎢0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0⎦⎥⎥
⎢⎣ 0 0⎥⎦ ⎢⎣1 0 0⎥⎦ ⎢⎣0 1 0⎦⎥ ⎣⎢0 −1
∂ 1 ∂ ∂ 1
= L1 + L2 + L3 + L4
∂z r ∂θ ∂r r
In view of the axisymmetric wave propagation problem, a dedicate
displacement field u is assumed:
⎡ur (r, θ, z, t)⎤ ⎡ur (r)⎤

⎢ ⎥ ⎢ ⎥
u(r, θ, z, t) = ⎢uθ (r, θ, z, t)⎥ = ⎢uθ (r)⎥ ei(nθ + kz − ω t) (3.52)
⎢ ⎥ ⎢ ⎥
⎢⎢uz (r, θ, z, t)⎥⎥ ⎢⎢uz (r)⎥⎥
⎣ ⎦ ⎣ ⎦
where the circumferential wavenumber or circumferential order number, n,
can only be either zero or integer, because the functions must be periodic
in the circumferential direction for wave propagating in the axial z-
84
direction. This particular form of the solution leads the displacement
components ur , uθ and uz to be function of the radius r only. As
already discussed linear viscoelastic materials are accounted for in the
model by Eq. (3.51) where C* is the complex stiffness constitutive tensor.
Eq. (3.51) is analogous to the elastic constitutive equation, and the

substitution between the elastic and complex tensor holds in force of the
correspondence principle (Auld, 1990).
Based on the axisymmetric nature of the problem, monodimensional
finite element can be used to represent the displacement amplitude in the
r, θ and z-direction along the radius r . The cross-section discretization
reduces then to a line of elements along the radius of the waveguide.
In the most generic case of bonded coaxial cylinders with distinct

material properties, any cylinder can be describer by one or more
dedicated finite elements throughout its thickness. The displacement field
(3.52) in the generic e-th finite element can be approximated as:
u(e) = N (r) q (e)ei (nθ + kz − ω t) (3.53)
where N (r) is the shape functions matrix and q (e) is the vector of nodal
displacement components. For mono-dimensional three nodes quadratic
elements, the above quantities are:
⎡N1 0 0 N2 0 0 N3 0 0⎤
⎢ ⎥
N(r) = ⎢ 0 N1 0 0 N2 0 0 N3 0 ⎥ (3.54)
⎢ ⎥
⎢0 0 N1 0 0 N2 0 0 N3 ⎥⎦
⎣
and
85
q (e) = [Ur1 Uθ1 Uz1 Ur2 Uθ 2 Uz2 Ur 3 Uθ 3 Uz3 ]

T
(3.55)
where N1(r) = 1 − 3r + 2r 2 , N2 (r) = 4r − 4r 2 and N3 (r) = −r + 2r 2 .
The stress within the element can now be written in terms of the
approximate displacement vector Eq. (3.51)-(3.52) as:
ε(e) = [ikB1 + inB2 + B3 ] q (e)ei(nθ + kz − ω t) (3.56)
where:
N(r) ∂N(r) N(r)

B1 = L1N(r), B2 = L2 , B3 = L3 + L4 (3.57)
r ∂r r
The dispersion equation is formulated by using the Hamilton’s principle.
The first variation of the Hamiltonian of the waveguide:
t2 nel
δH = ∫ ∪ ⎡⎣δΦ − δ T (e) ⎤⎦dt = 0
( e)
(3.58)
e =1
t1
can be thought as the summation over the number of elements nel of the
first variation element strain energy δΦ(e) and kinetic energy δ T (e) , that in
cylindrical coordinates assume this representation:
T
+∞ 2π ro( e )
δΦ(e) = ∫ ∫ ∫ δ ⎡⎣ ε(e) ⎤⎦ C*(e)ε(e)rdrdθdz (3.59)
−∞ 0 ri( e )
+∞ 2π ro( e )
⎡u(e) ⎤ u(e)ρ rdrdθdz
T
δ T (e ) = ∫ −∞ ∫ ∫
0 ri( e ) ⎣ ⎦ (3.60)
In Eq. (3.59) C*(e) is the complex tensor for the e-th element, the upper
script T means a transpose vector, ρ is the mass density, the overdot
86
represents a time derivative and ri(e) and ro(e) the e-th element inner and
outer radius.
Substituting the approximate expression for ε(e) in Eq. (3.59), yields:
δΦ(e) = δ q (e)T ⎡⎣k2k1(e) + knk 2(e) + ikk 3(e) +

(3.61)
n2k 4(e) + ink 5(e) + k 6(e)q (e) ⎤⎦
where:
ro( e )
k1(e) = ∫ r ⎡⎣B1T C*(e)B1 ⎤⎦dr
ri( e )
ro( e )
k2(e) = ∫ r ⎡⎣B2T C*(e)B1 + B1T C*(e)B2 ⎤⎦dr
ri( e )
ro( e )
k 3(e) = ∫ r ⎡⎣B3T C*(e)B1 − B1T C*(e)B3 ⎤⎦dr
ri( e )
(3.62)
ro( e )
k4 ( e)
= ∫ r ⎡⎣B2 CT
B2 ⎤⎦dr
*(e)
ri( e )
ro( e )
k5(e) = ∫ r ⎡⎣B3T C*(e)B2 − B2T C*(e)B3 ⎤⎦dr
ri( e )
ro( e )
k 6(e) = ∫ r ⎣⎡B3T C*(e)B3 ⎦⎤dr
ri( e )
and the integration over θ and z reduces to a unity factor due to the
complex conjugate terms ± exp [i(nθ + kz − ω t)] . Integrating by parts the
kinetic term (3.60) and by using the approximated displacement field
(3.53), leads to:
δ T (e) = δ q (e)T ⎡⎢ω 2m(e) ⎤⎥ q (e) (3.63)

⎣ ⎦
where:
ro( e )
m ( e)
= ∫ r ⎡⎣N T ρ (e)N ⎤⎦dr (3.64)
ri( e )
87
Substituting Eqs. (3.59)-(3.60) into the Hamilton’s equation (3.58), leads
to:
nel
∪ δq (e) T ⎡k2k1(e) + knk 2(e) + ikk 3(e) +

⎣
e =1 (3.65)
(e) ⎤
2
n k4 ( e)
+ ink (e)
+k ( e)
− ω mq ⎥ = 0
2
5 6
⎦
Performing the standard finite element assembling procedure the dispersive

wave equation is finally obtained as:
⎡k2K + knK + ikK + n2K + inK + K − ω 2M⎤ Q = 0 (3.66)

⎣⎢ 1 2 3 4 5 6 ⎦⎥ M
where Q is the global vector of nodal displacement components and the

subscript M indicates the problem dimension.
The governing equation of motion (3.66) has the form of a three-parameter

algebraic eigensystem, depending on the circular frequency ω and on both
longitudinal k and circumferential wavenumbers n. Herein, n is assigned
to obtain the n-th order axial symmetric modes and k is adopted as the
eigenvalue parameter for a given frequency ω .
The quadratic eigenproblem (3.66) for k can be recasted into a 2M first
order form as:
⎡A (n, ω ) − kB (n, ω )⎤ Q = 0 (3.67)

⎣ ⎦ 2M
where
⎡ 0 n2K4 + inK5 + K6 − ω 2M⎤⎥

⎢
A = 2 (3.68)
⎢n K + inK + K − ω 2M nK2 + iK3 ⎥
⎣ 4 5 6 ⎦ 2M
88
⎡n2K4 + inK5 + K6 − ω 2M 0 ⎤⎥ ⎡Q⎤

B = ⎢ Q = ⎢ ⎥ (3.69)
⎢ 0 −K1 ⎥⎦ 2M ⎢⎣ ξ Q⎥⎦
⎣ 2M
Once the circumferential order number n is assigned, solving Eq.
(3.59) yields 2M complex eigenvalues km = kRe + ikIm and the
corresponding 2M eigenvectors, Q , for a given input frequency ω . The
2M roots of the dispersion relation are not independent, that is, if km is
a root, then −km is also a root, where the overbar denotes complex
conjugate. Half of the 2M roots therefore physically relate to traveling or
standing damped modes along the positive z-axis and the other half are
their counterparts in the opposite direction.
Once the wavenumbers km = kRe + ikIm have been calculated, for
each guided wave the phase velocity can be evaluated by cph = ω / kRe
while the guided wave attenuation, in Nepers per meter (Np/m), is
given by kIm . The attenuation may also be easily expressed as decibels per
meter (dB/m), since a Neper can be converted to a decibel by multiplying
for 20 log10 (e1) ≅ 8.6859 .
For viscoelastic materials solving the Eq. (3) for a given input
frequency ω yields m = 1 ÷ 2M complex eigenvalues km = kRe

m
+ ikIm
m
and the corresponding eigenvectors Um = URe

m
+ iUIm
m
. Only half of the 2M
roots of the dispersion relation are independent ( km ). The remaining M
solutions are defined as −km , where the overbar means complex conjugate.
From the m-th eigenvalue km the phase velocity cph

m
= ω / ξRe
m
[ m / sec ]
and the attenuation attm = ξIm

m
[ Np / m ] for the m-th guided mode are
89
obtained. From the corresponding eigenvector Um , instead, the nodal
displacement vector q m for each finite element in the mesh ( e = 1 ÷ nel )
can be extracted and the displacement uhm , strain εhm and stress σhm at a
point reconstructed. The time averaged strain and kinetic energy densities
at a point are then obtained as:
Re ⎡⎢(σhm ) εhm ⎤⎥
1 T
Sm = (3.70)
T
4 ⎣ ⎦
1
ρ (uhm ) uhm
T
Km = (3.71)
T
4
T
where T
= 1 / T ∫ ( )dt is the time averaging operator and T = 2π / ω
0
is the period [33]. For the m-th mode, the power flow density vector at a
given point, also known as the Poynting vector, can be obtained by time-
averaging over a unit period the product of the stress tensor and the
velocity vector:
1
⎡Prm Pθm Pzm ⎤ Re (σhmuhm )
T
⎣ ⎦ = − (3.72)
2
where σhm is the conventional 3×3 stress tensor Achenbach (1984).
Note that for GUWs propagating in the z − direction, the Poynting
vector component in the circumferential direction is zero, Pθm = 0 . In
addition, if there is no leakage or damping, also the component in the

radial direction is zero, Prm = 0 , i.e. the power flow is entirely in the
direction of the wave propagation.

The rate at which m-th mode energy is transmitted along the
structure is indicated by the mode power flow, PF m , that can be

90
obtained by integrating over the waveguide cross-section Ω the component

of the Poynting vector in the propagation direction:
Ro
= ∫ (P ⋅ z )dΩ = 2π ∫ Pzmrdr
m m
PF (3.73)
Ω Ri
The wave energy velocity for the the m-th mode, Vem , can be finally
obtained as the ratio of the mode power flow and the total energy density
(kinetic and strain) for the whole cross-section:
−1
⎧⎪ Ro ⎫⎪
⎪ ⎤ rdr⎪⎬
= PF ⎨2π ∫ ⎡⎢⎣ S + K
m m m m
Ve (3.74)
⎪⎪ R T T ⎥⎦ ⎪⎪
⎩⎪ i ⎭⎪
Figure 3.13: (a) Schematic representation of a multilayered hollow cylinder,

(b) representation of the through-thickness mono-dimensional finite element
mesh, (c) e-th mono-dimensional quadratic finite element with three
degrees of freedom per node.
91
Since quadratic mono-dimensional elements were adopted in the

formulation (see Figure 3.13(c)), an exact three point Gaussian integration
was used in Eqs. (3.73) and (3.74) to perform the spatial integration
through the thickness of the waveguide.
3.5.2 Numerical validation
For the purpose of validating the accuracy of the proposed

m
formulation, the SAFE cut-off frequencies fcut of a scheduled 4-in. ANSI
40 steel pipe were compared with those calculated with a SPBW algorithm
(Seco et al., 2002).

Since cut-off frequencies do not have meaning in damped
waveguides, material damping was not considered. Geometrical and
material properties of the pipe are defined in Table 3.7. For several refined
meshes (nel=1,2,4,8,16,32) the SAFE cut-off frequencies were calculated for
a given circumferential wavenumber n assuming k = 0 . The SPBW cut-
off frequencies were calculated with a zero finding routine based on the
Ridder’s method considering a maximum frequency tolerance of
1 ⋅ 10−5 Hz . In Table 3.8 the SAFE and SPBW cut-off frequencies for the
co-existing Torsional T(0,j), Longitudinal L(0,j) and first-order Flexural
modes F(1,j), are given in the 0-1000kHz frequency range.
Torsional modes were analyzed considering only the dof in the
θ − direction. Complementarily, the r − z dof were used to describe the
Longitudinal modes. Therefore for a mesh with one quadratic finite
element (nel=1) only 3 dof were available to extract the pure Torsional
92
modes and the T(0,4) could not be calculated. Likewise, for nel=1, the
Longitudinal L(0,7), Flexural F(1,10) and F(1,11) modes were not found.
As expected, SAFE slightly overestimates the cut-off frequencies because
discretized systems are stiffer than real structures, producing higher natural
frequencies. However, by increasing the number of finite elements, the
SAFE solution converges rapidly to the SPBW one.

For the same waveguide, a second accuracy test was performed by
comparing the axial wavenumbers at the frequency of 1 MHz. The SAFE
wavenumbers were obtained for different meshes by solving Eq. (3.67) at
the desired frequency. For the SPBW algorithm a maximum frequency
tolerance of 1 ⋅ 10−5 Ηz and a maximum phase speed tolerance of
1 ⋅ 10−5 m /s were used in the roots searching routine. Results are

summarized in Table 3.9.
Table 3.7: Axisymmetric waveguides geometrical and acoustic material

properties
Long. Shear Long. bulk Shear bulk
Inner Density bulk bulk attenuation attenuation
radius Outer
Case Material radius Ro ρ speed speed κL κT
Ri cL cT
[mm] [mm] [kg/m3]
[m/s] [m/s] [Np/λ] [Np/λ]
TML 24515
45/60 0 6.8 970 - 430 - 1.3500
bitumen
5.1 core(a)
Elastic
copper 6.8 7.5 8900 - 2240 - 0
pipe(a)
Elastic 4-in
40 ANSI 51.181 57.150 7800 5900 3190 0 0
steel pipe(b)
5.2
Viscoelastic
bitumen 57.150 57.3024 1500 1860 750 0.2688 1.1310
coating(b)
Elastic
steel 0 7.62 7932 5960 3260 0.003 0.008
strand(c)
5.3 Viscoelastic 7.62 31.75 1600 2810 1700 0.043 0.100
grout(c)
Viscoelastic 31.75 76.20 2152 3758 2090 0.186 0.229
concrete(c,d)
(a) acoustic properties from Simonetti and Cawley (2003).
(b) acoustic properties obtained from Barshinger and Rose (2004)
(c) acoustic properties from Pavlakovic et al. (2001).
(d) acoustic properties from Bartoli et al. (2007).
93
Table 3.8: Cut-off frequencies for a 4 in. 40 ANSI steel pipe in the range
0-1000 kHz for an increasing number of finite elements nel used in the
SAFE mesh. Analytical solution is obtained by a SPBW formulation from
Seco et al. (2002).
Cut-off frequencies [kHz]

Guided
mode SAFE number of finite elements SPBW
nel=1 nel=2 nel=4 nel=8 nel=16 nel=32 solution
T(0,2) 295.3614 268.8374 267.8991 267.8348 267.8307 267.8306 267.8304
T(0,3) 659.2486 589.6497 536.7450 534.8738 534.7456 534.7373 534.7368
T(0,4) no data 965.2810 815.0922 802.8491 801.9137 801.8511 801.8479
L(0,2) 15.7913 15.7913 15.7913 15.7912 15.7913 15.7533 15.7913
L(0,3) 294.7276 268.3422 267.4058 267.3417 267.3376 267.3374 267.3374
L(0,4) 544.9717 496.3560 494.6273 494.5090 494.5014 494.4992 494.5009
L(0,5) 658.9309 589.3322 536.4968 534.6265 534.4984 534.4902 534.4897
L(0,6) 1218.7344 964.9626 814.9279 802.6840 801.7489 801.6872 801.8630
L(0,7) no data 1089.9189 992.2911 988.8335 988.5967 988.5805 988.5805
F(1,2) 9.3779 9.3779 9.3779 9.3779 9.3783 9.3708 9.3779
F(1,3) 22.2642 22.2641 22.2641 22.2642 22.2651 22.2590 22.2641
F(1,4) 294.8777 268.5068 267.5711 267.5070 267.5029 267.5025 267.5026
F(1,5) 295.9331 269.4079 268.4678 268.4033 268.3992 268.3987 268.3990
F(1,6) 544.8441 496.0422 493.9608 493.8121 493.8024 493.7999 493.8018
F(1,7) 658.9978 589.4070 536.5790 534.7090 534.5809 534.5725 534.5722
F(1,8) 659.6892 590.2145 537.6638 535.8220 535.6958 535.6870 535.6871
F(1,9) 1218.7931 965.0083 814.9812 802.7389 801.8038 801.7420 801.7380
F(1,10) no data 965.2204 815.0712 802.8292 801.8941 801.8311 801.8283
F(1,11) no data 1090.1726 992.4761 989.0179 988.7802 988.7638 988.7639
It can be seen in this case that the SAFE formulation underestimates the
wavenumbers. In fact, in discretized (stiffer) structures, guided modes
propagate with higher phase speed, i.e. lower wavenumber. Again, with
94
Table 3.9: Guided modes wavenumbers for a 4 in. 40 ANSI steel pipe at 1
MHz for increasing number of finite elements nel used in the SAFE mesh.
Analytical solution is obtained by a SPBW formulation from Seco et al.
(2002)
Modes Wavenumber [1/m]
@
SAFE number of finite elements SPBW
1000
kHz nel=1 nel=2 nel=4 nel=8 nel=16 nel=32 solution
T(0,1) 1969.6505 1969.6505 1969.6505 1969.6505 1969.6506 1969.6502 1969.6506
T(0,2) 1881.7758 1897.1390 1897.6538 1897.6891 1897.6913 1897.6913 1897.6915
T(0,3) 1481.0296 1590.8062 1661.8818 1664.2207 1664.3805 1664.3909 1664.3915
T(0,4) no data 514.4999 1141.0802 1174.2706 1176.7477 1176.9131 1176.9218
L(0,1) 2012.4827 2056.8344 2110.6264 2132.4272 2135.4840 2135.7271 2135.7426
L(0,2) 1901.4722 2023.9905 2085.9396 2110.1576 2113.4772 2113.7391 2113.7561
L(0,3) 1184.9864 1507.0610 1706.5366 1715.5892 1716.2380 1716.2822 1716.2832
L(0,4) 1030.3750 1123.8164 1205.9364 1230.2622 1232.3134 1232.4523 1232.4598
L(0,5) 719.0069 843.4473 977.7119 998.4361 1000.0818 1000.1917 1000.1992
L(0,6) no data 659.1574 958.2004 973.1884 974.4192 974.5025 974.5074
L(0,7) no data no data 120.4295 145.7714 147.7086 147.8419 147.8439
F(1,1) 2012.3988 2056.7524 2110.5467 2132.3483 2135.4052 2135.6484 2135.6638
F(1,2) 1969.5636 2023.9054 2085.8566 2110.0754 2113.3952 2113.6572 2113.6741
F(1,3) 1901.3838 1969.5640 1969.5641 1969.5641 1969.5641 1969.5642 1969.5641
F(1,4) 1881.6825 1897.0482 1897.5631 1897.5984 1897.6006 1897.6010 1897.6008
F(1,5) 1480.9147 1590.6996 1706.4379 1715.4908 1716.1396 1716.1841 1716.1849
F(1,6) 1184.8409 1506.9465 1661.7775 1664.1167 1664.2765 1664.2878 1664.2875
F(1,7) 1030.2084 1123.6643 1205.7965 1230.1255 1232.1769 1232.3159 1232.3233
F(1,8) 718.7688 843.2457 1140.9286 1174.1227 1176.6001 1176.7667 1176.7742
F(1,9) no data 658.8970 977.5362 998.2641 999.9101 1000.0202 1000.0276
F(1,10) no data 514.1672 958.0220 973.0130 974.2440 974.3275 974.3321
F(1,11) no data no data 119.0030 144.5963 146.5492 146.6842 146.6856

95
increasing number of elements in the mesh, the SAFE wavenumbers

converge rapidly to the SPBW solutions.
The results presented in Table 3.8 and Table 3.9 confirm that the
SAFE meshing criteria proposed by Galan and Alabascal (2002) for
homogeneous plate-like waveguides is also valid for the case of axis
symmetric homogeneous waveguides. This criterion guarantees SAFE

solution accuracy by adopting finite elements of maximum dimension
Lmax = λT / β , where λT = 2πcT / ωmax , β = 4 for quadratic finite
elements and ωmax = 2πfmax is the highest circular frequency for which the
solution is sought. In this example, where cT = 3190 m / s and
fmax = 1 MHz , the maximum element length is Lmax = 0.7975 mm .
Therefore, for the system under study (wall-thickness
Ro − Ri = 5.969 mm ) the above criteria is satisfied by considering a
minimum of eight finite elements in the mesh.
It can be observed in Table 3.8 that for nel=8, the maximum error
on the cut-off frequencies is around 0.12% for the modes T(0,4), F(1,9)
and F(1,10) while the maximum error on the wavenumbers, reported in
Table 3.9, is around 1.4% for the Longitudinal L(0,7) and Flexural F(1,11)
mode. The accuracy of the formulation was also verified by checking that
the boundary conditions for cross-sectional distribution of multiple guided
wave features (displacements, Poynting vector components, energy velocity,
etc.) were as expected.
3.5.3 Elastic copper tube filled with elastic and viscoelastic bitumen
The first example examines an isotropic elastic copper pipe filled
with viscoelastic TML 24515 45/60 bitumen. Wave propagation analysis in

96
this system was performed by a SPBW formulation (Simonetti and
Cawley, 2003). In this paper, Simonetti and Cawley (2003) proposed an
NDE technique to estimate the bulk properties of the viscoelastic core by
measuring the speed and attenuation of the fundamental Torsional guided
mode. In the 0-150 kHz frequency range the bitumen was found behaving
as a hysteretic media with constant complex shear moduli µ = ρcS2 .

Geometrical and material properties are given in Table 3.7.
For the SAFE calculation, the elastic stiffness tensor C for the
for the bitumen were
copper and the constant complex stiffness matrix C
obtained as indicated in section 3.4.1.1. Since the longitudinal bulk
properties do not have an effect on the T(0,j) modes for this system were
arbitrarily assumed equal to zero.
For a maximum frequency fmax = 150 kHz , eleven quadratic
elements, ten in the core and one in the pipe, were used to satisfy the
mesh criteria proposed by Galan and Abscal (2002).
Torsional modes were analyzed by considering only the 23 uθ dof over the
total [nel × (nn − 1) + 1] × ndof = 69 dof, where nn = 3 is the number of
nodes per element and ndof = 3 is the number of dof per node, i.e. ur ,
uθ and uz .
SAFE results for the Torsional modes are shown in Figure 3.14. On
the left column the phase velocity (a), energy velocity (b) and attenuation
(c) dispersion curves are given assuming the bitumen core as an elastic
media neglecting the bulk shear wave attenuation (undamped system). The
same dispersive features are presented on the right column considering the
bitumen core as a viscoelastic material (damped system). In all these plots,
roots with att > 130 dB m-1 were marked with a brighter color. The
97
dashed line in Figure 3.14(a) and (b) indicates the non-dispersive
fundamental Torsional mode for the empty copper pipe Te(0,1).
elastic bitumen core viscoelastic bitumen core

5 5 (d)
T(0,2) T(0,3) T(0,4) T(0,5)
(a)
cph [km/sec]
4 4
cph [km/sec]
3 T e(0,1) 3
2 2
T d(0,1)
1 T(0,1) 1
0 0
0 50 100 150 0 50 100 150
3 3 (e)
T e(0,1)
(b)
Ve [km/sec]
Ve [km/sec]
2 T(0,2) T(0,3) T(0,4) T(0,5) 2
T(0,1)
1 1 T d(0,1)
0 0
0 50 100 150 0 50 100 150
10 (c) 10 (f)
att [dB/mm]
att [dB/mm]
5 5
T d(0,1)
0 0
0 50 100 150 0 50 100 150
Figure 3.14: Torsional modes for a copper ( ρ = 8900 kg m−3 , cT = 2240 m/s ,
κT = 0 Np /λ ) pipe with inner radius 6.80 mm and wall thickness 0.70
mm, filled with bitumen. (a)-Phase velocity, (b)-energy velocity and (c)-
attenuation results considering the bitumen as an elastic medium
( ρ = 970 kg /m3 , cT = 430 m /s , κT = 0 Np / λ ). (d)-Phase velocity, (e)-energy
velocity and (f)-attenuation considering the bitumen as a viscoelastic
medium ( ρ = 970 kg / m3 , cT = 430 m/s , κT = 1.35 Np / λ ). The dashed line in
(a) and (b) correspond to the non-dispersive fundamental Torsional mode
for the empty copper pipe Te(0,1).
For the undamped system, among the 23 possible independent roots
at each frequency the number of propagative (real wavenumber) modes
vary from one, at near zero frequency, to five, at 150 kHz (see Figure
98
3.14 (a)). The remaining roots represent the evanescent modes (imaginary
wavenumber). In Figure 3.14(c), only the evanescent modes with
att < 10 dB mm-1 are shown.

The fundamental T(0,1) mode starts propagating at around 2030.2
m s-1 phase speed, it behaves as non dispersive up to 30 kHz and for

increasing frequency its velocity drops down to the shear bulk wave speed
of the bitumen. As it can be seen from Figure 3.14(b), the higher order
Torsional modes, i.e. T(0,2), T(0,3), T(0,4) and T(0,5), start propagating
at their cut-off frequencies T (0, j )

fcut T (0, 2)
( fcut = 40.62 kHz ,
T (0, 3)
fcut = 71.85 kHz , fcut
T (0, 4)
= 103.31 kHz and fcut
T (0, 5)
= 134.96 kHz ) and
for increasing frequency their energy velocity increases, reaches a maximum
and then decreases tending to the shear speed of the bitumen. Energy
velocity maxima occur where the phase velocity dispersion curves of the
T(0,j) modes intersect the phase velocity curve of the Te(0,1) mode.
In Figure 3.15 the mode shape, the power flow density component
PzT (0, 2) and the strain energy density S T (0, 2) for the T(0,2) mode were
T
calculated at five different frequencies to justify the energy velocity peaks.
The mode shapes were normalized with respect to their maximum value.
The cross-sectional distributions of power flow density and strain energy
density, calculated and represented within each element at the Gaussian
points, were normalized by the mode power flow PF T (0, 2) at the
calculation frequency fi . These normalizations were necessary in order to
compare results at different frequencies.
From Figure 3.15 it can be observed that at a frequency slightly
bigger than the mode cut-off frequency, f1 = 41.69 kHz , the T(0,2) mode
shows little motion in the tube wall and high strain energy density in the
99
core. In the proximity of the cut-off frequency, in fact, the T(0,2) mode
corresponds to a standing wave across the thickness, and due to the large
impedance difference between the core and the tube, the strain energy is
mainly concentrated to the core.
f 1=41.69 kHz f 2=45.01 kHz f 3=53.26 kHz f 4=69.83 kHz f 5=80.12 kHz
T(0,2)
7.5 7.5 7.5 7.5 7.5

[mm]
T(0,2)
radius [mm]
displ.
5 5 5 5 5
Uθ Uθ Uθ Uθ Uθ
radius
(a)
norm.
2.5 2.5 2.5 2.5 2.5

norm.
0 0 0 0 0
-1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1
7.5 7.5 7.5 7.5 7.5

norm. PT(0,2)
radius [mm]
5 5 5 5 5
z
(b)
2.5 2.5 2.5 2.5 2.5
0 0 0 0 0
0 1 2 3 0 1 2 3 0 1 2 3 0 1 2 3 0 1 2 3
4 4 4 4 4
x 10 x 10 x 10 x 10 x 10
7.5 7.5 7.5 7.5 7.5
norm. 〈S〉T(0,2)
radius [mm]
5 5 5 5 5
T
(c)
2.5 2.5 2.5 2.5 2.5
0 0 0 0 0
0 20 40 0 20 40 0 20 40 0 20 40 0 20 40
Figure guided mode for a copper ( ρ = 8900 kg /m3 ,

3.15: T(0,2)
cT = 2240 m /s , κT = 0 Np /λ ) pipe, with inner radius 6.80mm and wall
thickness 0.70mm, filled with elastic bitumen ( ρ = 970 kg m−3 , cT = 430 m s-1 ,
κT = 0 Np / λ ) at different frequencies. (a) normalized mode shape; (b)
normalized power flow density PzT (0,2) ; (c) normalized strain energy density
S T(0,2) . Frequency for each column is indicated at the top.
T
100
As the frequency increases, ( f2 = 45.01 kHz ), the amount of
displacement and power flow in the tube wall slightly increases while the
strain energy density decreases substantially. At f3 = 53.26 kHz , where
the T(0,2) and Te(0,1) phase velocity curves intersect, most of the energy
propagates in the tube rather than in the core where there is a minimum
of strain energy density. As the frequency increases further, displacement,
power flow density and strain energy density in the pipe wall decrease,
and for very high frequency they are primarily confined in the core. For
example, at f5 = 80.12 kHz the T(0,2) mode behaves as a shear wave in
the core with phase and energy velocities close to the shear speed of the
bitumen. Therefore, according to Eq. (3.74) where the energy velocity is
inversely proportional to the normalized energy density, the maximum of
this feature must occur around f3 where the normalized strain energy is
minimum. A similar behaviour, not reported here, has been obtained for
the Torsional T(0,3), T(0,4) and T(0,5) modes at different frequency
ranges.
When the bitumen core is considered as viscoelastic, Figure 3.14(d)-
(e)-(f), all the independent 23 Torsional modes Td(0,j) are damped
(complex wavenumber). Among these solutions, only the Td(0,1) mode has
attenuation smaller than 130 dB m-1 . This mode is generated by the
coupling of the formerly undamped modes branch due to the bitumen
viscosity. Similar branch connection phenomena were observed in
viscoelastic orthotropic and isotropic plates (Bartoli et al., 2005; Bernard
et al, 2001). This mode starts propagating with a phase speed around
2029.2 m/s. At increasing frequency its velocity tends to the bulk shear
101
speed of the copper as an opposite to the undamped T(0,i) modes that
were tending to the shear bulk wave speed of the bitumen.
The higher Td(0,j) j = 2, 3, .., 23 modes propagate in the 0-150
kHz frequency range with non zero phase and energy velocity, and extend
to the origin of the frequency axis at zero value.
2300
(a)
2200 Te(0,1)
cph [m/sec]
T(0,5)
2100 Td(0,1)
2000 T(0,2) T(0,3) T(0,4)
T(0,1)
1900
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Frequency [kHz]
130
(b)
100
att [dB/m]
Td(0,1)
50
f T(0,2)
cut
f T(0,3)
cut
f T(0,4) f T(0,5)
cut
cut
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Frequency [kHz]
Figure 3.16 Torsional Td(0,1) mode phase velocity (a) and attenuation (b)
dispersion curves ( • • • • ) for a copper ( ρ = 8900 kg /m3 , cT = 2240 m /s ,
κT = 0 Np / λ ) pipe, with inner radius 6.80 mm and wall thickness 0.70
mm, filled with viscoelastic bitumen ( ρ = 970 kg /m3 , cT = 430 m / s ,
κT = 1.35 Np / λ ). Also shown in (a) the dispersion curves ( D D D D ) for the
T(0,i) modes of the equivalent undamped system. The dashed line
corresponds to the non-dispersive fundamental Torsional mode for the
empty copper pipe Te(0,1).
102
T (0, i)
Each higher Td(0,j) mode, below the fcut cut-off frequencies of
the equivalent undamped mode T(0,j), is characterized by attenuation
values (Figure 3.14(f)) close to the one of the evanescent modes of the
undamped case (Figure 3.14 (c)), and very small energy velocity. Around
T (0, i)
the fcut cut-off frequency, the attenuation of the damped mode reaches
a minimum and the energy velocity starts to increase. For increasing
frequency, the mode attenuation grows linearly while the energy velocity
remains constant (see Figure 3.14 (e) and 2(f)).
In Figure 3.16 dispersion results for the undamped and damped
systems are overlapped and represented in the 1900 − 2300 m s-1 phase
speed range, Figure 3.16(a), and in the 0 − 130 dB m-1 attenuation range,
Figure 3.16(b). Modes with attenuation larger than 0 − 130 dB m-1 are
not shown. It can be noted that the behavior of the Td(0,1) mode is in
perfect agreement with the results proposed by Simonetti and Cawley
(2003), proving the reliability of the proposed SAFE damped formulation.
Local maxima in the Td(0,1) attenuation spectrum occur around the
mode cut-off frequencies of the undamped system. Interestingly, the
minima of Td(0,1) attenuation occur for those frequencies where the phase
velocity curves of T(0,j) and Te(0,1) mode intersect.
It is well known that in undamped waveguides the Poynting vector
is parallel to the z-axis while for damped media a non zero radial
component Prm appears. In the following, the link between Prm and mode
attenuation is analyzed. In Figure 3.17(a) the PrTd (0,1) component integrated
over the waveguide cross-section Ω and normalized with respect to the
mode power flow PF Td (0,1) is represented. It can be noted that this ratio
presents a similar behavior to the Td(0,1) mode attenuation shown in

103
Figure 3.16(b). In Figure 3.17(b) and (c), are represented the normalized
Td(0,1) mode shapes and cross sectional distributions of the radial power
flow density, respectively.
-0.1
norm. ∫ PrdΩ
(a)
Ω
-0.05
Td(0,1)
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Frequency [kHz]
fd1=40.97 kHz fd2=47.83 kHz fd3=55.24 kHz fd4=148.68 kHz
7.5 7.5 7.5 7.5
norm. Td(0,1)
radius [mm]
5 Uθ 5 Uθ 5 Uθ 5 Uθ
(b)
2.5 2.5 2.5 2.5
0 0 0 0
-1 1 -1 0 1 -1 0 1 -1 0 1
7.5 7.5 7.5 7.5

T (0,1)
radius [mm]
5 5 5 5
norm. Pr d
(c)
2.5 2.5 2.5 2.5
0 0 0 0
-500 0 -500 0 -500 0 -500 0
Figure Td(0,1) mode for a copper ( ρ = 8900 kg /m3 ,

3.17: Torsional
cT = 2240 m / s , κT = 0 Np / λ ) pipe with inner radius 6.80 mm and wall
thickness 0.70 mm filled with viscoelastic bitumen ( ρ = 970 kg /m3 ,
cT = 430 m / s , κT = 1.30 Np / λ ). (a) normalized radial power flow density
PrT (0,1) integrated over the waveguide cross-section Ω . (b) normalized mode
d
shape. (c) normalized radial power flow density PrT (0,1) . Frequency for each d
column is indicated at the top.

104
At fd1 = 40.97 kHz , where the Td(0,1) mode shows an attenuation
peak, the normalized cross sectional distribution of the radial power flow
density is maximum. In contrast at fd 3 = 55.24 kHz where a local
minimum of the normalized radial power flow takes place, the Td(0,1)
mode attenuation minimum occurs.
3.5.4 Elastic steel pipe coated by a thin layer of viscoelastic bitumen
An elastic scheduled 4-in. ANSI 40 steel pipe coated with a thin
layer of viscoelastic bitumen was chosen because studied in depth by
Barshinger and Rose (2004) and Barshinger et al. (2002). In these
references phase velocity and attenuation dispersion curves for the
Longitudinal modes with att(m) < 10 dB m-1 were obtained by using a
SPBW based formulation. Waveguide geometric and acoustic properties are
given in Table 3.7. The bulk wave attenuations for the viscoelastic
bitumen were calculated as κL,S = 2πcL,S (αL, S / ω ) , where αL,S / ω is
given by Barshinger et al. (2002).
According to the mesh criterion of given by Galan and Abascal
(2002) twelve finite elements across the steel pipe thickness and three
elements for the bitumen layer were used. For the Longitudinal modes,
only the 62 ur and uz dof of the total [nel × (nn − 1) + 1] × ndof = 93
dof were used ( nel = 15 ) to formulate the wave equation.

105
10000 0
L(0,4) (a) (c)
L(0,6) L(0,2)
L(0,3) L(0,1)
8000 1
L(0,5)
cph [m/sec]
6000 L(0,2)
2
4000 L(0,3)
L(0,1) 3 L(0,4)
2000 L(0,5)
4
0 0
att [dB/m]
0 200 400 600 800 1000
Frequency [kHz] 5
1
6000 6 L(0,1)
L(0,2)
(b)
L(0,2) 2
4000 L(0,1) L(0,5) 7
Ve [m/sec]
3
L(0,2)
2000 8
4
L(0,3) L(0,6) 10 15 20
0 9
L(0,4) L(0,1)
10
0 200 400 600 800 1000 0 200 400 600 800 1000

cL = 5900 m /s , cT = 3190 m /s ) 4-in. scheduled ANSI 40 steel pipe coated
with 0.1524 mm of viscoelastic bitumen ( ρ = 1500 kg /m3 , cL = 1860 m /s ,
cT = 750 m /s , κL = 0.2688 Np / λ , κT = 1.1310 Np / λ ) examined by Barshinger
and Rose (2002). (a) Phase velocity, (b) energy velocity and (c)
attenuation curves. Low attenuation points are highlighted with an ellipse.
In Figure 3.18 SAFE results in terms of phase velocity, energy
velocity and attenuation dispersion curves are shown in the 0 − 1 MHz
frequency range. The results are in excellent agreement with those
proposed by Barshinger and Rose (2004) and Barshinger et al. (2002).
Here, in addition, the roots with att(m) > 10 dB m-1 are represented with
a brighter marker.
106
In the examined frequency range, among the independent 62
complex roots, the number of damped modes with att(m) < 10 dB m-1
range from 1 to 5 for increasing frequency. The phase and energy velocity
dispersion curve of these modes look quite similar to the phase and group
velocity dispersion curves of an elastic 4-in 40 ANSI steel pipe. This is
due to the fact that the bitumen coating is very thin in comparison with
the elastic pipe layer.
It is interesting to note in Figure 3.18(c) a highly dispersive
behavior of the attenuation curves. In this plot, three frequency-mode
combinations with particularly low attenuation values are highlighted with
an ellipse. These combinations have great potential for performing NDE
tests in coated pipes as demonstrated experimentally by Barshinger et al.
(2002).
From the energy velocity dispersion curves Figure 3.18(b) it can be
noted that the L(0,4) mode exists also as a “backwards” wave where its
phase and energy velocity have opposite signs. The attenuation of this
“backwards” wave has negative values and it is not represented in Figure
3.18(c).
Also for this second example, it can be seen in Figure 3.19(a) that
the radial power flow PrL(0, i) integrated over the waveguide cross-section
and normalized with respect to the modes power flow PF L(0, i) look like
the dispersive GUWs attenuation curves of Figure 3.18(c). In Figure
3.19(b) and (c) the normalized mode shapes and the normalized cross
sectional distribution of the radial power flow density are represented at
three different frequencies.

107
At f2 = 749.69 kHz , where the L(0,3) attenuation curve has a
minimum, both the uz displacement component in the bitumen layer and
the radial power flow distribution have a very small magnitude. These
features, have larger values at f1 and f3 .
0
norm. ∫ PrdΩ
(a)
Ω
0.005
0 100 200 300 400 500 600 700 800 900 1000
Frequency [kHz]
f1=549.94[kHz] f2=749.69 [kHz] f3=950.70 [kHz]
56 56 56 ur
ur
norm. L(0,3)
radius [mm]
ur
uz (b)
54 54 54
uz
uz
52 52 52
-1 0 1 -1 0 1 -1 0 1
56 56 56
L(0,3)
radius [mm]
norm. Pr
54 54 (c)
54
52 52 52
-2 0 2 4 6 8 -2 0 2 4 6 8 -2 0 2 4 6 8
Figure modes for an elastic ( ρ = 7800 kg /m3 ,

3.19: Longitudinal
cL = 5900 m / s , cT = 3190 m / s ) 4-in. scheduled ANSI 40 steel pipe coated
with 0.1524 mm of viscoelastic bitumen ( ρ = 1500 kg /m3 , cL = 1860 m /s ,
cT = 750 m /s , κL = 0.2688 Np / λ , κT = 1.131Np / λ ) examined Barshinger and
Rose (2004). (a) normalized radial power flow density PrL(0,i) integrated
over the waveguide cross-section Ω . (b) normalized L(0,3) mode shapes. (c)
normalized L(0,3) radial power flow density PrL(0,3) . Frequency for each
column is indicated at the top.
108
Some minor differences can be noted by comparing the SPBW
results in obtained by Barshinger and Rose (2004) and Barshinger et al.
(2002) and the SAFE results of Figure 3.18. In the study by Barshinger et
al. (2002) some solutions of the attenuation curves are missing resulting in
interrupted or discontinuous branches. The SAFE results show no missing
roots. Furthermore, in the paper by Barshinger et al. (2002) the
attenuation for both L(0,1) and L(0,2) modes grows exponentially at
around the L(0,2) mode cutoff frequency for the equivalent undamped
L(0, 2)
system ( fcut = 15, 74 kHz ), while in the SAFE results the L(0,1)
attenuation branch extends to the origin of the frequency axis (see the
inset in Figure 3.18(c)). This behavior is expected since the fundamental
modes must have zero attenuation at zero frequency.
3.5.5 Viscoelastic steel strand embedded in concrete
Steel strands embedded in grout provide the necessary level of
prestress in post-tensioned concrete structures. Studying GUW propagation
in these components is important to design effective NDE methods for
detecting defects and monitoring loads Pavlakovic et al. (2001). The aim
of this SAFE simulation is to identify the modes which have low
attenuation as they propagate along the strand-grout-concrete waveguide
(see Figure 3.20) in order to maximize the inspection coverage.
In this example the waveguide consists of a 15.24 mm -diameter
viscoelastic steel bar embedded in a 63.50 mm outer diameter duct,
completely full of viscoelastic grout, surrounded by a layer of concrete

109
with external diameter equal to 152.40 mm . The present three layer axial
symmetric waveguide models the prestressed concrete specimen represented
in Figure 3.20(a) and 8(b) which was the subject of several experimental
tests reported elsewhere (Bartoli et al., 2007). The waveguide geometrical
and acoustic properties are given in Table 3.7.
For a maximum frequency fmax = 700 kHz , the mesh criteria
proposed by Galan and Abascal (2002) required 7 finite elements in the
steel bar, 40 elements in the grout layer and 58 elements in the concrete
layer. Only the 422 dof in the radial and axial direction were used for
the representation of the Longitudinal L(0,i) modes. Figure 3.21 shows the
SAFE dispersion results for the Longitudinal modes in terms of phase
velocity (a), energy velocity (b) and attenuation (c).
(a) (b)
6 inmm
152.4
2.5mm
63.5 in
7-wire strand
φ =15.24 mm z (c)
Ur Uz
i-th element
Uθ
Concrete
Grout
strand clamp Steel rod
Strand
Figure 3.20 (a) Seven wire strand embedded in grout and concrete block.
(b) Section of the specimen. (c) Three layer system considered in the
SAFE model.
110
290 305 320 290 305 320
3.5 200 65
3200
3.0 180 60
cph [km/sec]
3000
55
160 L(low-att)
2.5
2800 50
140
2.0
(a)
120
1.5
att [dB/m]
0 100 200 300 400 500 600 700
L(low-cph) 290 305 320 100
3.0 2500
80
Ve [km/sec]
2.0 2300 60
2100
40 L(low-att)
1.0
20
(b) (c)
0 0
0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700
Figure 3.21: Phase velocity (a), energy velocity (b) and attenuation (c)
dispersion curves for the Longitudinal L(0,i) guided modes in a three-layer
waveguide: viscoelastic steel bar ( ∅ = 15.24 mm , ρ = 7932 kg /m3 ,
cL = 5960 m /s , cT = 3260 m / s , κL = 0.003 Np / λ , κT = 0.008 Np / λ )
embedded in a viscoelastic layer of grout ( ∅ = 63.50 mm , ρ = 1600 kg /m3 ,
cL = 2810 m /s , cT = 1700 m /s , κL = 0.043 Np / λ , κT = 0.100 Np / λ ) and a
viscoelastic layer of concrete ( ∅ = 15.24 mm , ρ = 2200 kg /m3 , cL = 3900 m /s ,
cL = 2200 m /s , κL = 0.200 Np / λ , κT = 0.400 Np / λ ). Three low attenuated
modes at 305 kHz are highlighted in the insets.
111
76.2 76.2 76.2 76.2

ur
60 60 60 60
concrete uz
radius [mm]
L(0,low-cph)
40 40 40 40 (a)
20 grout 20 20 20
steel
0 0 0
-1 0 1 0 1 2 0 500 0 0.1 0.2
norm. mode shape norm. Pr norm. Pz norm. 〈S〉T
76.2 76.2 76.2 76.2

ur
60 60 60 60
concrete uz
radius [mm]
L(0,low-att)
40 40 40 40 (b)
grout
20 20 20 20
steel
0 0
-1 0 1 0 50 0 1000 2000 0 0.2 0.4 0.6
norm. mode shape norm. Pr norm. Pz norm. 〈S〉T
Figure 3.22 (a) Lowest phase velocity Longitudinal mode L(0,low-cph) and
(b) lowest attenuation Longitudinal mode L(0,low-att) in the three-layer
steel-grout-concrete viscoelastic waveguide at 305 kHz . First column:
normalized mode shape. Second column: normalized radial power flow
density PrL(0, i) . Third column: normalized axial power flow density PzL(0, i) .
Fourth column: normalized strain energy density S L(0, i) .
T
Roots with attenuation larger than 100 dB /m are marked with a
brighter line. The dispersion curves look quite complex if compared to the
ones of the previous examples. In this case, in fact, 422 damped roots are
obtained at each frequency with the adopted mesh. However, some
frequency-mode combination with low attenuation can be easily found.
For example, highlighted in the insets of Figure 3.21 are the three
low attenuated modes (●,▼,■) at 305 kHz :
● att = 51.3 dB m-1 , cph = 3323.3 m s-1 , Ve = 2408.6 m s-1 ;
▼ att = 59.3 dB m-1 , cph = 2855.1 m s-1 , Ve = 2488.2 m s-1 ;

112
■ att = 61.7 dB m-1 , cph = 3079.2 m s-1 , Ve = 2269.4 m s-1 .
It can be seen that at around 305 kHz these modes have a
maximum of energy velocity. This property is useful to limit the effect of
pulse dispersion and to reduce the risk of mode overlapping that generally
complicates the measurements. With low attenuation and maximum energy
velocity, these modes are good candidates for NDE of this system.
The mode with smallest attenuation (●) is also labeled as L(0, low − att)
in Figure 3.21(c), while the mode with lowest phase speed at 305 kHz is
highlighted L(0, low − cph ) in Figure 3.21(a). The L(0, low − cph ) mode
has a phase speed cph = 1713.6 m s-1 , attenuation = 157.9 dB m-1 and
energy velocity Ve = 1685.3 m s-1 .
Figure 3.22 presents the normalized mode shapes, the normalized
radial and axial power flow distribution, as well as the normalized strain
energy density for the L(0, low − cph ) and L(0, low − att) modes at
305 kHz . It can be seen that the L(0, low − cph ) mode displacement,
Figure 3.22(a), is mainly in the radial direction with very little
displacement in the axial direction. Radial power flow and energies are
confined to the grout layer. As a result, the mode behaves as a shear
wave traveling at a speed close to the shear bulk speed of the grout.
From Figure 3.22(b), it can be noted that the L(0, low − att)
displacements are concentrated within the steel rod and the grout layer.
The radial power flow is distributed between the steel and the grout with
a transmission peak at the interface. The energy for this mode is mainly
flowing in the strand while only a small amount is present in the grout.
Therefore, traveling mainly in the less attenuating media, the mode has
small attenuation at this particular frequency.

113
This kind of analysis can help designing an NDE system that makes
use of guided modes with ultrasonic energy mainly concentrated the within
the strand.
While SAFE computational time to obtain the results of the first and
second examples is negligible, on the order of a few seconds, it should be
noted that the larger mesh adopted in the current example results in a
significant calculation time. The analysis carried out on a 2.6 GHz
Pentium 4 processor with 1 GB of RAM typically employs 100-120
minutes in this case. The increase in the computational cost is mainly
related to the solution of the eigenvalue problems.
3.6 Forced responces
3.6.1 Response to harmonic force
In this paragraph the steady state response for a time harmonic load
acting on the section Ω located at x = xs is derived. The steady state
response must be periodic with the same frequency ω of the applied load.
To compute the contribution of the external load the potential of the
harmonic external load has to be accounted in the Hamilton’s principle.
Such potential associated to the external forces t is given by (Hayashi et
al., 2003)
Ve = ∫ δ u T t dΩ (3.75)
Ωe
The external traction vector can be described inside the single element as:
114
⎡ +∞ ⎤
t(e) (x, y, z, t) = N (y, z) T(e) (x) e−iω t = ⎢ ∫ NT(e) exp (ikx)dx⎥ e−iω t (3.76)
⎣⎢ −∞ ⎦⎥
In Eq. (3.76), exp[-iωt] accounts for the time harmonic behavior of the
load. T(e) is the nodal external traction vector acting on the e-th element
while T(e) represents the Fourier transform of T(e) . Substituting Eq. (3.76)
in Eq. (3.75) the potential of the load can be rewritten as:

nel
V e( e ) = ∪∫
e =1 Ω
δ u(e)T t(e)dΩ =
e
{⎡⎣e } {∫ }
nel +∞
= ∪∫ ⎤ δ q (e)T NT (y, z)
i(kx − ω t) T
N(y, z)T(e)ei(kx − ω t)dx dydz = (3.77)
⎦ −∞
e =1 Ω
e
{∫ }
nel +∞
⎡ei(kx − ω t) ⎤ ei(kx − ω t)dx
T
= ∪ δq ∫
e =1
( e )T
NT (y, z)N(y, z)T(e)dxdy
−∞ ⎣ ⎦
Ωe
The last term can be added to Eq. (3.22) and by performing minor
manipulations the governing equations comprehensive of the loading term,
are given as:
⎡K1 + ikK2 + k2K3 − ω 2M⎤ U = f (3.78)

⎣ ⎦M
where f is:
nel
f = ∪∫
e =1 Ω
N T (y, z)N(y, z)T(e)dxdy (3.79)
e
The discrete system of governing equations can be recast and the new
system of equations already derived in Eq. (3.31) can be finally
represented as:
(A − ξ B) Q = p (3.80)
115
where matrix A, B and vector Q are the same obtained in Eq. (3.32)-
(3.33) and p is given as:
⎡0⎤
p = ⎢ ⎥ (3.81)
⎢⎣ f ⎥⎦
The solution of Eq. (3.80) can be represented by the a linear combination
of the right eigenvectors ΦRm :
2M
Q = ∑Q
m =1
mΦRm (3.82)
In order to obtain the coefficients Qm this expression is substituted in the
governing equations (3.80) obtaining:
2M
(A − kB) ∑ QmΦRm = p (3.83)
m =1
Pre-multiplying by the left eigenvector ΦLl the last equation
2M
ΦLl (A − kB) ∑ QmΦRm = ΦlLp (3.84)
m =1
and considering the bi-orthogonality conditions:
⎧⎪0 l ≠ m,
ΦLl AΦRm = ⎪⎨ L
⎪⎪⎩ΦmAΦRm = kΦLmBΦRm l = m
(3.85)
⎪⎧0 l ≠ m,
ΦLl (−kB) ΦRm = ⎪⎨
⎪⎪⎩−kΦLmBΦRm l = m
the following expression can be obtained

116
ΦLm (A − kB) QmΦRm = ΦLmp

QmΦLmAΦRm − kQmΦLmBΦRm = kmQmΦLmBΦRm − kQmΦLmBΦRm = (3.86)
kmQmBm − kQmBm = (km − k) QmBm = ΦLmp
Finally the coefficients are given by:
ΦLmp
Qm = − (3.87)
(k − km ) Bm
and the explicit form of the solution vector can now be expressed as:
2M
ΦLmp
Q = ∑− ΦRm (3.88)
m =1 (k − km ) Bm
The nodal displacement vector U (M×1 vector, see Eq. (3.33) ) is
obtained by isolating the upper part of Q (2M×1 vector)
2M
ΦLmp
U (k, ω ) = ∑ ΦRup (3.89)
m = 1 (km − k) Bm
m
where ΦRup
m is the upper part of the m-th right eigenvector ΦRm . The
inverse Fourier transform of Eq. (3.89) must be computed in order to
obtain the displacement vector U. If point loads are considered, acting on
the surface at x=xs the forces can be represented by the Dirac delta
expression δ=δ(x-xs). Vector p contains the Fourier transform of the load
that can be computed as:

+∞
p = ∫
−∞
pδ (x − xS ) e−ikxdx = pe−ikxS (3.90)
where p contains the amplitudes of the nodal loads applied in x=xs. The
displacement vector U can be finally computed through the expression:

117
+∞ +∞ 2M ΦLm 1
U (x, ω ) = ∫ −∞
Ueikxdk = ∫ −∞
∑
m = 1 (km − k) Bm
ΦRup
m
2π
pe−ikxS eikxdk (3.91)
Finally, computing the integral in Eq. (3.91) by using the residue theorem,
the following expression can be obtained:
2M
ΦLmp Rup i ⎡⎣ξm (x − xS )⎤⎦
U (x, ω ) = ∑ − Φm e (3.92)
m =1 Bm
For x>xs Eq. (3.92) can be written as
∑α
i ⎡⎣ ξm (x − xS )⎤⎦
U (x, ω ) = mΦRup
m e (3.93)
m =1
Where the participation factor is
ΦLmp
αm = − (3.94)
Bm
It should be noted that in the Eq. (3.93) the summation is extended only
to the M forward wave modes.
A few examples are shown in the following. In Figure 3.23(a),
harmonic motion produced by a uniform vertical load in a plate is shown.
The system is an aluminum 3.175mm thick plate discretized by 4
quadratic elements (3 nodes per element) along the z direction. Motion
refers to the right portion x > 0 of an infinite plate. Harmonic load
(Figure 3.23(c)) is applied along the thickness at x=xs=0. At the
frequency considered (f=200 kHz) the motion is mainly produced by the
fundamental a0 mode.
118
Motion produced on the same plate by a uniform horizontal
harmonic load is shown in Figure 3.24(a). The harmonic force applied at
200 kHz excites primarily the fundamental s0 mode.
Finally, the case of a bonded joint is considered. The system is
represented by two 1.5875mm thick aluminum layers bonded by an epoxy
layer (thickness 0.2mm). The motion represented in Figure 3.25(a) shows
how the epoxy layer, that is assumed to have poor stiffness properties, is
subjected to a considerable shear deformation.
(a)
(b) p(t) (c)
x t
p(t)
z
Figure 3.23: (a) Deformation of a plate subjected to a uniform harmonic
vertical load (b), (c).
119
(a)
(b) p(t) (c)

p(t)
x t
Figure 3.24: (a) Deformation of a plate subjected to a uniform harmonic

horizontal load (b), (c).
(a)
xx
z Bond layer
(b) Bond
Bond
x
p(t)
Figure 3.25: (a) Deformation of a bonded joint subjected to a uniform

harmonic vertical load (b).
120
3.6.2 Response to arbitrary force
Equation (3.93) represents the response to a pure harmonic
excitation of unitary amplitude. The response to an excitation with
arbitrary time history can be computed as well. First the frequency
content of the excitation signal F (t) must be computed by applying the
Fourier transform:
∞
F (ω ) = ∫−∞
F (t)e−iω tdt (3.95)
The response in the frequency domain to the above force can be computed
simply as:
∑α
i ⎡⎣ ξm (x − xS )⎤⎦
V (x, ω ) = F (ω ) ⋅ U (x, ω ) = F (ω ) ⋅ ΦRup
mm e (3.96)
m =1
Finally, the time-domain response can be obtained using the inverse
Fourier Transform,
∞
1
V (x, t) =
2π ∫ V (x, ω )eiω tdω (3.97)
−∞
It should be noted that Standard Fast Fourier Transform algorithms can
be used to compute the integrals in Eqs. (3.95) and (3.97).
The procedure discussed was used for example in Chapter 7 to
compute the acceleration at one end in a steel bar due to a broadband
laser excitation (Section 7.2.2.1). In the same chapter, the displacement at
one end of a seven-wire steel strand, produced by a PZT actuator exciting
tone-burst pulses was predicted with the approach just discussed (Section
7.2.5).
121
3.7 Discussion and conclusions
In this chapter a general SAFE method was proposed and derived
for modeling wave propagation in waveguides of generally arbitrary cross-
sections. The main innovation over SAFE models proposed in the past is
accounting for viscoelastic material damping, and thus representing the
energy velocity and the attenuation curves. The hysteretic viscoelastic
model (frequency independent) and the Kelvin-Voigt viscoelastic model
(frequency dependent) were used in the formulation. The energy velocity
values (for damped waveguides) and the group velocity values (for
undamped waveguides) are obtained at each individual wavenumber-
frequency solution, without the need for tracking the modes and without
considering adjacent solutions in finite difference calculations. The
knowledge of the dispersive wave properties is relevant for NDE/SHM
testing for identifying propagating modes, locating defects, as well as
exciting low-loss mode-frequency combinations for increased ranges.
The method was validated on various examples, some of which were
examined previously by SPBW methods. The examples included plate
systems (viscoelastic orthotropic plates, laminated composite plates,
composite-to-composite adhesive bonds in UAV wings), arbitrary section
systems (railroad tracks) and axis-symmetric systems (coated pipe,
embedded rod).
The approximation of any SAFE method depends only on the
discretization of the waveguide’s cross-section. In the absence of rigorous
convergence criteria, the rule of thumb should be having the necessary
number of elements to properly represent the cross-sectional mode shapes
of the problem for the highest frequency of interest. Obviously the number
122
of elements to be used depends on the element type. For example, the
criterion exposed in Section 3.5.2 suggests to discretize the thickness of
plates and cylinders with quadratic elements smaller than 1/4 of the
minimum wavelength considered.
The solutions of the SAFE eigenvalue problem are the displacement
components. Consequently, if strain and stress components are of interest,
the mesh should be refined further to compensate for the loss of accuracy
of the differentiation process. This aspect is also important in the
evaluation of the energy velocity. Energy velocity is, in fact, a function of
the Poynting vector, Eq. (3.38), that, in turn, is a function of the stress
and the displacement derivatives, Eq. (3.39). Thus the accuracy of the
energy velocity solutions depends on the accuracy of the displacement
derivatives. Alternatively, the group velocity is a function of the
displacement eigenvectors, Eq. (3.37). In summary, the phase velocity, the
group velocity and the attenuation curves are calculated with the same
order of accuracy, while the energy velocity curves require a more refined
discretization to achieve the same accuracy.
In the following chapters some examples will be discussed where
guided waves have been extensively used as a structural health monitoring
tool.
3.8 Acknowledgements
Chapter 3, in part, has been published in the Journal of Sound and
Vibration, Bartoli, Ivan; Marzani, Alessandro; Lanza di Scalea, Francesco;
Viola, Erasmo (2006). The title of this paper is “Modeling wave

123
propagation in damped waveguides of arbitrary cross-section”. The
dissertation author was the primary investigator and primary author of
this paper.
OF COMPOSITE WING SKIN-TO-SPAR BONDED
JOINTS IN AEROSPACE STRUCTURES
4.1 Introduction
The first application discusses ultrasonic guided wave monitoring of
composite wing skin-to-spar bonded joints in aerospace structures.
Composite-to-composite joints representative of the wing skin-to-spar bonds
of Unmanned Aerial Vehicles (UAVs) are discussed. Two different lay-ups
have been examined for the wing skin and two different types of bond
defects, namely poorly-cured adhesive and disbonded interfaces have been
simulated. The assessment of bond state has been based on monitoring the
strength of transmission through the joints of selected guided modes. The
wave propagation problem has been studied numerically by the semi-
analytical finite element method accounting for viscoelastic damping, and
experimentally by ultrasonic testing that uses small PZT disks preferably
exciting and detecting the single-plate s0 mode.
Fiber-reinforced polymer composites have being increasingly utilized
in the aerospace industry due to their light weight and high strength. One
example of a heavy use of composites is found in Unmanned Aerial
Vehicles (UAVs), that are employed for both military and civil purposes
such as environmental monitoring. Adhesively-bonded joints are found in
the wing skin-to-spar assemblies of UAVs. These joints are critical
structural components, necessitating effective tools for assessing their
condition. A built-in structural health monitoring approach, rather than a
124
125
traditional NDT-type maintenance approach, would be most desirable for
this application.
Ultrasonic inspection of bonded joints has gone through decades of
improvement and has proven to be a very powerful method. Early studies
used normal-incidence ultrasonic testing to determine the modulus and
thickness of the bond layer through spectroscopy principles, monitoring the
through-thickness longitudinal resonances of the multilayered joint structure

(Guyott and Cawley, 1988a and 1988b). The reflection coefficients of bulk
ultrasonic waves at oblique incidence have also been shown to be sensitive
to bond conditions. A thorough theoretical analysis for this approach was
presented by Rokhlin and Wang (1991), where reflection coefficients were
predicted during the cure process of an epoxy resin layer between solid
semispaces.
An alternative ultrasonic bond testing approach uses guided waves
that exploit the natural waveguide geometry of most bonded aerospace
components. The guided wave method can be an effective bond diagnostic
tool due to its capability of long-range inspection as well as its flexibility
in selecting sensitive mode-frequency combinations. Generally, sensitive

combinations are those resulting in large strains/stresses at the bond layer.
In addition, through the use of built-in actuators and sensors, the guided
wave approach lends itself to the development of integrated systems for
continuous bond diagnostics as opposed to regularly-scheduled NDT
maintenance. Such an integrated health monitoring system is needed in
aerospace structures in order to implement a condition-based maintenance
philosophy.
126
There are two approaches traditionally used for guided wave
inspection of bonds. In the first approach, the waves are both generated
and detected in the bonded region (“within the bond” testing
configuration). In the second approach, the waves are generated in the
adherend on one side of the bond and received across the bond (“across
the bond” testing configuration). The main difference of the second
approach is the occurrence of mode conversion when the wave enters and
leaves the bond due to the transition from the single adherend geometry
to the bonded assembly geometry, and viceversa. Several previous studies
used the “within the bond” configuration to relate wave amplitude,
velocity and frequency to the elastic properties of the adhesive layer (Mal,
1988; Nagy and Adler, 1989; Mal et al., 1990; Xu et al., 1990; Pilarski
and Rose, 1992; Lowe and Cawley, 1994; Kundu and Maslov, 1997;
Chimenti, 1997; Kundu et al., 1998; Rose et al., 1998; Heller et al., 2000;
Cheng et al., 2001; Seifried et al., 2002). Recent applications of the
“within the bond” configuration have also examined the possibility of
inspecting the bond between a composite skin and a core in sandwich
aerospace panels (Castaings and Hosten, 2003; Hay et al., 2003). The
“across the bond” configuration was also used successfully for the
inspection of lap-shear joints, tear strap-to-skin joints, and bonded patch
repairs for damaged aircraft panels (Rohklin, 1991; Rose et al., 1995;
Chang and Mal, 1995; Mal et al., 1996; Lowe et al., 2000; Lanza di Scalea
et al., 2001; Sun et al., 2002; Lanza di Scalea et al., 2004).
These previous studies provided a great deal of knowledge on the
behavior and defect sensitivity of various guided wave modes propagating

127
in adhesively-bonded joints. However, none of these works examined the
case of composite-to-composite bonded joints, such as those found in wing
skin-to-spar bonds of UAVs. In addition, the majority of previous works
were not focused on built-in structural monitoring systems, an area that
has gained increasing interest only recently (Light et al., 2003). The use
of arrays of built-in piezoceramic elements (Pb(Zr-Ti)O3 - PZT) for

exciting and detecting ultrasonic guided waves goes back ten years (Keilers
and Chang, 1995). However, the preferred application has been the
detection of impact damage in FRP aerospace panels, rather than the
condition monitoring of bonds (Lakshmanan and Pipes, 1997; Wang and
Chang, 1999; Ihn et al., 2001; Lemistre and Balageas, 2001; Giurgiutiu and
Zagrai, 2002; Sohn et al., 2003; Giurgiutiu et al., 2003; Staszewski et al.,
2004).
In the following the propagation of ultrasonic guided waves in
composite-to-composite bonded joints as a function of bond condition is
investigated. The specific bond conditions considered are those of properly-
cured adhesive, poorly-cured adhesive and disbonds. The “within the bond”
configuration and the “across the bond” configuration have been examined
numerically by the semi-analytical finite element method that provides
modal solutions for the joints and accounts for the viscoelastic behavior of
the composites. The cross-sectional power flows of the carrier modes in the
bonded region (Poynting vectors), each weighted by an excitation factor
based on displacement mode shapes, have been used as the basis to
predict changes in the ultrasonic strength of transmission through the
different bond conditions. Experimental tests were conducted using PZT
discs bonded on two simulated wing skin-to-spar joints that were

128
constructed in the laboratory. The root mean square of the detected
signals was used to quantify the strength of transmission across the
varying bond conditions. Both the predictions and the experiments show
that the degradation of the bond has a marked effect on the strength of
transmission of the guided wave modes.
4.2 Problem statement and modeling formulation
A partial cross section of a typical UAV wing assembly is shown in
Figure 4.1(a). The wing skin is generally made of a Nomex or aluminum
honeycomb core sandwiched between two carbon fiber-reinforced plastic
(CFRP) laminated composite plates. Each skin is bonded using high
strength epoxy adhesive to a tubular composite spar that runs down the
length of the wing. The sandwiched skin tapers down in the bonded
region, where only the CFRP laminates are bonded to the spar. The
theoretical and experimental tests that follow were performed on CFRP
plates directly bonded to CFRP spars to represent the wing skin-to-spar
joint. The bond conditions that were examined included regions with
poorly-cured adhesive and disbonded regions, in addition to regions with

properly-cured adhesive. Analyses and experiments were performed in the
two testing configurations of “within the bond” and “across the bond,”
depicted in Figure 4.1(b).
Two different plate lay-ups were studied, namely a quasi-isotropic
[0/±45/90]S lay-up and a [0/±45/0]S lay-up, both consisting of eight plies
and bonded to a woven [0/90] tubular spar. The [0/±45/0]S lay-up is a
better representation of most UAVs, where the 0-deg fibers provide

129
flexural rigidity and the ±45-deg fibers provide torsional rigidity to the
wing. The quasi-isotropic lay-up was examined as a simplified model of
the UAV wing.
Modal solutions for the dispersive waves were obtained from the
SAFE approach described in the previous chapter. For the “within the
bond” case, Figure 4.1(a), the only model of interest was the bonded
region with the wave propagating along the lengthwise direction of the
spar.
90 deg. (b)
Spar
(a)
Actuator Sensor
Bonded joint 0 deg.
Actuator
Upper Wing Skin
Wing skin
Lower Wing Skin Lengthwise direction of wing Sensor

Spar
Figure 4.1: (a) Wing skin-to-spar bonded assembly. (b) Across bond and
within bond test configuration
130
For the “across the bond” configuration, Figure 4.1(b), the wing
skin model was examined separately in addition to the bonded region with
the wave propagating across the spar. In both cases the bond model
assumed a uniform thickness for the spar, neglecting the localized increase
in thickness at the bond edges found in the “across the bond” case due to
the tubular geometry. This approximation was considered acceptable given
that the frequency content of the waves examined was such that their
penetration beyond the wall thickness of the spar was negligible. Also,
mode conversion effects at the bond edges for the “across the bond” case
were predicted based on the similarities between cross-sectional mode
shapes of the excitation and the carrier modes found from the models,
following the same philosophy adopted in previous works (Auld, 1990;
Lowe et al., 2000, Lanza di Scalea et al., 2004).
The waveguide was modeled as a system of N homogeneous and
generally viscoelastic anisotropic layers. As shown in Figure 4.2(c), the
dimensions were considered infinite in the width direction, y, perpendicular
to the wave propagation direction, x.
Thus the wave propagation problem could be studied by simply
considering a longitudinal section in the x-z wave propagation plane. The
section was discretized in the thickness direction, z, by a set of one-
dimensional finite elements with quadratic shape functions and three nodes,
with three degrees of freedom per node. Under the hypothesis of time-
harmonic motion, the displacement vector u at any point within an
element can be written as

131
“WITHIN THE BOND” (a) “ACROSS THE BOND” (b)

PZT PZT PZT PZT
Wing skin
Adhesive
Spar
Lengthwise direction of wing
i th Quadratic Element
Ux1
Uy1
∞ x Uz1
Ux2
1 y i th
Uz2 Uy2 (c)
2 Ux3
3 ∞ Uz3 Uy3
Figure 4.2 Schematic of wave propagation for (a) “within bond” test
configuration, (b) “across bond” test configuration. (c) Diagram of SAFE
model and beam element adopted
⎡ 3 ⎤
⎢ ∑ N j (z) Uxj ⎥
⎢ j =1 ⎥
⎡ux ⎤ ⎢ ⎥
⎢ ⎥ ⎢ 3 ⎥
u(x, y, z, t) = ⎢uy ⎥ = ⎢⎢ ∑ N j (z) Uyj ⎥⎥ ei (kx − ω t) (4.1)
⎢ ⎥ ⎢ j =1 ⎥
⎢⎢u ⎥⎥ ⎢ 3 ⎥
⎣ z⎦
⎢ ⎥
⎢ ∑ N j (z) Uzj ⎥
⎢⎢⎣ j =1 ⎥⎥⎦
where t represents the time variable, Nj are the shape functions, k is the
wavenumber, ω is the circular frequency, and Uxj, Uyj and Uzj are the
nodal displacements along directions x, y, and z, respectively.
132
Each ply of the CFRP wing skin was modeled as a transversely
isotropic layer with the following standard constitutive laws in the global
laminate directions (x, y, z):
σ = C*ε = T1−1C*T2ε (4.2)
where C* is the complex stiffness matrix in the global directions of
the laminate, C* is the complex stiffness matrix in the principal material
directions of the individual ply, T1 and T2 are the transformation
matrices from the principal material directions to the global laminate
directions as defined in Jones (1975), and the stress vector σ and strain
vector ε are given by
σ = ⎡⎣σxx σyy σzz σyz σxz σxy ⎤⎦

T
(4.3)
ε = ⎡⎣ εxx εyy εzz γ yz γ xz γ xy ⎤⎦

T
(4.4)
The equations governing the wave propagation problem can be
obtained by introducing the following expression of the principle of virtual
works in absence of external forces:
∫ δ uT (ρ u)dV + ∫ δεT σdV = 0 (4.5)

V V
where the dot indicates differentiation with respect to time, ρ is the
density, and V is the volume of the waveguide. The components of the
vectors δu and δε represent the virtual displacements and the
corresponding virtual strains. By substituting in Eq. (4.5) the constitutive
laws of Eq. (4.2), the strain-displacement expressions, and introducing the

133
interpolation of Eq. (4.5), the following eigenvalue problem is finally
obtained for the entire structure after common finite element
manipulations:
(A − kB) Q = 0 (4.6)
where the matrices A and B have been defined in the previous Chapter.
When considering viscoelastic material damping the matrices A and
B become both complex as a result of the complex stiffness matrix C*.
Therefore all of the wavenumbers and the eigenvectors are complex. As
described in Chapter 3 the energy velocity (see Eq. 3.38), rather than the
group velocity, must be used as the only physically meaningful parameter
for viscoelastic media (Bernard et al., 2001). The expression for the energy
velocity for the case of the plate used in the present work is:
∫H P ⋅ xdz
1
Ve = H
( )
(4.7)
∫
1 1
T ∫T H H
Edz dt
where x is the vector of unit length along the wave propagation direction,
x, and 1/T∫T(..)dt denotes the average over the time period T. The
integral 1/H∫H(..)dz evaluates the average power computed from the
Poynting vector, P (real part only) and the average total energy (kinetic
and potential), E, over the entire waveguide thickness, H. The Poynting
vector can be calculated from the known relation:
⎛Px ⎞⎟
1 ⎜⎜ ⎟
P = − Re (συ ) = ⎜⎜Py ⎟⎟⎟
∗ (4.8)
2 ⎜⎜ ⎟⎟
⎝⎜Pz ⎠⎟
134
where σ is the classical 3×3 stress tensor. The numerator of Eq. (4.7) is of
particular interest for the results that follow. This term corresponds to the
average power flow in the direction of wave propagation carried by a
mode over a unit period of time, and it was used to predict the wave
transmission strength across the different bond conditions.
The cross-sectional distribution of strains were normalized by the
square root of the power flow through the entire thickness of a unit-width

waveguide and averaged over a temporal period, [∫HP• x dz]1/2. The cross-
sectional distributions of power flows were normalized by the square of

this factor, ∫HP• x dz. These normalizations were necessary to compare
results between different modes and frequencies.
Complex constitutive matrices were used to model viscoelastic
material damping in all components. The hysteretic viscoelastic model
introduced in Chapter 3 (see Eq. 3.4) was assumed where the imaginary
part of the stiffness matrix is independent of frequency. Thus:
Cij * = C′ij − iC′′ij (4.9)
The values of the real components, C′ij , and of the imaginary
components, C′′ij , of the stiffness matrix adopted in the models are shown
in Table 4.1. For the eight CFRP layers of the skin, the real components
of the stiffness matrix were based upon manufacturing specifications and
typical values for T300/5208 carbon epoxy. The imaginary components of
the stiffness matrix were assumed equal to the values used by Neau et al.
(2002).
The CFRP spar was modeled as one homogenous layer with
anisotropic properties equivalent to the multilayer [0/90] structure of

135
T800/924. The real and imaginary components of the stiffness matrix,
Ceq , for the equivalent spar layer were computed by averaging the
coefficients of the stiffness matrices of two adjacent layers, C0 and C90 ,
obtained from Eq. (4.2).
These equivalent coefficients were consistent with those obtained
from the independent approach of Karusena et al. (1991). The
homogeneous equivalent coefficients are accurate when the number of
layers is high as it was the case for the test spar.
Table 4.1: Real and imaginary stiffness coefficients, geometric and physic
properties for bonded multilayer models
C'11 [GPa] C'12 [GPa] C'13 [GPa] C'22 [GPa] C'23 [GPa] C'33 [GPa] C'44 [GPa] C'55 [GPa] C'66 [GPa] Density Thickness
Layer
(C''11) [GPa] (C''12) [GPa] (C''13) [GPa] (C''22) [GPa] (C''23) [GPa] (C''33) [GPa] (C''44) [GPa] (C''55) [GPa] (C''66) [GPa] [kg/m3] [mm]
135 5.70 5.70 14.2 8.51 14.2 2.87 4.55 4.55
CFRP Lamina (8.23)* (0.65)* (0.60)* (0.34)* (0.25)* (0.65)* (0.24)* (0.28)* (0.25)* 1530 0.133
88.0 5.45 5.09 88.0 5.09 11.3 4.64 4.64 6.00

CFRP Spar (4.28) (0.65) (0.425) (4.28) (0.425) (0.65) (0.26) (0.26) (0.25) 1530 5.235
Properly-cured 8.24 4.10 4.10 8.24 4.10 8.24 2.07 2.07 2.07
Bond (0.39) (0.028) (0.028) (0.39) (0.028) (0.39) (0.18) (0.18) (0.18) 1421 0.203
Poorly-cured 6.89 4.58 4.58 6.89 4.58 6.89 1.16 1.16 1.16
Bond (0.19) (0.064) (0.064) (0.19) (0.064) (0.19) (0.066) (0.066) (0.066) 1465 0.203
0.0697 0.0695 0.0695 0.0697 0.0695 0.0697 0.000118 0.000118 0.000118

Disbond 1421 0.203
(0.0352) (0.0349) (0.0349) (0.0352) (0.0349) (0.0352) (0.000128) (0.000128) (0.000128)
* From Neau et al., 2002
Table 4.2: Ultrasonic bulk longitudinal and shear velocities and material
attenuations for the bond.
Layer CL [m/s] CT [m/s] αL [Np/λ] αT [Np/λ]
Properly-cured
2410 1210 0.149 0.276
Bond
Poorly-cured
2170 890 0.089 0.178
Bond
Disbond 241 12.1 1.497 2.763
136
The epoxy adhesive was modeled as a viscoelastic isotropic layer. In
this case the viscoelastic matrix, C*, depends only on the two elastic
constants, Young’s modulus, E, and shear modulus, G, that were
calculated from the bulk longitudinal and shear wave velocities, cL* and
cS*. These complex velocities were calculated by the known expressions:
−1
⎛ ⎞
cL,S = cL,S
* ⎜⎜1 + i αL,S ⎟⎟ (4.10)
⎜⎝ 2π ⎠⎟⎟
where αL,S are the longitudinal and shear attenuation in the material,
expressed in Nepers per wavelength. The attenuation and elastic constants
for the properly-cured epoxy and for the poorly-cured epoxy were obtained
through normal-incidence ultrasonic tests on properly-mixed and poorly-
mixed bulk epoxy samples. The poorly-cured epoxy resulted in a 42%
degradation of the Young’s modulus and a 44% degradation of the shear
modulus compared to the properly-cured epoxy. The disbond was simulated
in the model by reducing the shear wave velocity (real part) of the
properly-cured adhesive by a factor of 100, reducing the longitudinal wave
velocity by a factor of 10, and increasing the longitudinal and shear
attenuation by a factor of 10. The properties assumed for the various
bond conditions are summarized in Table 4.1 in terms of stiffness
coefficients, and in Table 4.2 in terms of bulk ultrasonic velocities and
attenuations.
137
10000
00
10000
000
00 A1 000
(a) (b)
8000
00
8000
000
S0

00 S1 000
S0 A1
6000
00
6000
000
00
SH1 000
4000
00
4000
000 S0
A1
00
SH 0 000
SH 0
2000
00
2000
000
00 A0 A0
SH1 S1
000
000 50 100 150 200 250 300 000

0 50 100 150 200 250 300 50 100 150 200 250 300
0 50 100 150 200 250 300
Frequency [KHz]
Frequency [KHz]
200200
180
[0/±45/0]S Skin
160160
Attenuation [Np/m]
140
120120 S1
A1 (c)
100
Spar
8080
60 SH1 Properly-cured
A0 Bond
4040 SH0 A0
20 S0
0 00 50 100 150 200 250 300
0 50 100 150 200 250 300
A1
Frequency [KHz] S0
Figure 4.3: (a) Phase velocity, (b) energy velocity, (c) and attenuation
dispersion results for well bonded [0/±45/0]S composite plate to composite
spar joint. Wave propagation is along 90 degree direction relative to plate
fiber orientation.
Convergence of the dispersion solutions was found satisfactory using one

element for each ply in the wing skin, one element for the bond layer,
and five elements for the spar. The number of elements was doubled to
plot displacement, strain, and power flow cross-sectional shapes.
4.3 Dispersion results for different bond states
Phase velocity, energy velocity and attenuation curves were obtained
from the SAFE models for the three bond conditions examined (properly-
138
cured bond, poorly-cured bond and disbonded interface), each under the
two testing configurations (“within” and “across” the bond) and the two
wing skin lay-ups ([0/±45/90]S and [0/±45/0]S).
For the ease of the reader, throughout this paper the conventional
nomenclature of symmetric or antisymmetric character is used for all
guided modes. However, the only true symmetric and antisymmetric modes
exist when considering just the wing skin plate. The modes in skin-to-
spar bond are not truly symmetric nor antisymmetric because the cross-
section is not symmetric about its mid-plane. Modes propagating in the
single skin plate will be indicated with a lower case, si, ai and shi for the
symmetric, the antisymmetric, and the shear horizontal modes, respectively.
Modes propagating in the bonded region will be indicated with an upper
case, Si, Ai and SHi. Due to the presence of the ±45 deg plies, the
horizontally and vertically polarized partial waves are generally not de-
coupled, and thus the dispersion curves presented always include the shear
horizontal modes.
The results for the [0/±45/0]s plate bonded to the spar with the
properly-cured adhesive are shown in Figure 4.3 considering wave
propagation in the “across the bond” test configuration. The frequency
range shown is DC-300 kHz that was the operating range of the
experimental tests. The four modes of interest here are the zero-order
symmetric, S0, the zero- and first-order antisymmetric, A0 and A1, and the
zero-order shear horizontal, SH0.
Figure 4.3(b) and (c) show that A1 is propagative only above 135
kHz whereas the other three modes are propagative throughout the
frequency range. Although other higher-order modes exist with cut-on

139
frequencies above 135 kHz, Figure 4.3(b), they are not considered further
due to their large attenuation, Figure 4.3(c).
The two modes with minimum attenuation losses are S0 below 200
kHz and A1 above 200 kHz, Figure 4.3(c). Thus these two modes appear
preferred candidates for a bond monitoring system within the frequency
range examined. Moreover, PZT disks used for built-in ultrasonic structural
diagnostics are typically operated in the 3-1 electro-mechanical coupling
mode, i.e. to generate and receive in-plane, rather than out-of-plane
displacements (Giurgiutiu et al., 2003). The cross-sectional mode shapes
presented in the next section indicate that both S0 and A1 have

substantial in-plane displacements at the skin plate surface, and they are
thus coupled very effectively to the built-in PZTs. The opposite is true for
either A0 (predominant out-of-plane displacement) or SH0 (zero in-plane
displacement along the wave propagation direction).
A closer look at Figure 4.3 reveals a mode coupling effect occurring
for both S0 and A1 at around 200 kHz. Mode coupling is a known
phenomenon that can be caused by damping effects present in a single
layer (Bernard et al., 2001) or solely by geometrical effects in undamped
multilayered structures (Castaings and Hosten, 2003). In this study mode
coupling is due to both viscoelastic damping and multilayer geometrical
effects, as seen in Figure 4.3.
In this figure the phase velocity results are obtained from the SAFE
model of the bonded joint with and without damping losses. The
examination of the displacement mode shapes of S0,undamped and A1,undamped
reveals that apparently different branches on either side of the 200 kHz
frequency value are indeed the same mode even in the undamped case. In
140
other words, the branches are already coupled in the undamped case
although they are not physically connected around 200 kHz. Once
damping is included in the model, S0 and A1 essentially retrace the
corresponding S0,undamped and A1,undamped, and connect the branches at the
mode coupling points around 200 kHz.
Mode coupling is generally associated to sharp changes in group or
energy velocity, increased attenuation, and large transfer of energy across
the thickness of the waveguide. The first two phenomena are clearly
visible in Figure 4.4(b) and (c) for both S0 and A1 at around 200 kHz.
The third phenomenon is seen in the plots of Figure 4.4(b), (c) and (d),
comparing the normalized through-thickness power flow, Pz, of S0 at the
three frequencies of 155 kHz, 205 kHz and 255 kHz, respectively. At the
mode coupling frequency of 205 kHz, Figure 4.4(c), a large exchange of
energy occurs between the top skin and the bottom spar. Contrarily, most
of the energy flow occurs within the bottom spar at 155 kHz, Figure
4.4(b), and within the top skin at 255 kHz, Figure 4.4(d), with little
interlayer flow. Similar Poynting vector results, not shown here, can be
found for A1. The large transfer of energy in the thickness direction is
particularly relevant for a bond monitoring system as the one proposed
here, that is based on relating bond defects to an increased strength of
transmission through the joint. Exciting modes with large energy transfer
between adherents, such as S0 and A1 at mode coupling frequencies, would
clearly be beneficial in this case.

141
[0/±45/0]S Skin
x
y
z
Spar
Properly-cured
Bond
(a)
(b) (c) (d)
Figure 4.4: (a) Phase velocity dispersion curves for the [0/±45/0]S skin-to-
spar joint (“across the bond” testing configuration) with damping and
without damping. Through-thickness Poynting vector for S0 at (b) 155
kHz, (c) 205 kHz, and (d) 255 kHz.
142
The dispersion curves for the case of the poorly-cured bond, where
only the material properties within the thin bond layer were altered
according to Table 4.2, did not show any notable change from those of
the properly-cured bond. This is because the thickness of the bond layer
was very small compared to the thickness of the entire skin-to-spar
assembly that dominates the dispersion. However, substantial changes in
the wave power flow between the two cases were predicted, and this will
be discussed in detail in the following sections.
10000 10000
A1,spar (a) (b)
8000 S0,spar 8000 S0,spar

6000 S0,plate
6000
SH0,plate A1,spar
S0,plate
4000 4000
A0,plate A
SH 0,spar A0,spar SH0,plate 0,spar A0,plate
SH0,spar
2000 2000
0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
80
Single [0/±45/0]S plate
70 (c)
Disbonded joint
A0,plate
60
Attenuation [Np/m]
50
A1,spar a0 [0/±45/0]S Skin
40 A0,spar
30
SH0,plate
20
SH0,spar
10 S0,plate Spar
Disbond
S0,spar
0
0 50 100 150 200 250 300
Frequency [KHz]
Figure 4.5: (a) Phase velocity, (b) energy velocity, and (c) attenuation
curves for the [0/±45/0]S skin-to-spar joint with a disbonded interface and
for the single [0/±45/0]S plate (“across the bond” testing configuration).
143
As for the disbond case, shown in Figure 4.5(a)-(c), dramatic
changes in the phase velocity, energy velocity and attenuation curves are
evident. The main observation in this case is the appearance of additional
modes that did not exist for either of the two bond cases discussed
previously. For most of the frequency range examined, the additional
modes of the disbonded joint essentially coincide with the solutions of the
single wing skin plate, represented in the plots by the open dots.
Therefore, the introduction of a disbond allows for the propagation of two
separate types of modes, namely those whose energy is mainly
concentrated within the upper plate above the bondline (identified in
Figure 4.5 by S0,plate, A0,plate, SH0,plate , etc.), and those whose energy is
mainly concentrated within the spar below the bondline (identified by
S0,spar, A0,spar, SH0,spar, etc.). Mode S0,spar of the disbonded joint corresponds
to mode S0 of the properly-cured joint. Similarly, A1,spar corresponds to A1
of the properly-cured joint. A comparison of Figure 4.3 to Figure 4.5
indicates that the character of S0 does not change substantially between
the two bond cases. However, the mode coupling for A1 disappears when a
disbond is present. Under the testing conditions used in this study, the
predominant carrier of energy through the disbonded joint is S0,plate, that
should thus be considered the most relevant mode in Figure 4.5. The
dispersion curves of S0,plate in the disbonded joint are very similar to those
of the single-plate s0 mode above 100 kHz. Below this frequency, the mode
deviates from the single-plate behavior as more energy leaks through the
spar.
144
4.4 Identification of Carrier Modes
Mode conversion effects as the wave travels through the joints were
evaluated on the basis of the similarity between the displacement mode
shapes of an incoming mode and those of a carrier mode. Predictions of
energy transmission across the bond in the “across the bond” test
configuration were based upon initial excitation of the s0 mode and
subsequent mode conversion in the bonded region. Likewise, the predictions
for the “within the bond” test configuration were based upon initial
excitation of those overlap modes with similar modes shapes to those of
the single-plate s0 mode. These assumptions were consistent with the in-
plane (3-1) electro-mechanical coupling characterizing the operation of the
PZT disks employed in the experiments.
Figure 4.6(a) shows the displacement mode shapes for the [0/±45/0]S
skin-to-spar bond tested in the “across the bond” configuration at the
mode coupling frequency of 205 kHz. The left-hand plot in this figure
shows the incoming s0 mode in the wing skin. The plots to the right show
the four possible carrier modes in the bond, namely S0, A0, A1 and SH0,
considering the properly-cured bond condition. The notation for the
displacement components is consistent with the reference system in Figure
4.2(c). The dominant displacement for s0 is the in-plane component in the

wave propagation direction, ux, that thus also dominates the mode
conversion process. The ux displacement is also the dominant component
for the S0 and A1 carrier modes, and its symmetry within the upper plate
in both carrier modes is similar to that of the incoming mode. Conversely,
ux is not dominant for A0 and it is altogether absent for the shear

145
horizontal SH0. It is thus suggested that S0 and A1 will act as the

primary energy carriers across the joint at this frequency.
Figure 4.6(b) shows the cross-sectional strain components in the
plane of propagation of the wave for the two carrier modes S0 and A1 at
205 kHz.
(a)
(b)
(c)
Figure 4.6: The [0/±45/0]S skin-to-spar joint in the “across the bond”
testing configuration: (a) displacement mode shapes at 205 kHz in the
properly-cured bond, (b) strain profiles for modes S0 and A1 at 205 kHz in
the properly-cured bond, (c) displacement mode shapes at 205 kHz in the
disbonded joint. Also shown are the corresponding mode shapes for the
incoming s0 mode in the single plate (a) and the S0 and A1 modes in the
properly cured joint (c).
146
To emphasize the bonded region, only the upper portion of
composite spar wall is shown. It can be seen that both modes show a
concentration of both normal strain, εzz, and shear strain, γxz, within the
adhesive layer.
Thus both modes are suitable candidates for detecting changes in
both longitudinal and shear stiffness of the adhesive layer. Between the
two modes, it is S0 that produces the larger concentration of strains at the
bondline, suggesting a larger sensitivity to bond conditions. This result will
be confirmed by power flow considerations in the next section.
Similar results were found when analyzing the mode shapes in the
poorly-cured joint. This is because the changes in the bond stiffness that
were considered had little influence on the shapes of the modes that
remained dominated by the skin and the spar adherends.
The next step is the case of wave propagation across a central
disbonded portion of the skin-to-spar joint (again, [0/±45/0]S skin lay-up in

the “across the bond” configuration). The s0 mode is still assumed to be
initially excited outside of the bond region. Mode conversion into the S0
and A1 modes occurs at the joint boundary. In turn, these two modes will
mode convert once they reach the disbond. Figure 4.6(c) shows the S0 and
A1 mode shapes for the properly-cured region together with the S0,spar,
S0,plate and A1,spar mode shapes for the disbonded region at 205 kHz. For
S0,spar and A1,spar the largest displacements occur below the bondline in the
spar wall. Conversely, the largest displacements occur in the top plate for
S0,plate. Since the initial wave excitation is in the top plate, it is expected
that the majority of the wave energy will remain confined to this
component for short propagation distances. In the experimental setup

147
adopted the propagation distance to a centralized disbond was less than
half of either S0 or A1 wavelengths at 205 kHz. Thus it can be assumed
that S0,plate is the predominant carrier mode through the disbonded region.
4.5 Strength of Transmission as a Function of Bond Condition
The change in strength of transmission as a function of bond
condition was predicted from the cross-sectional averaged power flow
computed by the numerator of Eq. (4.7). This quantity was divided by

the normalization factor ∫HP• x dz to enable direct comparisons between
the various bond conditions and carrier modes.
Figure 4.7 shows the real Px component profiles for the [0/±45/0]S
skin-to-spar joint probed in the “across the bond” configuration for the
different bond conditions. The properly-cured bond and the poorly-cured
bond are compared in the top four plots for the two carrier modes S0 and
A1. Figure 4.7(a) and (b) refer to S0 propagating at 155 kHz and 205
kHz, respectively. It can be seen that the S0 strength of transmission is
larger in the poor bond as compared to the good bond. This phenomenon
manifests itself mainly within the upper plate layer. The power flow profile
justifies the commonly identified reduction in “energy leakage” that occurs
when guided waves are transmitted across poor adhesive joints. It is also
evident in Figure 4.7(a) and (b) that the normalized power flow for a
given bond condition is larger at 205 kHz than it is at 155 kHz, and the
increase in transmission strength with degrading bond conditions is also
larger at 205 kHz than it is at 155 kHz. The change in transmission
strength between the two bond conditions in much less pronounced for the
148
A1 carrier mode shown in Figure 4.7(c) and (d). In fact, a slight decrease
in transmission strength within the upper plate occurs at 205 kHz as the
bond is degraded. It can be further deduced that the S0 mode would be
primarily responsible for the sensitivity to bond stiffness based on
transmission measurements. This conclusion is consistent with the cross-
sectional strain profiles of Figure 4.6(b) where S0 is seen to produce larger
strains than A1 at the bond layer.
As for the disbonded case, shown in Figure 4.6(e) and (f) for 155
kHz and 205 kHz, the power flow results confirm that S0,plate is completely
confined to the top plate. As a result, the strength of transmission should
be expected much larger than what seen in either of the two carrier
modes, S0 or A1, in the previous two bond conditions. Significant changes
in the Px component of the Poynting vector in the disbonded case were
not observed over the frequency range of interest. In fact, the result at
155 kHz in Figure 4.6(e) is indistinguishable from the one at 205 kHz in
Figure 4.6(f).
The cross-sectional power flow information was used to predict
changes in transmission strength as a function of bond condition over the
frequency range 100 kHz – 300 kHz, for both plate lay-ups ([0/±45/0]S and
[0/±45/90]S) and both testing configurations (“within” and “across” the
bond).
149
Figure 4.7: Px component of the Poynting vector in the properly-cured

bond and in the poorly-cured bond for (a) S0 at 155 kHz, (b) S0 at 205
kHz, (c) A1 at 155 kHz, and (d) A1 at 205 kHz. Same quantity in the
disbonded case for (e) S0,plate at 155 kHz, and (f) S0,plate at 205 kHz.
150
Figure 4.8(a) and (b) plot the strength of transmission in terms of
normalized power flow, averaged over the thickness of the upper plate and
bond layer. This quantity was calculated by:
⎡ ⎤
⎢ Px ⎥
⎢ ⎥dz
P = ∫ (4.11)
h ⎢
⎢⎢
⎣
( ∫H P i x )
dz
⎥
⎥⎥
⎦
where h is the combined thickness of the upper plate and the bond layer.
Figure 4.8(a) compares A1 and S0 propagating across the properly-cured
and the poorly-cured bonds, for the [0/±45/0]S lay-up. It can be seen that
S0 is the main energy carrier in most of the frequency range, with its
transmission strength increasing with degrading adhesive properties. A1 is
seen to have a minor role in the energy transmission and it is also seen
less sensitive to the change in bond condition. Also notice that the A1
contribution is eliminated below the 135 kHz cut-on frequency. The same
general behavior can be observed for the other plate lay-up, [0/±45/90]S,
shown in Figure 4.8(b). The main difference is this case is the increased
transmission strength of A1 with degrading adhesive properties below 200

kHz.
The sensitivity to bond condition was compared directly by
computing the relative differences in transmission strength between the
degraded bonds and the properly-cured bond. First, the relative excitability
of the individual carrier modes needed to be taken into account
considering the single incoming mode, s0, for the “across the bond”
configuration. An excitation factor for each potential carrier mode was
evaluated based on the similarity between the displacement mode shapes of

151
the incoming mode and those of the carrier mode. These displacements

were normalized by the factor [∫HP• x dz]1/2 in order to capture their
relative magnitude at the various frequencies and for the various modes.
The maximum of the cross-correlation function was used as the similarity
index. The following Excitation Factor was defined for an i-th carrier
mode (CMi) under a single incoming mode (IM):
Excitation Factor (f ) =
∑ k = x, y, z
Max uk, IM (f ) ⊗ uk, CMi (f )
(4.12)
∑ ∑
CMi N
i =1 k = x, y, z
Max uk, IM (f ) ⊗ uk, CMi (f )
where ⊗ is the cross-correlation operation between the k-th cross-sectional
displacement component of the incoming mode, uk,IM, and that of the i-th
carrier mode, uk,CMi, computed at the various frequencies and in the
upper plate only. In Eq. (4.12) the Excitation Factor is normalized by
the contribution of all possible N carrier modes existing at the incoming
frequency. For the subject tests, the excitability remains dominated by the
ux displacement component overweighing the other components of the
incoming s0 mode. The values of the Excitation Factor for the properly-
cured and the poorly-cured bonds were close to 0.5 for both S0 and A1 in
the 100 kHz – 300 kHz range, with minima and maxima at 0.46 and 0.55,
respectively.
Generally, the transmission strength will also be affected by
attenuation losses. This effect was only factored explicitly into the
contribution of the A1 mode below 135 kHz where a zero Excitation

Factor was assumed to reflect non-propagative conditions (Figure 4.2).
152
Across Bond
(a) [0/±45/0]s (b) [0/±45/90]s
1 1
- - - Properly-cured - - - Properly-cured
Normalized Power Flow, P
Normalized Power Flow, P

S0 Poorly-cured Poorly-cured
0.8 0.8 S0
0.6 0.6
A1
0.4 A1 0.4
0.2 0.2
0 0
100 150 200 250 300 100 150 200 250 300
Frequency (KHz) Frequency (KHz)
Across Bond
(c) [0/±45/0]s (d) [0/±45/90]s
(Poorly-cured) – (Properly-cured)
Relative Transmission Strength

0.30 0.30
Disbond – (Properly-cured)
Poorly-cured - Properly-cured 8 8
0.25 0.25
0.20 6 0.20 Poorly-cured - Properly-cured 6
0.15 0.15
4 4
0.10 Disbond - Properly-cured 0.10
2 2
0.05 0.05
Disbond - Properly-cured
0.00 0 0.00 0
100 150 200 250 300 100 150 200 250 300
(e) [0/±45/0]s (f) [0/±45/90]s


0.30 0.30
8 8
0.25 0.25
0.20 Disbond - Properly-cured 6 0.20 Poorly-cured - Properly-cured 6
0.15 0.15
4 4
0.10 0.10
2 2
0.05 0.05
Poorly-cured - Properly-cured Disbond - Properly-cured
0.00 0 0.00 0
100 150 200 250 300 100 150 200 250 300
Figure 4.8: Above spar power flow of S0 and A1 modes for well cured and
poorly cured cases. 90 deg. wave propagation direction in (a) the
[0/±45/0]s and (b) the [0/±45/90]s bonded composite plate. Power flow
difference results for both defects in “across bond” configuration of (c) the
[0/±45/0]S (d), [0/±45/90]s and for “within bond” configuration of (e) the
[0/±45/0]s and (f) the [0/±45/90]s multilayer models.
Above 135 kHz, since the attenuation values for A1 and S0 were close, the
relative weights of Eq. (4.12) were considered for the perfectly elastic case
without loss of accuracy.

153
For the “within the bond” configuration, values of 0.5 were
maintained for the Excitation Factors of S0 and A1 (with a zero value for
A1 below 135 kHz), whereas an Excitation Factor of 1 was assigned to the
S0,plate mode, being the only one relevant for the disbonded case.
Finally, the excitation-adjusted power flow differences, computed
after weighting P of Eq. (4.11) with the respective Excitation Factors, are
shown in the plots of Figure 4.8(c) through (f). These results were also
normalized by the maximum power flow of the properly-cured bond to
provide a relative transmission change. There are three main observations

that can be made from these plots. First, the transmission strength is
always larger for the degraded bond cases relative to the properly-cured
bond for each plate lay-up and direction of propagation considered. Notice
that the jump at 135 kHz in all plots is due to the zero weight assigned
to A1 below this frequency. Second, the difference in transmission strength
for the disbonded joint is larger than that for the poorly-cured bond in all
cases (notice the different values in the left-hand and the right-hand axes).
This is a consequence of the unique character of the S0,plate mode that
carries the incoming s0 mode energy very efficiently. It should be reminded

that the current simulation considers the limit case of a disbond extending
for the entire joint width. Clearly, the effect will be much reduced for a
localized disbond within a joint as is the case for the experimental tests
that follow. Third, comparing the poorly-cured to the properly-cured bond,
an abrupt increase in sensitivity is seen to occur around 200 kHz in all
cases. The increase in sensitivity is due to the mode coupling phenomenon
discussed earlier, affecting both S0 and A1 near this frequency value. The
increased energy transfer through the thickness of the waveguide at the
154
mode coupling frequencies is the physical basis for the abrupt increase in
sensitivity.
4.6 Experiments on simulated wing skin-to-spar joints
4.6.1 Test Specimens
Two specimens representative of UAV skin-to-spar joints were

fabricated in the laboratory. The specimens matched closely the models
analyzed numerically. One of the fabricated joints is shown in Figure 4.9.
Two composite plates were fabricated using T300/5208 carbon epoxy
prepreg with the two lay-ups of [0/ ± 45/0]S and [0/ ± 45/90]S. Each plate
had a total thickness of 1.067 mm. The plate dimensions were cut to 330
mm × 330 mm. The composite piece used to replicate the wing spar was
a woven T800/924 carbon epoxy square tube with nominal outer
dimensions of 50.8 mm × 50.8 mm and a wall thickness of 6.35 mm. The
composite tube was cut into two, 330 mm long sections to match the
length of each plate. A two-part Hysol 9394 epoxy adhesive was used to
bond the spar to the individual plates. Each of the two specimens was
manufactured identically with the exception of the different plate lay-ups.
Bonding of the plates was done such that the 0-deg fiber direction ran
along the lengthwise direction of the spar.
The same type of bond conditions considered in the model were
artificially created prior to assembling the joints (Figure 4.9). The majority
of the bond was comprised of a properly mixed epoxy, representing the
well-bonded region. In order to form a region of degraded bond stiffness,
an improperly mixed sample of epoxy was prepared resulting in an

155
approximate 50% stiffness degradation compared to the properly-cured case
(the exact values from ultrasonic through-transmission testing of the mix
were 42% degradation for the longitudinal stiffness and 44% degradation
for the shear stiffness as reflected by the velocity values in Table 4.2).
Finally, two disbonded regions of different sizes were created by inserting
Teflon release film with a thickness of 0.025 mm at the centerline of the
bond. The release film was expected to severly degrade the shear stiffness
of the bond while degrading its longitudinal stiffness to a smaller extent.
The two simulated disbonds had dimensions of 12.7 mm × 12.7 mm and
25.4 mm × 25.4 mm, respectively. The total bond thickess was measured
to be 0.203 mm in both specimens.
Properly- Poorly-cured
(a)
cured bond bond
disbonds
(b)
CFRP plate
CFRP spar
Figure 4.9: [0/±45/0]s plate bonded to spar (a), top view (b), side view
156
4.6.2 Experimental setup and procedure
Guided wave testing was performed in the joints with the objective
of relating changes in energy transmission strength to the presence and
type of the simulated bond defects. PZT disks with a 12.7 mm diameter
were used as wave actuators and sensors for the “across the bond” test
configuration and the “within the bond” test configuration. A sensor
spacing of 84 mm and 70 mm were used for the former configuration and
for the latter configuration, respectively. Two PZTs are shown in the
pictures of Figure 4.9. The disks were bonded to the structure using a
thermally-activated film adhesive. To assist in the normalization of the
testing procedure, the same pair of sensors was used throughout the tests.
Tests run on a single composite plate confirmed that s0 was the mode
generated and detected by the PZTs with the greatest efficiency.
The signal generation and data acquisition system was a National
Instruments PXI-1010 running under LabVIEW that was assembled and
programmed in-house. The system used an arbitrary function generator to
allow for swept frequency tests using Hanning windowed tonebursts sent to
the actuating PZT. The frequency sweep was performed between 100 kHz
and 300 kHz, stepping in 1 kHz increments. At each frequency, the root
mean square (RMS) of the received time-domain signal (corresponding to
the energy of the signal) was computed following the known relation:
∑x
i =1
i
2
(4.13)
RMS =
Z
where xi is a single measure within a collection of Z measurements.

157
Sample waveforms collected in the “within the bond” test of the
[0/±45/0]S skin-to-spar joint are shown in Figure 4.2 at 205 kHz
corresponding to the properly-cured bond, the poorly-cured bond, and the
large disbond. The time window where the RMS of the signal was
evaluated is identified on the plot by two vertical lines. Notice that the
signal energy within this time gate is larger for the two degraded bonds
relative to the properly-cured bond. The time gates were maintained
constant for each frequency sweep in a given test configuration.
The relative change in energy transmission was quantified by taking
the RMS spectra difference between each defected bond and the properly-
cured bond, and then normalizing the results by the RMS spectrum of the
properly-cured bond. This procedure enabled a qualitative comparison with
the power flow results from the model in Figure 4.2(c) through (f).
4.6.3 Experimental results
The relative RMS changes as a function of bond state measured in
each test are shown in Figure 4.11. It can be seen that each of the
spectra are positive, and thus the energy transmission is strengthened in
the presence of any of the simulated bond defects. This general
observation is consistent with the predictions from the models in Figure
4.8. It can also be seen in the experimental results that the peak relative
RMS change occurs at around 200 kHz in all cases, and thus this
frequency value is confirmed as the most sensitive to the bond defects
considered in this study. This result was predicted by the model based on
mode coupling occurring for S0 and A1 that are the primary energy
carriers for the properly-cured and for the poorly-cured bonds.
158
Other identifiable trends can be observed in Figure 4.11 when
comparing the results amongst each defect case. First, in three of the
four test sets, the strength of transmission for the poor bond is larger
than that for the two disbonds across the entire frequency spectrum. The
only exception is for the [0/±45/90]S in the “across the bond” test shown
in Figure 4.11(b). Alternatively, the predictions from the model in Figure
4.8 indicate that the power flow difference is larger for the disbond case
compared to the poor bond case. However, issues regarding the defect size
were not included in the model (that assumed a disbond extended for the
entire width of the joint). Consequently it is expected that the strength of
transmission will be somewhat reduced from the predicted values when the
disbonds only extend for a portion of the joint.
Figure 4.10: Typical waveforms from the “within the bond” test of the
[0/±45/0]S skin-tospar specimen at 205 kHz.
159
Across Bond
(a) [0/±45/0]s (b) [0/±45/90]s
1.2 1.2
Poorly-cured - Properly-cured 25.4mm x 25.4mm
1 1 Disbond - Properly-cured
Normalized RMS
Normalized RMS
Difference
Difference
0.8 25.4mm x 25.4mm 0.8 12.7mm x 12.7mm
Disbond - Properly-cured Disbond - Properly-cured
0.6 0.6
0.4 0.4
12.7mm x 12.7mm
0.2 0.2
Disbond - Properly-cured Poorly-cured - Properly-cured
0 0
100 150 200 250 300 100 150 200 250 300
Within Bond
(c) [0/±45/0]s (d) [0/±45/90]s
1.2 1.2
Poorly-cured - Properly-cured
1 1 Poorly-cured - Properly-cured
Normalized RMS
Normalized RMS
Difference
Difference
0.8 25.4mm x 25.4mm 0.8 25.4mm x 25.4mm

Disbond - Properly-cured
0.6 Disbond - Properly-cured 0.6
0.4 0.4
12.7mm x 12.7mm 12.7mm x 12.7mm
0.2 0.2
Disbond - Properly-cured Disbond - Properly-cured
0 0
100 150 200 250 300 100 150 200 250 300
Figure 4.11: Normalized RMS difference results for each defect in “across
bond” configuration of (a), [0/±45/0]s (b), [0/±45/90]s and for “within
bond” configuration of (c), [0/±45/0]s (d), [0/±45/90]s test specimen.
Second, in every test set of Figure 4.11 the relative RMS difference
associated with the large disbond is larger than that associated to the
small disbond, with the exception of the usual [0/±45/90]S “across the
bond” test above 220 kHz. That the strength of transmission increases
with increasing disbond size is due to the favorable Px component of the

Poynting vector of the S0,plate mode (Figure 4.7(e) and (f)) that isolates
the majority of the wave energy within the same upper plate where the
PZT actuator and sensor are located.
Third, the relative RMS difference for each defect is of the same
order of magnitude in each test. The smallest RMS differences are seen for
160
the [0/±45/90]S “across the bond” test. Thus this test is the one with the
smallest sensitivity to bond state. In terms of discriminating between the
defect cases, the [0/±45/90]S “across the bond” test remains the least
favorable since the three defects yield close results. The testing
configuration providing maximum discrimination between the poor bond
and the disbond defects changes depending on the plate lay-up: the
“within the bond” test is the most discriminating for the [0/±45/90]S lay-
up, Figure 4.11(d), whereas the “across the bond” test is the most
discriminating for the [0/±45/0]S lay-up, Figure 4.11(a).
4.7 Summary and conclusions
The subject of this chapter is the propagation of ultrasonic guided
waves across composite adhesively-bonded joints. The specific component
investigated was representative of a skin-to-spar joint of UAVs. The
research is part of a broader project aimed at developing a built-in
diagnostics system for monitoring the structural integrity of UAVs. Two
lay-ups for the composite skin were investigated, namely a quasi-isotropic
[0/±45/90]S lay-up and a [0/±45/0]S lay-up. Two types of bond defects
were considered, namely a poorly-cured bond (~50% drop in longitudinal
and shear stiffness), and a disbond where the shear stiffness was nominally
lost.
The first portion of this study demonstrated the application of a
semi-analytical finite element method that included viscoelastic material
damping to predict modal solutions for the joints when probed “across”
161
and “within” the bond. An excitation factor for the possible carrier modes
in the joints was defined based on the similarity (cross-correlation)
between the displacement mode shapes of the incoming mode and those of
the possible carrier modes. Given an incoming s0 mode which was known
to be generated most efficiently by the PZT actuators used for the
experimental component of this study, it was found that S0 and A1 act as
the primary carrier modes through the properly-cured and the poorly-cured
bonds. Since both of these modes produce increased normal and shear
strains at the bondline, they are suitable for monitoring changes in bond
stiffness. The S0,plate mode will instead act as the primary carrier mode
across disbonded regions, since its energy is confined within the top skin
plate. The model indicated that frequency values around 200 kHz would
produce mode coupling for S0 and A1. Mode coupling results into a large
interlayer energy transfer, and it was thus expected to provide maximum
sensitivity to the bond monitoring strategy adopted in this study. The
strength of transmission across the defected joints was quantified based on
the changes in the power flow (Poynting vector) along the wave
propagation direction relative to the properly-cured bond. For all the cases
considered the power flow results predicted that the strength of
transmission would be larger for the defected joints compared to the
properly-cured joint.
Experimental tests were conducted on two specimens constructed in
the laboratory and consisting of [0/±45/90]S and [0/±45/0]S T300/5208
plates bonded to T800/924 woven tubular spars. The joints were

manufactured by creating a region of poorly-cured adhesive and two
regions with isolated disbonds of two different sizes. The specimens were
162
tested in the “within the bond” and the “across the bond” configurations
using PZT disks. The strength of transmission was quantified by
computing the RMS of the detected signals in the 100kHz-300kHz range.
The experimental results confirmed that the strength of transmission
increases in the presence of the two types of bond defects compared to the
properly-cured bond. The best sensitivity to the defects was measured at
around 200kHz corresponding to the mode coupling point of the S0 and A1
carrier modes. In most of the cases, it was found that the poorly-cured
joint resulted in a larger transmission strength than the disbonded joints.
This result was opposite to what found by the SAFE power flow
predictions, that however assumed an idealized disbond extending for the
entire width of the joint. As expected, the measured strength of
transmission generally increased with increasing disbond size. Among the
cases examined, the “across the bond” configuration for the [0/±45/90]S
joint was the least favorable one, resulting in the smallest RMS changes
and the smallest ability to discriminate among defect types and sizes. The
most favorable testing configurations, providing maximum discrimination
between the poor bond and the disbond defects, were found to be the
“within the bond” test for the [0/±45/90]S lay-up, and the “across the
bond” test for the [0/±45/0]S lay-up.
Clearly, the results presented are strictly applicable to the joints
examined in this study. However, the general conclusions of an increased
strength of transmission in the presence of bond defects and a large
sensitivity at mode coupling points, can be extended to other geometries
or materials. Improvements can be made in the predictions of the power
flow through the defected joints, particularly for the disbonded case. A
163
complete power flow analysis should account for the finite size of the
disbond and for the attenuation of each of the carrier modes involved in
the energy transfer.
A factor that will play a substantial role in any field
implementation of the bond monitoring system is the operating
temperature. The nominal temperature range for UAVs is -40°F to 165°F
(-40 °C to 74 °C) for high altitude flights and storage in closed hangers,
respectively. Further tests should be conducted to characterize the effects
of temperature on the wave transmission strength.
4.8 Acknowledgements
Chapter 4, in part, has been published in the Journal of Acoustical
Society of America, Matt, Howard; Bartoli, Ivan; Lanza di Scalea,
Francesco (2005). The title of this paper is “Ultrasonic guided wave

monitoring of composite wing skin-to-spar bonded joints in aerospace
structures”. The dissertation author was co-author of this paper.

OF RAILROAD TRACKS (MODELLING)
5.1 Introduction
Various numerical and experimental studies have examined the
propagation of ultrasonic guided waves in railroad tracks. This topic iras
of interest in the area of noise generated by a passing train and, most
important, in the context of rail defect detection by long-range ultrasonic
inspection. Recent train accidents and associated direct and indirect repair
costs have reaffirmed the need for developing rail defect detection systems
more effective than those used today. In fact, rail defect detection has
been identified as a priority area in the U.S. Federal Railroad
Administration 5-year R&D plan. Lately researchers at SHM & NDE
laboratory at UCSD have been working on new inspection systems that
are targeted to the detection of transverse-type cracks in the rail head,
notoriously the most dangerous flaws in rails. Safety statistics data from
the US Federal Railroad Administration (FRA Safety Statistics Data)
indicate that train accidents caused by track failures including rail, joint
bars and anchoring resulted in 2,700 derailments and $441M in associated
damage costs during the decade 1992-2002. The first leading cause of these
accidents is the “transverse defect” type that was found responsible for 541
derailments and $91M in cost during the same time period ($17M in year
2001 alone). Transverse defects are cracks developing in a direction
perpendicular to the rail running direction.
164
165
The next two chapters will discuss some of the aspects of the
guided wave structural monitoring system for railroad tracks developed at
the NDE & SHM laboratory at UCSD. Chapter 6 will be focused on
experimental tests and signal processing to enhance the detection of
transverse defects in railroad tracks. The present chapter discusses
primarily the theoretical aspects of the guided wave propagation in
railroad tracks.
The classical theories of Euler-Bernoulli and Timoshenko for rail
vibrations are only accurate at frequencies below 0.5 kHz and 1.5 kHz,
respectively. At higher frequencies these theories cannot account for the
significant cross-sectional deformations of the rail (Thompson, 1993; 1997).
Numerical approaches have been proposed to predict the acoustic and
ultrasonic modal properties of rails as dispersion curves. In the work by
Thompson (1993) the finite element method was used to model transient
vibrations in rails with beam and plate elements at frequencies as high as
6 kHz. A double Timoshenko beam model that allowed relative
displacement between the head and the foot of the rail was used by Wu
and Thompson (1999) to model vertical waves, again below 6 kHz.
In the area of rail defect detection by long-range ultrasonic waves,
however, frequencies in the range 10 kHz – 50 kHz have shown the best
promise (Wilcox et al., 2003; Cawley et al., 2003; Lanza di Scalea and
McNamara, 2003; McNamara, 2003). In this case, complications arise from
the multimode and dispersive character of high frequency guided waves in
rails that sometimes makes the mode identification challenging. For
example, at 50 kHz there are around 20 vibrational modes theoretically
propagating in typical rails. In the papers by Gavric¶ (1994, 1995) the

166
complexity of finite element modeling of waves in acoustic waveguides
including rails was relaxed by proposing a discretization in the rail cross-
sectional plane only. The three-dimensional problem was thus successfully
treated as a bi-dimensional one with significant savings in computational
efforts. The same bi-dimensional method was employed by Wilcox et al.
(2003) where a cyclic symmetry condition was imposed in the wave
propagation direction. With the cyclic symmetry approximation, Gavric’s
bi-dimensional method could be implemented in standard finite element
programs. These techniques are very efficient to calculate the wave modal
solutions in terms of dispersion curves and cross-sectional mode shapes.
However, predicting the interaction of the waves with structural
defects generally requires a three-dimensional model. Sophisticated
numerical methods have been used to predict reflections of ultrasonic
waves from rail defects, including three-dimensional time marching models
(Wilcox et al., 2003; Cawley et al., 2003) and adaptive mesh refinement
models (Trivedi et al., 2004). The reflection coefficients of guided waves
from defects are the basis of defect detection in rails by long-range
inspection. Defects that develop transversely to the rail running direction

are notoriously the most dangerous in rails. According to the Federal
Railroad Administration safety data for the decade 1992-2002, transverse
defects were the first leading cause of track failures in the US with $91M
in associated damage and repair costs. Reflection coefficients for
longitudinal, lateral and vertical waves interacting with transverse head
defects of four different sizes and three different orientations were recently
determined experimentally by a broadband, impulse excitation of a 115lb
A.R.E.M.A. rail (Lanza di Scalea and McNamara, 2003; MaNamara, 2003).

167
In these studies a joint time-frequency analysis based on the Wavelet
transform was employed to analyze the defect reflections in the 10 kHz –
40 kHz frequency range.
The present chapter demonstrates the use of a commercial finite
element code (ABAQUS EXPLICIT) to model the reflection of vertical
bending waves from transverse defects in a rail that is subject to
impulsive excitation. The excitation used in the rail model simulates the
signal of the hammer used in the experiment. Defects examined are
transverse-type flaws in the rail head of four different sizes and three
different orientations. Reflection coefficients were predicted in the 20 kHz
– 45 kHz frequency range through Wavelet transform analysis and good
agreement is shown with a limited set of experimental data. The accuracy
of the simulation is first assessed in a pilot study conducted on the
classical case of Lamb waves in plates. Results from the simulation could
be used to generate a map of reflection coefficients for a variety of defects

in rails for the purpose of identifying optimum inspection frequencies or
training an automatic defect classification algorithm based on supervised
learning. The simulations prove also useful towards the ultimate goal of
developing a practical defect detection system for transverse rail defects
that can be operated at high speeds (~70 mph).
5.2 Finite element modeling with application to Lamb waves in a plate
The effectiveness of conventional finite element modeling of elastic
waves propagating in structural components has been shown in the past.
The case of Lamb waves in free plates is a classical example (Alleyne and
168
Cawley, 1991; Moser et al., 1999). The package used in the present
study, ABAQUS EXPLICIT, uses an explicit integration based on a
central difference method (Abaqus User’s manual).
The stability of the numerical solution is dependent upon the
temporal and the spatial resolution of the analysis. To avoid numerical
instability ABAQUS EXPLICIT recommends a stability limit for the
integration time step ∆t equal to
∆t = Lmin / cL (5.1)
where Lmin is the smallest dimension of the smallest finite element of the
model and cL is the bulk longitudinal wave velocity through the material.
This limit represents the time of travel of a longitudinal wave across the
element.
The maximum frequency of the dynamic problem, fmax, limits both
the integration time step and the element size. Depending from the
accuracy required a good rule is to use a minimum of 20 points per cycle
at the highest frequency (Moser et al., 1999), that is
1
∆t = (5.2)
20fmax
The size of the finite element, Le, is typically derived from the
smallest wavelength to be analyzed, λmin. For a good spatial resolution 10
nodes per wavelength are normally required (Alleyne and Cawley, 1991),
although some studies (Moser et al., 1999) recommend a more stringent

169
condition of 20 nodes per wavelength. The latter condition can be written
as
Le = λmin / 20 (5.3)
A pilot study of Lamb waves propagating in a free, isotropic plate
was conducted with the purpose of verifying the proposed limits of
temporal and spatial resolution in ABAQUS EXPLICIT as well as to show
the potential of the Wavelet transform to analyze dispersive waves in a
broad frequency domain. This work was also aimed at identifying criteria
to be extended to the case of wave propagation in railroad tracks.
The well known Rayleigh-Lamb solutions of Lamb waves are (Rose,
1999)
tan(qh) 4k2pq
= − 2 for the symmetric modes, and
tan(ph) (q − k2 )2
(5.4)
tan(qh) (q2 − k2 )2
= − for the antisymmetric modes
tan(ph) 4k2pq
where k is the wavenumber and 2h is the plate thickness. The variables p
and q are found from the circular frequency, ω, the wavenumber, k, and
the bulk longitudinal and shear wave velocities, cL and cT, following
⎛ω⎞ ⎛ω ⎞⎟
2 2
p = ⎜⎜ ⎟⎟⎟ − k2
2
and q = ⎜⎜
2
⎟⎟ − k2 (5.5)
⎜⎝ c ⎠⎟
L ⎝⎜ c
T ⎠
⎟
The ABAQUS model of the plate is shown in Figure 5.1. The material is
steel, with the following nominal properties: Young’s modulus E=209×109
N/m2; Poisson’s ratio ν=0.3; density ρ=7800 Kg/m3; bulk longitudinal
wave velocity cL=5.98 km/sec; bulk shear wave velocity cT=3.20 km/sec.
Damping effects were neglected. A thickness of 20 mm was chosen for the

170
plate based on the medium width of the web of the railroad track
subsequently examined. In Figure 5.1 the plate is infinite along z so the
problem is a plane strain one. A finite length of 3 m was considered for
the model in the x direction. Such length allowed avoiding reflections from
the right end. The plate was discretized by 4-node bilinear plane strain
quadrilateral elements with two degrees of freedom per node. Both square
and rectangular elements of various sizes were considered as schematized in
the table of Figure 5.1, where a and b are the element sizes along the
longitudinal direction x and the thickness direction y, respectively. The
integration time step was set to ∆t = 0.2 µsec throughout the analyses.
In the frequency range of interest (tens of kiloHertz), only the zero-
order symmetric mode, S0, and anti-symmetric mode, A0, are present.
These two modes were selectively excited in the model by applying
appropriate nodal displacements. A triangular forcing function d(t) was
applied as a uniform x-displacement distribution on the left edge of the
plate to excite S0 (Figure 5.2(a)). A similar displacement distribution, only
in the vertical direction y, was applied to the same nodes to excite A0
(Figure 5.2(b)). Typical results obtained by the model are shown in Figure
5.2(a) and Figure 5.2(b) for S0 and A0. The variable monitored was the
vertical, y-displacement monitored at mid-span at the top surface of the
plate (node 1). In Figure 5.2(a) the reflection from the right-end of the
plate is also visible in addition to the first arrival owing to the larger
group velocity of S0 when compared to A0 at these frequencies.
The continuous Wavelet transform (Appendix A) was used to
perform the joint time-frequency analysis of the time signals. As a good
compromise between time and frequency resolution, the Gabor wavelet

171
transform (GWT) was employed with a center frequency of 2π and a
shape factor of 5.336. GWT scalograms, representing the energy of the
signal in the joint time-frequency domain, are shown in Figure 5.2 for the
corresponding time histories. The GWT analysis allows to extract the
group velocity dispersion curves as well as the reflection coefficients from
geometrical discontinuities in a broad frequency range at once (Lanza di
Scalea and McNamara, 2003; MaNamara, 2003).
The wavelet analysis simply requires a single generation point and a
single detection point. Group velocity dispersion curves for the plate model
were calculated by comparing the scalograms of node 1 with those of node
2, located on the same top edge of the plate at a distance L = 750 mm
from node 1. The frequency-dependent velocity was then obtained through
the arrival times of the scalogram peaks for the two nodes following
L
Cg (f ) = (5.6)
t2 (f ) − t1(f )
Figure 5.1: ABAQUS model of the plate with meshes examined.

172
F.E.M. Time Signal
Displacement Node 1
(a)
Mode So
0 0.5 1
GWT of F.E.M. Signal
60
Frequency (KHz)
y 50
Incident Mode
40
x 30
Reflected Mode
20
10
0 0.5 1
Time (msec)
F.E.M. Time Signal

Displacement Node 1
(b)
Mode Ao
0 0.5 1
GWT of F.E.M. Signal
60
Frequency (KHz)
50
40
x 30
20
10
0 0.5 1
Time (msec)
Figure 5.2: Abaqus simulations of Lamb waves in a plate excited by an

impulse. (a) Case of the S0 mode; (b) case of the A0 mode.
173
The dispersion results from the finite element model are shown in
Figure 5.3 and compared to the exact Rayleigh-Lamb solutions from Eq.
(5.4). In this figure the convergence of the model is shown for the
different discretization meshes employed. The analysis was aimed at
examining the effect of the element size both in the longitudinal direction
x and in the cross-sectional direction z. There is a lack of
recommendations in the literature on the mesh refinement requirements
along the cross-section of the waveguides. As the present study shows, an
appropriate mesh refinement in the cross-section is as important as the
mesh refinement in the longitudinal, wave traveling direction. The
appropriate cross-sectional refinement is dependent on the mode shapes of
the guided wave in a similar way as the longitudinal refinement is
dependent on the wavelength from Eq. (5.3).
Figure 5.3: Group velocity dispersion curves extracted from the model for
varying mesh refinements. Plot also shows the Rayleigh-Lamb exact
solution.
174
Figure 5.3 shows that by doubling the refinement in the thickness
direction (mesh 2 compared to mesh 1), there is an appreciable
improvement in the group velocity prediction. By increasing the mesh
refinement also in the longitudinal direction (mesh 3), a more substantial
improvement is obtained with maximum errors below 3% of the exact
solution. The increasingly refined meshes 4 through 6 coincided with the
theoretical result in the frequency range examined.
Results were also obtained in terms of energy reflection coefficients
from the plate edge. Both a perpendicular edge and a 35 deg inclined
edge were considered in analogy with the reflections from perfectly
transverse and inclined defects in the rail. The finite elements were
distorted near the inclined edge. The convergence criterion indicated
below accounts for this distortion. It is well known that for a
perpendicular edge reflection no mode conversion is expected, meaning that
a given incident mode (whether S0 or A0) would produce the same
reflected mode below the cut-off frequencies of higher modes. An inclined
reflector will instead produce mode-converted reflections necessary to satisfy
stress-free boundary conditions. Energy reflection coefficients were obtained
from the GWT scalogram ridges of the vertical displacements of node 1 in
Figure 5.2 for the first arrival and the edge reflection. To compensate for
dispersion effects, the areas under the scalogram ridges were considered
rather than the scalogram peaks. Reflection coefficient spectra were then
calculated using
175
∫ GWT2 (t, f )dt

∆t2
R(f ) = (5.7)
∫ GWT1(t, f )dt
∆t`1
where GWT(t,f) is the scalogram map, ∆t is the entire duration of the
pulse at frequency f, and the subscripts 1 and 2 refer to the first arrival
and the edge reflection, respectively.
The results are presented in Figure 5.4 for the different mesh
refinements and monitoring the x longitudinal displacement in all cases.
The case of mesh 6 (a=b=1mm) is here considered an exact solution. For
the perpendicular edge, Figure 5.4(a) and (b), the reflection coefficient for
both S0 and A0 is one throughout the frequency range and no substantial

effect of the mesh refinement can be seen. For the 35 degree inclined
edge, mode conversions take place. The same-mode, S0 reflection (Figure
5.4 (c)) decreases with increasing frequency because the remaining energy
is converted to A0 (Figure 5.4 (e)). A similar trend is observed for the
A0 mode incident on the inclined edge (Figure 5.4 (d) and (f)). Contrarily
to the perpendicular reflection, the mesh refinement has a substantial
effect on the inclined reflection in all cases. The reason for this is that it
is more challenging to accurately represent mode converted modes
including the nonpropagating (evanescent) modes that exist near the
reflector (Torvik, 1967). The importance of the mesh refinement in the
cross-sectional direction, in addition to the longitudinal direction, is
reaffirmed by comparing mesh 3 (a=b=5mm) to mesh 4 (a=5mm;
b=2mm) for the oblique reflections in Figure 5.4(c) through Figure 5.4(f).
Convergence up to 60 kHz is obtained by further refining the mesh in the

176
longitudinal direction with mesh 5 (a=b=2mm). This suggests that an
appropriate cross-sectional spatial resolution requires 10 nodes across the
plate thickness, thus
Le,cs = d/10 (5.8)
where Le,cs is the cross-sectional size of the finite element and d is the
plate thickness. The plate results confirm qualitatively the validity of the
criteria expressed in Eqs. (5.1)-(5.3). For a maximum frequency of 60 kHz,
for example, the minimum wavelength is for A0 and it is given by λmin=
c/fmax =40 mm, considering a theoretical phase velocity of c =2.45
km/sec. From Eq. (5.3), the corresponding limit on the element size is Le
=2 mm that corresponds to the size of the mesh 5 element.
It should be noted that this criteria is analogous to the one
discussed in Section 3.5.2 for the case of axis-symmetric waveguides
modeled with the SAFE method. Since quadratic elements were used in
Chapter 3, the mentioned criteria used in the SAFE required only 4
elements to discretize a length equivalent to the minimum wavelength. The
more stringent criteria (10-20 elements per wavelength) used here are
imposed by the poorer interpolation (linear interpolation) of the
displacement in the finite elements used for the current ABAQUS
simulations.
As for the temporal resolution, with an element size of 2 mm Eq.
(5.1) recommends a maximum integration time step of ∆t = 0.33 µsec
that is satisfied with the value of 0.2 µsec used for the analysis.
It should also be noted that if lower frequencies are of interest,
these convergence criteria along the longitudinal direction can be relaxed.

177
For example, taking 45 kHz as the highest frequency of interest (as in the
rail application that follows), Figure 5.4 shows that mesh 4 gives adequate
results. In this case the largest error, in Figure 5.4(f), is within 8 % of
the exact solution. Therefore, considering that λmin=54.4 mm for fmax=45
kHz, ten nodes per wavelength in the longitudinal direction provide an
acceptable accuracy. Mesh 4 still satisfies the cross-sectional criterion of
Eq. (5.8).
(a) (b)
(c) (d)
(e) (f)
Figure 5.4: Energy reflection coefficients in a plate from a perpendicular

edge (a and b) and from a 35 deg inclined edge (c through f).
178
5.3 Application to rails
The use of guided waves in rails, as discussed above, is of interest
in the context of long-range defect detection. The potential is to detect
defects as far away as tens or even hundreds of feet from the sensoring
system.
In this study the attention was focused on the vertical bending
vibrational mode that provides good defect sensitivity and it is easily
generated and detected from the top of the rail. Access to the top of the
rail only is required for any practical defect detection system that operates
at high speeds in rails. The vertical bending mode can be generated by an
impulse excitation at the top of the rail head in the vertical direction.
Figure 5.5: Geometry of the rail studied with the different transverse head
defects.
179
Figure 5.6: Experimental setup for measuring reflections of the vertical

bending mode from transverse head defects.
Both numerical and experimental analyses were conducted to study
the reflection of the vertical bending mode from transverse-type defects in
the rail head. These are among the most dangerous flaws in rails as
mentioned above. Figure 5.5 schematizes the 115 lb A.R.E.M.A. rail
section that was studied. The elastodynamic properties of the rail steel
material were the same as the ones assumed for the plate. Four defect
sizes were studied numerically, herein indicated by the percentage of the
head section that was cut, i.e. ‘15% defect‘, ‘50% defect‘, ‘85% defect‘, and
‘100% defect‘. For all four sizes, three different orientations were
considered, namely a perfectly transverse direction, a direction inclined at
20 deg from the transverse direction, and one inclined at 35 deg from the
transverse direction. In total, twelve cases of head defects were examined
by the finite element analysis.

180
Results from the experimental testing are here shown only for the
100% defect (cut through the entire head) at the three orientations of
transverse, 20% oblique and 35% oblique. Also, only same-mode reflections
were measured. The experimental setup is shown in Figure 5.6. A high-
frequency impulse hammer (PCB 086D80 mini hammer), effectively excited
frequencies as high as 50 kHz with appreciable energy. The hammer struck
the top of the rail head at its left end along the vertical direction y. This
excitation produced effectively a vertical bending mode that is largely
predominant over other modes theoretically possible in the frequency range
considered (Lanza di Scalea and McNamara, 2003; McNamara, 2003; Lanza
di Scalea and McNamara, 2003). A typical time history of the hammer
impulse is shown in Figure 5.6. The same time history was used as a
force input in the finite element analysis. An accelerometer (PCB 352C67)
recorded the vertical bending mode at the top of the rail head at a
distance L2= 812.8 mm from the defect. The perfectly transverse 100%
defect was located at L1= 1828 mm from the impact location whereas the
100% inclined defects were located at L1= 1574.8 mm from the impact
location.
In the finite element model, the rail was discretized in ABAQUS
EXPLICIT by 8-node brick elements with linear deformation, lumped mass
matrix, and 3 degrees of freedom per node. Figure 5.7 shows the mesh
employed with a typical head defect. The size of the elements used was 4
mm in the longitudinal (running) direction z, and between 2 mm and 4
mm in the cross-sectional directions, x and y. The finite elements were
distorted near the inclined defects. The longitudinal direction refinement
followed the ten node-per-wavelength rule found acceptable for the plate
181
below 45 kHz (in this case λmin = 44.4 mm considering c=2 km/sec). The
cross-sectional refinement of the rail head also followed the ten-node-per-
thickness rule expressed in Eq. (5.8). To save computational efforts, the
cross-sectional refinement criterion applied to the rail head was not strictly
observed for the rail web and base as seen in Figure 5.7. This
approximation was considered acceptable since the defects were all located
in the rail head. The rail section modeled was 2.4 m long. This length
successfully isolated the defect reflection from the far end reflection. The
total number of nodes was on the order of 300,000. The integration time
step was set at 0.2 µsec. As for the plate case, this value satisfied the
temporal criterion in Eq. (5.1). All analyses were carried out on a 2.6
GHz Pentium IV with 1 GByte of RAM. A typical analysis for each of
the defect cases took 150-200 minutes.
The accelerations of the node at the top of the rail located at L2

from the defects were monitored in the three Cartesian directions. Figure
5.8 shows two time signals for the vertical acceleration with the
corresponding GWT scalograms. The signal in Figure 5.8(a) was obtained
by the model whereas the one in Figure 5.8(b) is experimental. The two
signals were obtained with the 100% defect in the perfectly transverse
orientation. As indicated by the scalograms, the first arrival of the vertical
bending wave and its reflection from the defect can be clearly identified.
The split seen at high frequencies in the predictions of Figure 5.8(c) is
either another mode or the result of spurious numerical oscillations. The
split contained energy that was more than one order of magnitude smaller
than that of the dominant mode, and it was thus neglected.

182
Figure 5.7: Finite element mesh employed for the rail.
The group velocity dispersion curves were extracted from Eq. (5.5)
where now the subscripts 1 and 2 refer to the first arrival and the defect
echo, respectively. Also, the distance L in Eq. (5.6) is 2L2 in Figure 5.6
and Figure 5.7. The group velocity results are shown in Figure 5.8(e)
confirming the expected character of the vertical bending mode with a
slight dispersion at the low frequencies. Also, the agreement between the
numerical and the experimental results is satisfactory with a maximum
discrepancy of 4%.
183
F.E.M. Time Signal Experimental Time Signal
Vertical Acceleration
Vertical Acceleration
(a) (b)
0 0.5 1 0 0.5 1
GWT of F.E.M. Signal GWT of Experimental Signal

50 50
Frequency (KHz)
Frequency (KHz)
40 40
(c) 30 30 (d)
20 20
0 0.5 1 0 0.5 1
Time (msec) Time (msec)
(e)
Figure 5.8: Time histories and GWT scalograms of the vertical acceleration
at the top of the rail head from the simulation (a and c) and from the
experiment (b and d). Resulting group velocity shown in (e).
Energy reflection coefficients for all 12 defect cases were obtained by
using Eq. (5.7) and the results are presented in Figure 5.9 and Figure
5.10. Figure 5.9 shows the same-mode reflection results, meaning the
reflected vertical bending mode when the same mode is incident. The
experimental result for the 100% defect is presented as a solid curve in
these figures. It should be noted that the experimental data presented here
were corrected for material attenuation effects since no attenuation was
included in the model. The following correction equation can be easily
derived:
184
α (f ) (2L2 )
R(f )corrected = R(f ) × 10 10

(5.9)
where R(f) is the raw, experimentally-derived reflection coefficient
spectrum from Eq. (5.7), α(f) is the linear attenuation spectrum of the
vertical mode in dB/m, and L2 is the accelerometer-defect distance in
meters.
Figure 5.9: Same-mode reflection coefficients for the vertical bending mode
from the perfectly-transverse defects (a), the 20 deg inclined defects (b),
and the 35 deg inclined defects (c).
185
By using this correction equation the results can be converted to
general reflection coefficients that are not bound by the experimental setup
used. In other words, the results correspond to what would be measured if
the detection accelerometer was positioned at the same location as the
defect, resulting in stronger reflections than the raw data would indicate.
The values of α(f) used in Eq. (5.9) were those previously
measured for the vertical bending mode (Lanza di Scalea and McNamara,
2003). In that study the attenuation was found highly varying with
frequency, and generally smaller than 1 dB/m below 50 kHz.
Figure 5.9 shows a slight discrepancy between the numerical and
the experimental data. This is probably the result of slight differences
between the real defects obtained by saw cutting and the flaws simulated
in the model. Appreciable reflections are seen in the entire frequency
range examined (20 kHz – 45 kHz), confirming the suitability of these
ultrasonic frequencies for long-range defect detection. The general trend
that emerges from Figure 5.9 is an expected increase in reflection strength
with increasing defect size for all three orientations of transverse (Figure
5.9(a)), 20 deg oblique (Figure 5.9(b)) and 35 deg oblique (Figure 5.9(c)).
For the 20 deg oblique defects, however, the reflection strength is less
dependent on defect size above 30 kHz. A more unexpected result is a
general decrease in reflection strength for the 20 deg oblique defects when
compared to the other two orientations of transverse and 35 deg oblique.
The effect can be explained by severe mode conversions for the 20 deg
cuts that take energy away from the same-mode, primary reflection.
186
The mode-converted reflection coefficients are presented in Figure
5.10 where the reflected lateral bending mode is now being monitored with
the usual incident vertical mode. Mode conversion between a symmetrical
mode (vertical bending) and anti-symmetrical mode (lateral bending) is
expected when either the size or the orientation of the defect are such
that the flaw is not symmetrical with respect to the longitudinal plane of
symmetry of the rail. This occurs for all defect cases examined with the
exception of the 100% defect in the perfectly-transverse orientation. The
difficulty of monitoring mode conversion is that the lateral bending mode
has the same group velocity as the vertical bending mode above 25 kHz,
and thus the two are indistinguishable in the time domain (Lanza di
Scalea and McNamara, 2003). In this study the reflected lateral mode was
identified by monitoring the transverse acceleration at the detection node
(direction x in Figure 5.7), rather than the vertical acceleration, along
direction y. Since this node is on the plane of symmetry of the rail, the
acceleration along x is zero for the vertical mode and it is only

”activated” by the reflected lateral mode. Mode-converted reflections were
computed from the usual Eq. (5.7), where now GWT2 refers to the
scalogram of the transverse acceleration for the reflection (lateral mode),
and GWT1 refers to the scalogram of the vertical acceleration for the first
arrival (vertical mode).
The results in Figure 5.10 show that mode-converted reflections are
generally weaker than the primary, same-mode reflections in Figure 5.9.
Also, mode conversions are stronger at the low frequencies, rapidly
decaying above 30 kHz. Figure 5.10(a) for the transverse orientation
confirms that mode conversion is absent for the largest 100% defect. In
187
the same figure, the reflection is stronger for the 50% defect when
compared to the 85% defect. This apparently unexpected behavior can be
explained by noting that the 50% defect is more un-symmetric than the
85% defect, and it thus generates a larger mode conversion. The same
trend is observed for the 20 deg oblique reflections in Figure 5.10(b). In
this case the strongest reflections occur for the largest 100% defect. The
35 deg oblique orientation, Figure 5.10(c), indicates a more intuitive trend
of increasing reflections with increasing defect size.
Figure 5.10: Mode-converted reflection coefficients for the vertical bending

mode from the perfectly-transverse defects (a), the 20 deg inclined defects
(b), and the 35 deg inclined defects (c).
188
5.4 Conclusions
This study has demonstrated the use of a commercial finite element
package, ABAQUS EXPLICIT, to model guided ultrasonic waves
propagating at frequencies of tens of kiloHertz. The waves, generated by
impulsive excitation, are examined in a broad frequency range with the aid
of joint time-frequency (wavelet) analysis.
The case of Lamb waves propagating in a 20mm-thick free plate is
first examined with the purpose of checking the validity of existing
recommendations for the spatial and temporal resolution of the finite
element analysis as applied to ABAQUS. Modal properties of the Lamb
waves in terms of group velocity as well as reflection coefficients from
perpendicular and inclined edges are studied. The modal solutions and the
reflection from the perpendicular edge are found less sensitive to the mesh
refinement than the reflections from the oblique edge. The reason is the
increased accuracy required to account for the presence of mode
conversions and evanescent modes. Adequate accuracy in modeling edge
reflections in the plate was achieved with 10 nodes per wavelength below
45 kHz and 20 nodes per wavelength below 60 kHz. As for the cross-
sectional refinement, it was found that 10 nodes across the thickness are
required for both the 45 kHz and the 60 kHz limits.
These guidelines were then applied to modeling guided waves
propagating in 115lb A.R.E.M.A. rails in the context of rail defect
detection by long-range ultrasonic inspection. The vertical bending mode
was examined for its ease of generation and detection in the field. This
mode was generated by an impulse excitation at the top of the rail head
in the vertical direction. The defects examined included four different sizes
189
of transverse head flaws at three different orientations, for a total of
twelve cases. The model extracted same-mode reflections (vertical incident
and reflected) and mode-converted reflections (vertical incident and lateral
reflected). A limited set of experimental data gathered in the laboratory
was also obtained for comparison with the finite element predictions. The
study shows that appreciable reflections from defects as small as 15% of
the rail head are found in the 20 kHz-45 kHz range. As expected, the
same-mode reflection strength generally increases with increasing defect
size, although this dependence is not marked for the 20 deg oblique flaws
above 30 kHz. Also, the same-mode reflections from the 20 deg flaw
orientations are weaker than those from the 35 deg orientations. Below 30
kHz, the mode-converted reflections are found to be substantial, their
strength rapidly decreasing at the higher frequencies. The amount of mode
conversion depends on the symmetry of the flaw with respect to the
longitudinal plane of symmetry of the rail. This consideration helps
explaining why in some cases stronger mode conversions are predicted for
smaller defects.
This study is part of a larger project aimed at developing a long-
range inspection system for rails that is particularly sensitive to transverse-
type flaws and that can be operated at high speeds. Details of the
development of an experimental prototype for high speed rail inspection
are given in the next chapter. Finite element modeling becomes an
essential tool to predict the wave interactions with a variety of defects
that would be impractical to replicate experimentally. The reflection
coefficient spectra from this variety of defect sizes and orientations can
then be fed to an Automatic Defect Classification (ADC) algorithm that

190
operates in a supervised learning mode. The first implementation using
Support Vector Machines for ADC of transverse defects in rails has been
recently demonstrated in the laboratory (McNamara at al., 2004).

OF RAILROAD TRACKS (EXPERIMENT)
6.1 Introduction
It was discussed earlier that the detection of cracks in rail tracks,

particularly transverse cracks in the rail head, is a topic that necessitates
further improvements over current rail inspection technologies. The most

popular rail inspection method is ultrasonic testing, conventionally
performed from the top of the rail head in a pulse-echo configuration. In
this system ultrasonic transducers are located inside a water-filled wheel
and are oriented at 0° from the surface of the rail head to detect
horizontal cracks, and at 70° to detect transverse cracks. Such approach
suffers from a limited inspection speed (typically 20 mph in the U.S.) and
from other drawbacks associated to the requirement for contact between
the rail and the inspection wheel. More importantly, horizontal surface
cracks such as shelling can prevent the ultrasonic beams from reaching the
internal defects resulting in false negative readings. The problem of surface
shelling was highlighted in the June 1992 train derailment in Superior,
Wisconsin, caused by a transverse crack that was missed by the
inspection, where an entire town had to be evacuated due to concerns
from spilling of hazardous materials.
Guided waves propagate along, rather than across the rail, and are
thus ideal for detecting the critical transverse defects. These waves are
also not sensitive to surface shelling since they can run underneath these
191
192
discontinuities. Finally, guided waves propagate at the speed of sound in
steel (on the order of thousands of miles per hour), resulting in the
potential for extremely large inspection speeds. This would make it
possible to reach inspection speeds as large as regular passenger speeds (up
to 100 mph) so as to avoid any traffic disruption during testing. Another
important aspect that is being addressed by researchers at UCSD, Johns
Hopkins University and other institutions is the development of practical

ways to generate and detect ultrasonic waves in rails. The use of
techniques that do not require contact with the rail is being investigated
as a promising solution. Non-contact rail testing has been demonstrated by
the use of lasers and air-coupled transducers [Lanza di Scalea and
McNamara, 2003; McNamara, 2003; Lanza di Scalea, 2000; Kenderian et
al., 2002; Kenderian et al., 2003; McNamara and Lanza di Scalea, 2004) as
well as electro-mechanical-acoustic transducers (Alers, 1988). However, the
drawback of any non-contact ultrasonic testing is a reduced signal-to-noise
ratio of the defect detection procedure when compared to the conventional
contact testing. The use of signal processing based on the wavelet
transform was recently demonstrated to enhance the performance of non-

contact rail testing (McNamara and Lanza di Scalea, 2004).
In this section the application of ultrasonic guided waves to the
detection of transverse-type defects in rails is described. The
complementary use of low-frequency waves (below 50 kHz) and high-
frequency waves (above 100 kHz) is investigated by using, respectively, an
impulse hammer and a pulsed laser for excitation. Air-coupled transducers
are used for detection of the waves at distances much larger than those
shown previously. Particularly, the sensors stay beyond the generally

193
recommended clearance envelope of 65 mm (2.5”) from the top of the rail.
Defects are detected by monitoring both reflected and transmitted waves.
This dual detection provides robustness and redundancy to the inspection.
The ultrasonic signals are processed by the wavelet transform to increase
the defect detection reliability, the inspection range and the inspection
speed.
(a)
(b) (c)
Figure 6.1: System for detecting transverse cracks in rails by generating

ultrasonic guided waves with a hammer and detecting the waves with a
pair of air-coupled sensors.
194
6.2 Defect detection using hammer generation/air-coupled detection
Rail sections of the 115lb A.R.E.M.A. type were tested in the
laboratory. The experimental setup used for the first system is shown in
Figure 6.1(a). An impulse hammer (PCB 086D80 mini hammer) was
employed to excite guided waves at frequencies below 50 kHz by striking
vertically the top of the rail head. A pair of air-coupled transducers with
broadband response in the DC-2MHz range was used to detect the waves.
Using two sensors allowed to extract both reflection coefficients and
transmission coefficients for the target defects. Pictures of the sensors
located prior to and across a transverse defect are shown in Figure 6.1(b)
and Figure 6.1(c), respectively. In these pictures the large lift-off distance
of the sensors can be appreciated. In this portion of the study results are
presented for a transverse crack that runs for the entire section of the rail
head.
6.2.1 Effect of Sensor Inclination
Determining the role of the sensor orientation is critical to maximize
the sensitivity to the defects as well as to characterize the system’s
tolerance to accidental misalignments in the field. The optimum air-coupled
detection angle from the normal to the rail surface, α, is given by Snell’s
law of refraction (Figure 6.3):
⎛c ⎞⎟
α = arcsin ⎜⎜⎜ air sin θp ⎟⎟ (6.1)
⎜⎝ cp ⎠⎟
195
where cp is the phase velocity of the guided wave in the rail, cair =330
m/sec is the wave velocity in air and θp = 90º for a wave propagating
parallel to the rail surface. Considering that a vertical-type guided wave
propagates at cp = 3,000 m/sec in the frequency range of interest, Eq.
(6.1) gives α = 6.3º. This is the optimum sensor angle to detect crack
reflections. The convention here is positive angle towards the defect (i.e.
away from the generating hammer). However, it is also important to
detect the wave incoming from the hammer that travels in an opposite
direction from the crack reflection. In this case the optimum detection
angle would be –6.3º, implying sensor oriented towards the hammer. A
compromise is to orient the sensor parallel to the rail surface (α = 0º).
Figure 6.2(a) shows the signal detected by such an orientation
where the sensor lift-off distance, h, is fixed at 76 mm (3”) and the
distance sensor-defect is 1.120 mm (44”). The first arrival and the
reflection from the defect are clearly visible. Figure 6.2(b) is the
continuous wavelet transform (CWT) scalogram of the time signal
revealing its time-frequency spectrum. Figure 6.2(c) illustrates the
extraction of the reflection coefficient at a frequency of 30 kHz from the
wavelet scalogram. The reflection coefficient at frequency f, R(f), is here
defined as the ratio between the reflected energy, A2, and the incoming
energy, A1, following:
A2 (f )
R(f ) = (6.2)
A1 (f )
196
Time Signal
50
Nei casi di seguito esaminati si e’
First Arrival Defect echo
Vertical displacement
(a)
-50
0 0.5 1
Time (msec)
GWT of Signal
45
40
Frequency (KHz)
35
30
25 (b)
0 0.5 1
Time (msec)
f = 30 kHz
First Arrival
GWT norm. amplitude
Reflection
A1
A2
(c)
0 0.5 1 1.5
Time (msec)
Figure 6.2: Detection of transverse crack by reflection measurements

(detection angle 0º). (a) time history; (b) continuos wavelet transform; (c)
wavelet peaks at 30 kHz.
197
T
2
2
Air
Steel
1
T
1
Figure 6.3: Snell’s law of refraction applied to air-coupled detection of

guided waves.
Figure 6.4: Formation of resonances in the lift-off air gap.
A value of R(f)=1 thus means that all of the incoming energy is reflected
from the defect, whereas R(f)=0 means that no energy is reflected and it
is all transmitted. Reflection coefficients were measured by varying the
sensor orientation in the range – 50º to 50 º. The results are shown in
Figure 6.5. Here R was calculated at a frequency of 30 kHz that provided
strong reflections. It is remarkable that it is possible to detect defect
reflections in a large range of sensor orientations, especially considering the
large lift-off distance of 76 mm (3”) used in the tests. As expected, the

198
reflection strength increases when the sensor is oriented towards the defect.
This suggests that in those cases where signal-to-noise ratio may be a
problem, such as when reflections are sought from small defects, the
sensors should be oriented away from the generation hammer. This point
will be demonstrated further when discussing the laser/air-coupled system
in the following paragraphs.
1.4
1.2
Reflection Coefficient
0.8
0.6
0.4
0.2
0
-50 -40 -30 -20 -10 0 10 20 30 40 50
Angle D
Figure 6.5: Variation of the reflection coefficient for transverse crack (100%
Head area reduction) as a function of detection angle.
199
An optimum angle to detect the defect reflection is identified at 35º in
Figure 6.5. The optimum detection angle depends on the phase velocity
of the guided waves which, in turn, is a function of the wave frequency.
Consequently, the value of 35º may not be appropriate for frequencies
beyond the 45 kHz limit examined here. Values of R larger than 1, are a
result of the more favorable detection of defect-reflected signals rather than
incoming signals at large sensor inclinations. Also, the variations of R are
very small for sensor angles between – 10º and 10º. This implies that some
transducer misalignment would be well tolerated by the system in the
field.
6.2.2 Effect of Sensor Lift-off Distance
Positioning the sensor as far away as possible from the top of the
rail head was the target of this study. A clearance envelope of 65 mm
(2.5”) is generally recommended for new rail inspection systems. Larger
sensor distances, however, tend to degrade the defect detection sensitivity
due to the losses in air, particularly severe for high-frequency waves.

Reflection coefficients from the large transverse defect positioned
1.120 mm (44”) away from the probes were obtained while varying the
sensor lift-off distance, h. The results of these tests, only summarized here
for the sake of space, indicated some unexpected trends for h < 25.4 mm
(1”). It was found that for small h, standing waves form between the rail
and the sensor face producing interference patterns that can give false
indications on the severity of the defect. Optimum lift-off distances for a

200
stable reflection from the large transverse defect were identified between 76
mm (3”) and 90 mm (3.5”) for the DC - 50 kHz frequency range under
hammer excitation. Previous studies (Lanza di Scalea and McNamara,
2003; McNamara et al., 2004; McNamara, 2003; Bartoli, 2004) indicate
that such frequencies are appropriate for detecting transverse head defects
as small as 15% of the head cross-sectional area. The potential thus
exists for a truly “non-contact” inspection system that stays outside the
recommended clearance envelope.
6.2.3 Transmission Approach
By using a pair of sensors as shown in Figure 6.1, it is possible to
detect a defect by monitoring the transmission coefficient between the two
sensors. The transmission coefficient, T(f), can be extracted from the
CWT scalograms of the signals detected by sensor #1 and sensor #2
following:
Asensor # 2 (f )
T (f ) = (6.3)
Asensor # 1 (f )
where A indicates the energy of the detected signals at frequency f and
sensor #1 is the closest to the impact hammer. When both sensors are
located prior to the defect, they should detect the same signal (neglecting
attenuation losses in the rail material) and thus T=1. When the sensors
are on either side of the defect, part of the energy is reflected back to
sensor #1 and T is substantially reduced. Once the sensor pair passes the
defect, T returns to its initial value of 1. Monitoring transmission
coefficients at short ranges can thus provide a defect detection tool in

201
addition to monitoring reflection coefficients at long ranges. Hence the
potential for an improved reliability of defect detection. Figure 6.6
demonstrates the transmission approach.
Time Signal 1 Time Signal 2

60 60
40 First Arrival 40
(a) First Arrival
20 20 (b)
0 0
-20 -20
-40 -40
-60 -60
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
GWT of Signal 1 GWT of Signal 2

45 45
40 40
Frequency (KHz)
Frequency (KHz)
First Arrival First Arrival

35 35
(c) 30 30 (d)
25 Defect echo 25
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Figure 6.6: Detection of transverse crack by transmission measurements

(detection angle 0º). (a) and (b) time histories; (c) and (d) continuous
wavelet scalograms.
202
The sensor orientation angle is 0º and lift-off distance is 76 mm
(3”). Figure 6.6(a) and (b) show the time signals detected by the two
sensors when they are positioned on either side of the 100% head crack
defect. The signal at sensor #1 is clearly larger than that at sensor #2
because much of the energy is reflected, rather than transmitted past the
defect. The corresponding wavelet scalograms are shown in Figure 6.6(c)
and Figure 6.6(d), respectively. The defect reflection is clearly visible at
sensor #1 besides the first arrival. This demonstrates that reflection data
can be monitored concurrently with transmission data. Figure 6.7 gives the
transmission coefficients measured at 30 kHz as the sensor pair was
scanned over the defect (the distance from sensor #2 to the defect, L2, is
shown in Figure 6.6). As expected, T is close to 1 when both of the
sensors are on the same side of the defect. T decreases substantially when
the sensors are on either side of the defect, providing a clear indication of
its presence.
The same tests were performed on 100% head cracks oriented at
10º, 20º and 35º from the transverse rail direction. These oblique defects
produced the same trend of T as a function of sensor position, although
the results were more scattered than for the perfectly-transverse crack.
6.3 Use of the Discrete Wavelet Transform to Enhance Defect Detection
The extraction of the reflection and the transmission coefficients
presented so far was performed by transforming the signal in the CWT
time-frequency domain. Although efficient in the laboratory, this approach
is too slow for field application. The Discrete Wavelet Transform (DWT)
203
is the fast version of the CWT and it can be implemented in real-time as
required by high-speed inspections in the field (Appendix A). Besides
increasing speed, the DWT performs an excellent de-noising of the signals
thereby increasing the chances of properly identifying the signature of a
potential defect that would be otherwise missed or misclassified. This is
particularly relevant in the case of small defects, discussed in the
following, whose signatures may be buried into noise.

The DWT operated here uses a series of low-pass and high-pass
filters to decompose the original time signal into various “levels”
representing different frequency bands (filter bank decomposition) (Mallat,
1999). Each decomposition level, j, corresponds to a center frequency, f(j),
given by:
∆×F
f (j) = (6.4)
2j
where ∆ is the sampling frequency of the original signal and F is the
center frequency of the particular mother wavelet used.
1.2
Transmission coefficient
0.8
0.6
DEFECT
0.4
NO DEFECT NO DEFECT
0.2
0
1500 1000 500 0 -500 -1000
L2 (mm)
Figure 6.7: Transmission coefficient for transverse crack at 30 kHz as a

function of position of the sensor pair relative to the defect.
204
0.04
Detail Level 6 Thresholding
0.02
(a) 0
-0.02
-0.04
0 1000 2000 3000 4000
Detail Level 6 First arrival

0.04
0.02 Defect echo

(b) 0
-0.02
-0.04
0 1000 2000 3000 4000
1.2
Transmission coefficient
0.8
0.6
DEFECT
0.4
NO DEFECT NO DEFECT
0.2
(c)
0
60 40 20 0 -20 -40
L2 (mm)
Figure 6.8: Use of the discrete wavelet transform for signal de-noising.
(a) signal reconstructed from level 6; (b) signal after thresholding
wavelet coefficients at level 6; (c) transmission coefficient after wavelet
processing as a function of position of the sensor pair relative to the
defect.
205
Figure 6.8(a) shows the time signal reconstructed from the level 6
DWT decomposition using the Daubechies wavelet of order 10 (db10).
The air-coupled sensor was 1,120 mm (44”) away from the 100% head
crack with a lift-off of 76 mm (3”) from the rail top surface. From Eq.
(6.4), level 6 is centered at 26 kHz that contains most of the reflected
energy. The plot in Figure 6.8(a) resembles closely the raw time history
recorded by the sensor. The defect reflection is not easily distinguishable
from the first arrival due to noise and reverberations. If only the largest
wavelet coefficients are retained, the reconstructed signal is de-noised and
the defect reflection now appears clearly (Figure 6.8(b)). This is called
thresholding procedure within a given decomposition level. In Figure 6.8
the threshold was set to 60% of the maximum value of the wavelet
coefficients. Another important product of the DWT is signal compression.
At each decomposition level, the number of points is halved
(downsampling). Thus an original signal that had 4,000 points is reduced
to only 60 point at level 6 and much less after thresholding. The task of
extracting reflection and transmission coefficients from such compressed
signals results in a very efficient process. By using DWT processing,
reflection and transmission coefficients were extracted for perfectly-
transverse head cracks, and oblique head cracks at 10º, 20º and 35º
orientations.
An example of transmission coefficients extracted by thresholding
the DWT level 6 is given in Figure 6.8(c) as the sensor pair is moved
over the transverse defect. The trend is similar to that shown in Figure
206
6.7. The main difference with the previous plot is the fact that these
results were obtained in real-time owing to the DWT processing. Real-
time defect detection capability is a necessary requirement for high-speed
inspections in the field.
6.3.1 Defect detection using laser generation/air-coupled detection
For the detection of small defects, on the order of a few millimeters
in depth, ultrasonic frequencies higher than those excited by an impulse
hammer are needed. High-frequency testing was accomplished by the use
of a pulsed, Nd:YAG laser operating at 1064 nm with a pulse duration of
8 nsec. This broadband source can generate effectively frequencies as high
as a few tens of Megahertz. As it will be shown later, the need for large
inspection ranges and the use of air-coupled detectors positioned at large
lift-off distances limit the operating frequency range to 100 kHz – 1 MHz.
The schematic of the laser/air-coupled system is shown in Figure
6.9(a). The laser was focused on a line to guide the generated ultrasonic
field. A pair of air-coupled sensors was used for monitoring reflection and
transmission coefficients from the defects (Figure 6.9(b)). The system was
tested on two types of small, surface-breaking transverse defects, namely a
horizontal crack at the top of the rail head (Figure 6.9(c)), and a 45º
oblique crack at the gage-side corner of the rail head (Figure 6.9(d)). The
cracks were tested at increasing depths from 1 mm to a maximum of 10
mm. A 50 kHz high-pass filter was employed for the signals shown in the
following sections.
207
6.3.2 Transverse Crack at the Top of Rail Head
Figure 6.10 illustrates the detection of the horizontal crack at a
depth of 5 mm (0.2”). The time waveforms, the corresponding CWT
scalograms, and the DWT de-noised signals are presented in the left
column for sensor #1, prior to the defect, and on the right column for
sensor #2, past the defect. The sensor orientation was 0º and the lift-off
distance was 76 mm (3”). The time waveforms (Figure 6.10(a) and Figure
6.10 (b) appear very noisy. The scalograms (Figure 6.10 (c) and Figure
6.10 (d)) indicate that most of the signal energy is in the 100 kHz – 700
kHz range.
The reflection from the crack is clearly seen in the DWT
reconstructed signal for sensor #1 in Figure 6.10 (e). In this case the de-
noising was performed after thresholding decomposition levels 3 and 4 that
corresponded to frequencies of 425 kHz and 210 kHz, respectively. The
db10 mother wavelet was used for the processing. The DWT reconstructed
signal for sensor #2 in Figure 6.10 (f) shows a disappearance of the defect
reflection and a decrease of the first arrival as expected. When the sensor
orientation is 0º, the smallest detectable crack is 4 mm (0.15”) in depth.
However, by optimizing the sensor orientation, a substantial increase in
defect detection sensitivity can be achieved. Figure 6.11 demonstrates the
detection of the crack at a depth of only 1 mm by reflection
measurements. In this case the sensor orientation was set to 6º to optimize
the sensitivity to the defect reflection following Eq. (6.1).
The DWT reconstructed traces (thresholding levels 3 and 4) for the
sensors (Figure 6.11(e) and Figure 6.11(f)), now positioned both prior to
208
the crack, clearly show the defect reflection signature after the first arrival.
Once the sensor pair was moved over the crack, the usual decrease in
transmission coefficient was observed. The laser/air-coupled setup is thus
effective to detect transverse cracks as shallow as 1 mm by both reflection
measurements and transmission measurements. Figure 6.11 also shows that
such a crack can be detected as far away as 500 mm (20”) from the
detecting sensor. Such distance was the largest allowed by the
experimental setup, and it thus should not be considered an absolute limit
on the range achievable by the inspection. Clearly, cracks deeper than 1
mm can be detected at even larger ranges. Considering that the system is
completely non-contact and the inspection probes are located far away
from the rail surface, these results are very encouraging for meeting the
objectives of this study.
Figure 6.12 illustrates the effect of crack depth on the frequency
content of the reflected signal. The reflection coefficients are here

obtained from the CWT of sensor #1. It can be seen that the 5 mm
and 7 mm deep cracks reflect mostly energy at 200 kHz. The smaller, 2
mm deep crack, reflects instead mostly at 600 kHz. This confirms the
assumption that higher frequencies are needed to detect smaller defects.
As commented previously, values of R larger than 1 result from the
particular sensor inclination adopted. Figure 6.12 also suggests that
monitoring the frequency content of the reflected signal can provide a
means for sizing the crack, beyond simply detecting it.

209
6.3.3 Transverse Crack at the Gage Corner of the Rail Head
Figure 6.13 shows the detection of the oblique crack at a 1 mm
depth located at the gage-side corner of the rail. The sensor orientation of
6º was utilized for optimum detection of the defect reflection. The DWT
reconstructed signals clearly show the presence of the defect reflection in
sensor #1 prior to the defect (Figure 6.13(e)), and a decrease in the first
arrival energy in sensor # 2 past the defect (Figure 6.13(f)). It is thus
confirmed that it is possible to detect the corner crack at the shallow
depth of 1 mm from both reflection and transmission measurements.
(a)
(b) (c) (d)
Figure 6.9: Laser/air-coupled system for detecting small transverse cracks.

(a) and (b) experimental setup; (c) and (d) types of defects considered.
210

1 1
0.5 0.5
0 0
(a) (b)
-0.5 -0.5
-1 -1
0 0.5 1 1.5 0 0.5 1 1.5

1000 1000
First Arrival
800 Defect Echo 800
Frequency (KHz)
Frequency (KHz)
First Arrival
600 Lift-off Echo 600

End Echo
(c) (d)
400 400
200 200
0 0.5 1 1.5 0 0.5 1 1.5

Reconstructed signal #1 (Level 3 and 4) Reconstructed signal #2 (Level 3 and 4)

1 1
First Arrival Defect Echo
First Arrival
0.5 Lift-off Echo 0.5
(e) 0 0 (f)
-0.5 -0.5
-1 -1
0 0.5 1 1.5 0 0.5 1 1.5
Figure 6.10: Detection of 5 mm crack by laser/air-coupled system

(detection angle 0°). (a) and (b) time histories; (c) and (d) continuous
wavelet scalograms; (e) and (f) de-noised signals after discrete wavelet
processing.
211

0.6 0.6
0.4 0.4
0.2 0.2
0 0
(a) (b)
-0.2 -0.2
-0.4 First Defect Echo -0.4

First Defect Echo
Arrival
Arrival
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8

1000 1000
800 800
Frequency (KHz)
Frequency (KHz)
First Defect Echo
First
Arrival
Arrival
600 600
(c) (d)
400 400
Defect Echo
200 200
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8


0.6 0.6
0.4 First 0.4 First

Arrival Arrival
0.2 0.2
(e) 0 0 (f)
-0.2 -0.2
Defect
Echo Defect
-0.4 -0.4
Echo
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8

T= 6°
Figure 6.11: Detection of 1 mm crack by laser/air-coupled system

processing.
212
Figure 6.13 also shows that this crack is detectable at distances of
at least 300 mm (12”) from a sensor. Cracks deeper than 1 mm, up to a
depth of 10 mm, produced even more pronounced signatures and can be
detected at even larger distances.
10.0
8.0
s = 2 mm
Reflection coefficient
6.0
s = 5 mm
s = 7 mm
4.0
2.0
0.0
0 200 400 600 800 1000
Frequency (kHz)
T= 6°
Figure 6.12: Reflection coefficient spectra for horizontal crack as a function

of crack depth (detection angle 6°).
213

1 1
Defect Echo
First Arrival
0.5 First Arrival 0.5
(a) 0 0 (b)
-0.5 -0.5
-1 -1
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

1000 1000
Defect Echo
800 800
Frequency (kHz)
Frequency (kHz)
First Arrival First Arrival
600 600
(c) (d)
400 400
200 200
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5

1 1
Defect Echo
First Arrival
0.5 First Arrival 0.5
(e) 0 0 (f)
-0.5 -0.5
-1 -1
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
T= 6°
Figure 6.13: Detection of 1-mm corner crack by laser/air-coupled system

processing.
214
In analogy with Figure 6.12 for the horizontal crack, Figure 6.14
shows the effect of corner crack depth on the frequency content of the
reflected signal monitored by sensor #1. It can be seen that all crack
depths tend to reflect most efficiently in the 200 kHz – 400 kHz range,
with the reflections generally increasing with increasing crack depths.
Frequencies higher than 500 kHz did not produce an increase in defect
detection sensitivity in this case, even for the smallest flaws. The changes
in reflection coefficient can be used to size the crack.
5.0
s = 1.5 mm
4.0
s = 3 mm
Reflection coefficient
s = 4 mm
3.0 s = 8 mm
s = 10 mm
2.0
1.0
0.0
0 200 400 600 800 1000
Frequency (kHz)
T= 6°
Figure 6.14: Reflection coefficient spectra for corner crack as a function of

crack depth (detection angle 6°).
215
6.4 Conclusions
This study demonstrates two inspection systems based on ultrasonic
guided waves for the detection of transverse head cracks in rails that
potentially surpass the capabilities of current rail defect detection systems.
The main advantages of the proposed technology include 1) an inherent
sensitivity to transverse defects despite the presence of surface shelling, 2)
an increased inspection speed due to non-contact conditions and large

inspection ranges, and 3) an increased defect detection reliability due to a
dual detection scheme based on both reflection and transmission
measurements.
A pair of air-coupled sensors is used to provide both reflection and
transmission information on a crack. The ultrasonic signals are processed
by the discrete wavelet transform in real time to enhance the defect
detection sensitivity. Transverse cracks larger than 15% of the head cross
sectional area can be detected by using ultrasonic frequencies below 50
kHz that can be excited by an impulse hammer. Smaller transverse cracks,
as shallow as 1 mm in depth, can be detected by using ultrasonic
frequencies between 100 kHz and 600 kHz that can be excited by a pulsed
laser.
The inspection range (distance between a detectable crack and the
inspection probes) achievable by the proposed technology is at least 10 m
(32’) for cracks larger than 15% head area and at least 500 mm (20”) for
surface-breaking cracks as shallow as 1 mm. It is thus possible to probe a
large portion of the rail at once, thereby increasing inspection speed. This
is in contrast to the conventional rail inspection that is carried out more
locally on a single cross-section at a time. Successful results were shown

216
for lift-off distances of the detecting sensors as large as 76 mm (3”) from
the top of the rail head, satisfying the recommended clearance envelope for
new rail inspection systems.
A prototype based on the laser/air-coupled combination has been
assembled and field tested twice. A summary of prototype components and
results of field tests is given in a recent technical report submitted to the
Federal Railroad Administration (Coccia et al., 2007).

OF MULTI-WIRE STRANDS
7.1 Introduction
Multi-wire steel strands are used in civil engineering as the
tensioning components of prestressed concrete structures and in cable
systems of cable-stayed and suspension bridges. The loss of prestress as
well as the presence of defects or the tendon breakage can be catastrophic
for the entire structure. Some of the documented collapses of post-
tensioned structures caused by tendon failures have occurred in Wales
(Woodward, 1988), Palau (Parker, 1996), North Carolina (Chase, 2001),
and Berlin (Schlaich et al. 1980). Monitoring of the applied loads and
assessment of the structural integrity of strands are engineering tasks that
remain a challenge in the NDE and SHM communities.
Many techniques have been applied to the defect detection and load
monitoring of prestressing tendons in prestressed and postensioned concrete
structures. The visual inspection technique especially used in bridge
engineering is probably the simplest and oldest form of inspection
(Williams and Hulse, 1995). It is considered efficient only when the
degradation is visible at the surface of the structure; this may be too late
for a cost-effective repair strategy.
Although very effective in many applications, radiography (Kear and
Leeming, 1994) encounters several limitations in concrete structures due to
the highly scattering nature of concrete. Also, safety issues dissuade from
its use in open and large areas.
217
218
Global approaches have been employed for the measurement of stay
cable loads which relies on modal-analysis techniques (Casas 1994,
Tabatabai et al. 1998, Cunha et al. 2001). In this case the cable natural
frequencies are correlated to the applied loads following the vibrating chord
theory. The technique, unfortunately, is not sensitive to small defects such
as corrosion and it is not applicable to embedded PS tendons.
For the detection of small defects in cables of stay-cable and
suspension bridges, the Magnetic Flux Leakage (MFL) method has gained
some interest especially in Europe (Weischedel, 1995). The MFL method
relies on the fact that a defect in a ferromagnetic object introduces a
localized discontinuity of the magnetic properties of the system. The MFL
method cannot be used for continuous health monitoring and it does not
provide any information on the level of stress in the cable.
An alternative method, recently presented by Bouchilloux et al.
(1999) and by Wang et al. (2001), is based on measuring the magnetic

permeability of tendons and cables to determine the applied stress. Current
configurations of the magnetic permeability sensors do not allow their use
in embedded PS tendons. Also, this technique does not provide information
on localized defects in the tendons.
Time-domain Reflectometry (TDR), a technique developed for
locating discontinuities in electrical transmission lines, has been used to
characterize and locate faults in suspended bridge cables (Liu et al., 1998)
and, more recently, to detect voids and corrosion in grouted PS tendons of
post-tensioned bridges (Chajes et al., 2003). The method can be considered
as the electrical equivalent of the known pulse-echo testing method of
ultrasonic testing where an electric pulse substitutes the ultrasonic pulse.

219
Although the potential for void detection, the TDR cannot be used to
monitor prestress levels in the tendons.
Elastic stress waves technique have been successfully used,
particularly the Impact Echo (IE) method (Sansalone et al. 1996,
Watanabe et al. 1999), to detect voids in tendon ducts of post-tensioned
concrete slabs. In IE, a short duration mechanical impact at a free surface
of the slab generates low-frequency waves that propagate through the
thickness; voids are detected by early wave echoes detected at the free
surface. The IE method cannot detect defects within the tendons, e.g.
corrosion or wire breakage, and it cannot monitor stress levels in the
tendons.
A technique that shows promises for the simultaneous detection of
defects and monitoring of prestress levels in PS tendons is based on
Guided Ultrasonic Waves. As opposed to the waves used in traditional IE
that propagate in 3-D within the post-tensioned concrete, GUW propagate

along the tendon itself by exploiting its waveguide geometry.
The advantages of this technique over the others mentioned above
include: (1) the possibility of using transducers permanently attached to
the strand for continuous structural monitoring, (2) the potential for
providing simultaneous defect detection and stress monitoring capabilities
for the strands with the same sensing system, and (3) the possibility for
detecting both active defects and pre-existing defects toggling between the
modes of “passive” acoustic emission testing and “active” ultrasonic testing
within the same sensing system.
Recent applications of the GUW technique were demonstrated for
the evaluation of stress levels in post-tensioning rods and multi-wire

220
strands (Kwun et al., 1998; Chen and Wissawapaisal, 2002; Washer et al.,
2002; Lanza di Scalea et al., 2003; Rizzo and Lanza di Scalea, 2004), as
well as for the detection of isolated defects in these components (Kwun
and Teller, 1995; Kwun et al., 1998; Beard et al., 2003; Rizzo and Lanza
di Scalea, 2004). The GUW technique can be also used in a passive
mode, for the real-time monitoring of active defects using acoustic emission
principles. Successful use of the passive approach has been demonstrated
for both steel and composite cables (Kwun and Teller, 1995; Casey and
Laura, 1997; Rizzo and Lanza di Scalea, 2001).
This chapter presents recent advances in the area of multi-wire
strand monitoring by Guided Ultrasonic Waves, focusing on the “active”
mode involving external generation and detection of waves. Wave
propagation models are first presented to predict the complicated dispersive
solutions of the strand waveguide in both free and embedded
configurations. Experimental results are then presented with application to

defect detection and stress monitoring in free strands. For defect detection
purposes, the proposed system uses magnetostrictive transducers, wavelet-
based signal processing, and statistical pattern recognition. For stress
monitoring purposes, the proposed approach uses piezoelectric transducers
which monitor the ultrasonic “cross-talk” between individual wires
comprising the strand.

221
7.2 Wave propagation in helical waveguides
7.2.1 Geometry of the strand
The typical geometry of the strand is represented in Figure 7.1. The
strands considered in the experimental tests and modeled in the numerical
simulations are formed by 6 helical wires wrapped around a core. Each
helical wire is assumed to have a circular cross section in a plane normal
to its axis, the diameter of which is smaller in comparison to the pitch of
the helix. Depending on the load conditions, each helical wire can be in
contact with the two adjacent wires, the core, or with both core and
adjacent wires. Strands for post-tensioning concrete structures are under
axial loading (pulling) that is typically associated with torsion, depending
on the boundary conditions. A detailed study of the static behavior of the
strand subjected to combined torsion and axial tension was provided by
Machida and Durelli (1973). The strand specimens tested, provided by
“Dsiamerica” http://www.dsiamerica.com/products/Multistrand_System.html)
are Grade 270, 15.2 mm (0.6”) diameter, seven-wire strands with a cross
sectional area of 140 mm2.
Principal geometric properties are dc=dh=0.2”=5.08mm with dc and
dh diameters of core and helical wires respectively; the lay angle β of the
helical wire can be evaluated as:
⎛ 2πR ⎞⎟
β = arctan ⎜⎜ ⎟ = 7.9 deg (7.1)
⎜⎝ p ⎠⎟⎟
222
where R=0.2”=5.08mm is the radius measured from the center of the
strand to the centre of helical wire and p=230mm is the pitch of the
helical wire.
Helical wire Core wire

dh dc
Axis of reference
of cylinder
Wire cross
ρ p
section s β
r
Axial line of
helical wire R
R β=Lay angle of helical wire

ρ=Radius of curvature of the
axis of helical wire
s=Arc Length of helical wire
R=Radius of strand (radius
of reference cylinder)
dc p=Pitch of helical wire
A=cross sectional area of wire
R d=Wire diameter
Subscripts c (core wire)

and h (helical wire)
Figure 7.1: Geometry of the strand: helical wire wrapped around a core.
223
7.2.2 Dispersion properties of unloaded strands
An unloaded strand behaves quite differently compared to a loaded
one. In the absence of axial load, the six helical wires and the core wire
can be seen as independent waveguides. The following experimental results
were extracted from a series of through transmission Laser Ultrasound
(LU) tests. A pulsed Nd:Yag laser operating at 1064 mm with a pulse
duration of ~8ns was used to generate at one end of the strand, while at
the other end, a ultra-mini broadband sensor of the Physical Acoustic
Corporation (PICO type) detected the incoming waves. A similar approach
was used by Rizzo and Lanza di Scalea (2004) that used a Michelson
interferometer with balance detectors to detect the ultrasonic guide waves
generated by a pulsed laser. Compared to the experimental study by Rizzo
and Lanza di Scalea (2004), the following results cover a broader frequency
range (f<1MHz).
7.2.2.1 Single wire
Two wires, the straight core and a peripheral helical wire, were
isolated by disassembling a 720mm long, 15.24mm (0.6”) diameter strand
and tested in a simply supported configuration.
A SAFE numerical simulation was performed to predict the
acceleration time history due to a broadband source and to help analyzing
the experimental results shown in the following. The cross section of a
5.08mm diameter steel bar was meshed using 720 triangular elements.
Computational effort was considerably reduced considering the symmetry of
the problem. An impulsive concentrated force was applied in the central
node of the circular section and inclined of 45deg with respect to the
224
longitudinal axis z, to excite longitudinal and flexural modes (Figure
7.2(a)). The acceleration time history and its corresponding continuous
wavelet scalogram are shown in Figure 7.2(b)-(c). To represent the
response of the PICO sensors, the acceleration was computed as the sum
of the accelerations of the nodes at z=0.72m. Each acceleration was
computed by numerically differentiating the displacement time histories
evaluated as in Chapter 3 (Eq. 3.97).
The joint time frequency scalogram shown in Figure 7.2(c) was
carried out adopting the Complex Morlet Wavelet (center frequency equal
to 5, bandwidth parameter equal to 2). As shown in the scalogram, the
Complex Morlet Wavelet transform successfully isolates the first two
longitudinal L(0,1), L(0,2) and flexural F(1,1), F(1,2) modes. It can be
noted the strong dispersive behavior of the first flexural mode F(1,1) at
low frequency as well as the sudden velocity change of the fundamental
longitudinal mode L(0,1) at around 600kHz. From the scalogram shown in

Figure 7.2(c) the group velocity dispersion curve can be extracted for all
the modes excited by the impulsive force shown in Figure 7.2(a) as
cg = D / t(f ) (7.2)
where t(f) is the arrival time of the scalogram local peak corresponding to
the generic mode at frequency f and D is the length of the rod.
Theoretical arrival times were also computed directly from the
dispersion group velocities computed with the SAFE. The curves obtained
were superimposed to the scalogram of the experimental signals. The time
waveforms and the scalograms recorded by the LU through-transmission

225
setup are shown in Figure 7.3 for a single straight wire (central core) and
in Figure 7.4 for a helical (peripheral) wire.
Both time histories include a reflection from the wire end of the
fastest fundamental longitudinal mode L(0,1) that appear at around
420µsec in Figure 7.4(a) and in Figure 7.5(a). The reflected L(0,1) mode is
also evident in the continuous wavelet scalogram where it can be noted
that only the low frequency portion appears. The higher but slower
frequency portion of the L(0,1) mode (f>300kHz) arrives after 450µsec and
consequently is not captured by the time window considered. It should be
noted that reflection cannot be predicted by the SAFE numerical
simulation previously discussed. The simulated response is in fact a
summation of all the modes carrying energy and propagating in the right
direction in a semi-infinite circular rod.
From the scalograms shown in Figure 7.3(b) and Figure 7.4(b) it is
observed that the modes signatures are, as expected, slightly more

scattered when compared to the correspondent traces in the scalogram of
Figure 7.2(c). The differences depend mainly on the uncertainty introduced
by the coupling between PICO sensors and wire surface. Furthermore, the
sensors are affected by reflections and mode conversions occurring on the
wire end that are not predicted by the theoretical SAFE model and.
Finally, the laser source is another source of uncertainty. However, the
simulated scalograms successfully identify at least four propagating modes
measured in the experiment.
It should also be noted that the arrival times predicted by the
SAFE method (black curves) match considerably well the scalogram traces
of the two measured fundamental modes L(0,1), F(1,1) and higher order
226
modes L(0,2) and F(1,2). Differences between simulated and measured
arrival times are slightly more significant for the case of the peripheral
helical wire. However, comparing the scalograms in Figure 7.3(b) and
Figure 7.4(b) it can be concluded that the helical waveguide has dispersion
properties very similar to those of the straight central core waveguide in
the frequency range considered.
In the next section, waveforms acquired by the LU testing of a free
unloaded strand will be shown and their joint time-frequency transforms
analyzed.
7.2.2.2 Unloaded Strand
The study of an unloaded, seven wire strand was performed by
exciting the central core wire only. A general study of the unloaded and
loaded strand with laser ultrasound testing was previously done (Rizzo and
Lanza di Scalea, 2004). The interest here is only in the coupling between
the center and helical wires in the absence of load. Furthermore, the
strand considered is free and simply supported without wedge tendon
couplers that could produce some additional effect on the behavior of the
waveguide (analyzed later).
The tests shown here include a C-C (central-central) test where the
wave is detected in the central core and a C-P test where the wave is
received with the PICO sensor positioned on the helical wire. The
waveforms recorded for test C-C and C-P are very different (compare
Figure 7.5(a) to Figure 7.6(a)). For example, even though it could be

227
(a)
-3
x 10
2.5
F(t)
2
z 1.5
3µsec
t 0.5
0 x
-0.5
45 x -1
-1.5
F(t) -2
y -2.5
-3 -2 -1 0 1 2 3
-3
y x 10
-9
x 10
(b) 6
-2
-4
-6
0 100 200 300 400
Time (µsec)
(c)
L(0,2)
L(0,1) F(1,2)
F(1,1)
Figure 7.2: (a) Core wire modeled with SAFE. (b) Acceleration time
history obtained with SAFE numerical simulation for a steel circular bar
(E=195GPa, ν=0.29, ρ=7700kg/m3, Length=720mm) subjected to impulsive
load. (c) Normalized Continuous wavelet scalogram of time waveform.
228
0.03
(a)
0.02
0.01
-0.01
-0.02
-0.03
0 100 200 300 400
Time (µsec)
(b)
F(1,2)
L(0,1) F(1,1) Reflected
L(0,1)
Figure 7.3: (a) Measured time waveform from Laser Ultrasound testing of
single steel wire (central core, Length=720mm). (b) Normalized (Complex
Morlet) Continuous Wavelet Transform scalogram of time waveform.
Theoretical arrival times are superimposed to experimental time-frequency
transform.
229
0.3
0.2
(a)
0.1
-0.1
-0.2
0 100 200 300 400

Time (µsec)
L(0,2)
(b)
F(1,2)
L(0,1) Reflected
F(1,1)
L(0,1)
Time (µsec)
single steel wire (helical wire, Length=720mm). (b) Normalized (Complex
Theoretical arrival times are superimposed to the scalogram.
230
argued that different testing conditions can affect the strength of the
signals, the peak to peak amplitude of the waveform acquired in test C-P
is considerably smaller than the corresponding feature computed on the C-
C test signal. Moreover, while the scalogram of the waveform registered
during test C-C (Figure 7.5(b)) isolates the two principal longitudinal
modes and a small portion of F(1,1), the scalogram of the signal from C-P
test, shown in Figure 7.6(b), is extremely scattered and the identification
of guided modes is no longer immediate. Portion of the two longitudinal
modes are still visible in Figure 7.6(b) but longitudinal and flexural modes
are either poorly excited by the test configuration or severely attenuated.
It can be concluded that when the cable is unloaded, each wire of
the pretwisted seven-wire strand behaves as an independent waveguide.
Energy introduced in one of the wires can leak in the adjacent ones but it
stays mainly confined within the excited wire. A completely different
behaviour is expected when load is applied to the strand. In the next

section a brief discussion on the inter-wire contact stresses generated in a
loaded strand will be provided.
7.2.3 Interwire contact stresses in axially loaded seven-wire strands
The seven wire strand has a manufacturer’s specified Young’s
modulus of 195GPa a Poisson’s ratio of 0.29 and a characteristic breaking
load of 260.7kN. The diameters of the strand core wire and the individual
helical wires were measured as 5.08mm (0.2”), resulting in an effective
steel cross sectional area of 140mm2. From the area, the material ultimate
tensile strength can be calculated as 1.86GPa.
231
0.03
(a) 0.02
0.01
-0.01
-0.02
-0.03
0 100 200 300 400
Time (µsec)
(b) L(0,2)
L(0,1) F(1,1)
unloaded strand. Guided wave excited in central wire and received in the
same central wire (Test C-C, Strand length=640mm). (b) Normalized
(Complex Morlet) Continuous Wavelet Transform scalogram of time
waveform. Theoretical arrival times are superimposed to the scalogram.
232
-3
x 10
8
-2
-4
-6
0 100 200 300 400
Time (µsec)
L(0,2)
L(0,1)
Time (µsec)
unloaded strand. Guided waves excited in central wire and received in
helical wire (Test C-P, Strand length=640mm). (b) Normalized (Complex
Theoretical arrival times are superimposed to the scalogram.
233
Following Machida and Durelli (1973), the length s of the axial line
of a helical wire in an undeformed strand of length l can be expressed by
l
p2 + (2πR )
2
s= (7.3)
p
where p=pitch of the helical wire and R=strand radius. Applying Eq.
(7.3) to the test strand, the increase in axial line length of the helical wire
with respect to the central wire is a small as 0.76%. This length difference
can be neglected as signals propagating in the core wire and in the helical
wires are practically detected simultaneously.
When an axial load is applied to a strand whose ends are restricted
from rotating, the following forces develop: axial force Nc in the core wire;
and axial force Nh, bending moment Mh, and twisting moment Th in the
helical wires. The ratio between the axial force in the core wire and the
axial force in a helical wire is given by (Machida and Durelli, 1973)
Nc A 1
= c (7.4)
Nh Ah cos2 β
where Ac and Ah are the cross-sectional areas of the core wire and the
helical wire, respectively and β is the lay angle of the helical wire. For a
lay angle of 7.9° in the subject strands, Eq. (7.4) predicts that Nc is only
1% larger than Nh. The effects of bending and twisting moments in the
helical wires can be expressed as

σ bmax 2r sin2 β (7.5)
=
σa R

τ max
t
Gr sin 4β
= (7.6)
σ a
4RE cos2 β
where σbmax is the maximum cross-sectional normal stress due to the
bending moment; τtmax is the maximum cross sectional shear stress due to
234
twisting moment and σa represents the uniform normal stress due to the

axial force. In Eqs. (7.5) and (7.6), r is the radius of the helical wire,
and E and G are the material Young’s modulus and shear modulus,
respectively. For the test strand, Eqs. (7.5) and (7.6) yield the values of
σbmax/ σa=1.5x10-2 and τtmax/ σa=10-5, indicating that bending and twisting
stresses in the helical wires can be neglected. The test strands can be
therefore assumed subject to an axial load distributed uniformly over the
wires cross-section and producing only a state of uniform normal stress, σ,
in each individual wire.
The stress distribution produced by the contact between the seven
wires was also studied by Machida and Durelli (1973). One assumption in
their study is that the contact between helical wires and the core under
tension and torsion loadings generate a negligible relative motion along the
contact lines. This is in agreement with the maximum relative motion
between the wires previously estimated (0.76%). Motion is assumed
possible between adjacent helical wires but with minor frictional forces
involved. Consequently, the resultant force due to contact is directed
outward normally to the axial line of the helical wire as illustrated in
Figure 7.7(c).
According to Machida and Durelli (1973), the resultant of the
contact forces per unit length, P, is given by
P = 2Ph cos 60D + Pc = Ph + Pc (7.7)
where Ph is the resultant of the contact forces between adjacent helical

wires while Pc is resultant force due to contact between helical wire and
235
core. Equilibrium of an infinitesimal element of the helical wire allows to
compute P as
Nh
P = (7.8)
ρ
where the radius of curvature is
⎛ p ⎞⎟
2
1 ⎜⎜ ⎟
ρ = R+ (7.9)
R ⎜⎝ 2π ⎠⎟
Stress distribution and contact area can then be estimated according to
Hertz theory that models the interaction between two frictionless elastic
cylinders in contact. Referring to Figure 7.8, directions r and y define,

respectively, the horizontal axis (contact arc direction) and the vertical
axis (cylinder radial direction). The distribution of pressure exerted
between the two solids under a force per unit length P is given by:
2P r2
σz (r) = 1− 2 (7.10)
πa a
The half length of the contact arc can be evaluated with the following
expression:
8P (1 − ν2 ) dcdh
a = (7.11)
πE l (dc + dh )
where l is the length of contact along the axis of the wires. From Eq.
(7.11) it is observed that the contact length a is proportional to the
square root of the resultant P.

236
Pds=P ρ d φ
(a) Nc
ds (b)
Nh
Tc
Mh
Th
Nh Nh
ρ
Pds
ds
Th
ρ 2Nh d φ =Pds=P ρ d φ
2
Mh dφ
Nh
(c)
Ph
P 60 Pc
60
P=2Ph cos60 +P=
c
Ph =Ph + Pc
Figure 7.7: (a) Axial forces, bending and twisting moments acting on the
helical and central wires of a loaded strand. (b) Equilibrium of force in an
element of a helical wire. (c) Resultant contact force in the transverse
cross section of helical wire.
A set of 2-D non-linear finite element simulations were performed using
ABAQUS implicit to compute contact areas and stresses in between the
cross-section of the wires. A first simulation was carried out by modeling
two of the seven wires comprising the 0.6-in strand, as shown in Figure
7.8. The purpose of this analysis was to determine the level of mesh
refinement to properly represent the interwire contact stresses by
comparison with the Hertzian theoretical results. The interwire radial force
237
was calculated according to Eq. (7.8) where the radius of curvature of the
helical wire, ρ , is given in Eq. (7.9). For example, for the subject strand
loaded at 70% of U.T.S., the helical wire radius of curvature is ρ =0.27m,
and the helical wire axial force is Nh = 21 kN, resulting in an interwire
force of P≈96.5kN. The length of the contact arc from Hertz theory in
this case is a=0.073mm.
The contact model was treated as a classical case of non-linear
geometry using ABAQUS Newton-Raphson algorithm. Triangular, 3-node
plane-strain elements were used in the model with two degrees of freedom
per node. Once an appropriate mesh size was chosen to model the most
critical area surrounding the contact arc, the attention was focused on the
entire strand cross section.
The particular load conditions induced by the axial load (i.e.

resultant interwire force P every 60 degree angle) allowed to exploit the
symmetry of the strand cross section. As shown in Figure 7.9 only a 60
degree circular sector comprising the central wire and two peripheral wires
was analyzed and appropriate boundary conditions were assumed. The
results allowed to confirm that stresses are mainly concentrated in the
contact areas. For example, as predicted by Hertz theory, stresses along

the radial and normal directions decay past a radial depth ~3a.
The stresses induced by the axial load applied to the strand change
the behavior of the waveguide. It was already demonstrated in Section

7.2.2.2 that the individual wires are free to vibrate as independent
waveguides when the strand is unloaded; the contact interwire stresses

caused by the axial load, limit the relative motion between core and
helical wires. The contact stresses are in general considerably larger than
238
the stresses induced by the ultrasonic wave motion. Consequently the
contact stresses force the wires to remain “attached to each other”. In the
following sections the loaded strand will be considered as a global
waveguide and analyzed by SAFE accounting for the contact stresses.
σr
C1
P
r σy
C2
Figure 7.8: Reference system of Hertz contact problem
(a) (b)
Figure 7.9: Three-wire contact problem studied by ABAQUS to compute

the contact stresses in the seven-wire strand induced by axial loads. (a)
Schematic, (b) FE model with Von Mises equivalent stresses for axial load
equal to 70% of ultimate load.
239
7.2.4 SAFE dispersion properties of loaded strands (pre-twisted waveguide)
The SAFE numerical simulations reported in the following, considers
the contact between the cross section of the wires. This interaction is
modeled by imposing a contact length estimated according to Hertz theory
and confirmed by the 2-D numerical FE simulations performed with
ABAQUS. Consequently, the increase of the axial load is analyzed by
changing the cross section of the strand i.e. increasing the length of the
arcs of contact. In the SAFE discretizazion of the cross section, the
displacement of nodes shared by adjacent wires and belonging to a contact
arc are assumed equal.
A contact line lying on the reference cylinder is considered between
peripheral wires. A helical contact line is considered also between adjacent
peripheral wires.
The first SAFE simulations was simply performed to study the
effect of the rate of pretwist on the dispersion properties of the strand.
The SAFE formulation of pre-twisted waveguides was proposed by Onipede
and Dong (1996). The equations of motion for this structural component
can be written using a body coordinate system (ξ, η, ζ) related to the fixed
Cartesian framework (x, y, z) through the relation (see Figure 7.10):

ξ = x cos (β z) + y sin (β z)
η = −x sin (β z) + y cos (β z) (7.12)
ζ = z
where β is the uniform rate of pretwist in the axial direction. The above
coordinate transformation allows to write displacement vector (u), strain

240
vector (ε) and stress field (σ) with respect to the body coordinate system
as:
u (ξ, η, ζ , t) = ⎡⎣uξ w⎤⎦

T
uη
ε = ⎡⎣ εξξ εηη εζζ εηζ εξζ εξη ⎤⎦

T
(7.13)
σ = ⎡⎣σξξ σηη σζζ σηζ σξζ σξη ⎤⎦

T
Linear strain relations can be summarized as:
ε = Lξη u + Lζ u (7.14)
where
⎡ ∂ ⎤
⎢ 0 0 ⎥
⎢ ∂ξ ⎥
⎢ ⎥
⎢ ∂ ⎥
⎢ 0 0 ⎥
⎢ ∂η ⎥
⎢ ⎥
⎢ 0 0 β D⎥
⎢ ⎥ ⎛ ∂ ∂ ⎞⎟
Lξη = ⎢⎢ ∂ ⎥⎥ D = ⎜⎜η −ξ ⎟ (7.15)
⎢ β β D ∂η ⎥
⎜⎝ ∂ξ ∂η ⎠⎟⎟
⎢ ⎥
⎢ ∂ ⎥
⎢ β D −β ⎥
⎢ ∂ξ ⎥⎥
⎢
⎢ ∂ ∂ ⎥
⎢ 0 ⎥
⎢ ∂η ∂ξ ⎥
⎢⎣ ⎥⎦
and
241
⎡0 0 0⎤ ⎡ 0 0 0⎤
⎢ ⎥ ⎢ ⎥
⎢0 0 0⎥ ⎢ 0 0 0⎥
⎢ ⎥ ⎢ ⎥
⎢ ∂ ⎥⎥ ⎢
⎢0 0 ⎢ 0 ∂ ⎥⎥
⎢ 0
⎢ ∂ζ ⎥⎥ ⎢ ∂z ⎥⎥
Lζ = ⎢⎢ ∂ ⎥ = ⎢⎢ ∂ ⎥ (7.16)
⎢0 0 ⎥⎥ ⎢ 0 0⎥
⎢ ∂ζ ⎥ ⎢ ∂z ⎥
⎢ ⎥ ⎢ ⎥
⎢∂ ⎢ ∂ ⎥
⎢ ∂ζ 0 0 ⎥⎥ ⎢ 0 0⎥
⎢ ⎥ ⎢ ∂z ⎥
⎢0 ⎢ ⎥
⎣⎢ 0 0 ⎥⎦⎥ ⎣⎢ 0 0 0 ⎥⎦
ε ) for an isotropic waveguide are

Constitutive equations ( σ = C
⎡ σξξ ⎤ ⎡1 − ν ν ν 0 0 0 ⎤ ⎡ εξξ ⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢σηη ⎥ ⎢ ν 1−ν ν 0 0 0 ⎥ ⎢εηη ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ σζζ ⎥ E ⎢ ν ν 1 − ν 0 0 0 ⎥ ⎢ εζζ ⎥
⎢ ⎥ = ⎢ ⎥ ⎢ ⎥
⎢σ ⎥ (1 + ν ) (1 − 2ν ) ⎢ 0 ⎥ ⎢ ⎥
⎢ ηζ ⎥ ⎢ 0 0 1 − 2ν 0 0 ⎥ ⎢εηζ ⎥
⎢σ ⎥ ⎢ 0 0 0 0 1 − 2ν 0 ⎥⎥ ⎢ε ⎥
⎢ ξζ ⎥ ⎢ ⎢ ξζ ⎥
⎢σ ⎥ ⎢ ⎢ε ⎥
⎢⎣ ξη ⎥⎦ ⎢⎣ 0 0 0 0 0 1 − 2ν ⎥⎥⎦ ⎢⎣ ξη ⎥⎦
(7.17)
Hamilton’s principle can be used to generate the governing equations of
motion:
⎛1 ⎪⎧⎪ ⎪⎫ ⎞
⎜⎜ εdξ dη⎪⎬ dζ ⎟⎟⎟dt = 0
t2
δ∫ ∫ ⎪⎪∫∫ ∫∫
⎨ ρ u T u dξ dη − εT C (7.18)
⎜ ⎪ ⎟
t1 ⎜⎝⎜ 2 ζ ⎩ Ω Ω ⎭⎪ ⎠⎟
where a dot indicates derivative with respect to the time variable t, ρ is
the mass density and Ω is the area of the waveguide cross section.
Assuming a harmonic motion along the propagation direction, z≡ζ,
and interpolating the displacement components over the cross section by
standard finite element method, the approximated displacement field in the
e-th element takes the form

242
⎡N ⎤( )
e
⎢ ∑ N j (ξ , η ) U ξ j ⎥
⎢ j =1 ⎥
⎢ ⎥
⎢N ⎥
u(e) (ξ, η, ζ , t) = ⎢⎢ ∑ N j (ξ, η ) Uη j ⎥⎥ ei (kζ − ω t) = N (ξ, η ) q (e) ei (kζ − ω t) (7.19)
⎢ j =1 ⎥
⎢N ⎥
⎢ ⎥
⎢ ∑ N j ( ξ , η ) Uζ j ⎥
⎢⎣⎢ j =1 ⎥⎦⎥
where i represents the imaginary unit, q(e) is the displacement vector of
the e-th element with nodal displacement components Uj=Uξj, Vj=Uηj,
Wj=Uζj and N is the shape function matrix. Substitution of the last
expression in the Hamilton’s principle and subsequent standard finite
element assembling result in the following wave equation already derived
in Chapter 3 and here repeated for completeness
⎡K1 + ikK2 + k2K3 − ω 2M⎤ Q = 0 (7.20)

⎣ ⎦M
with M the number of total degrees of freedom of the cross-sectional mesh.
For the sake of brevity, expressions of the stiffness matrices K1, K2, K3
and mass matrix M are not shown here but were derived for example by
Onipede and Dong (1996). The eigenvalue problem in Eq. (7.20) is a two
parameter (k-ω) eigensystem. If material damping is neglected and only

propagative modes are of interest, the wavenumber can be assigned as a
real number and ω is adopted as eigenvalue parameter (see Chapter 3).
This formulation was adopted to study the dispersion properties of
the structural component shown in Figure 7.10(a). The material has the
following nominal properties: Young modulus E=195GPa, Poisson’s ration
ν=0.29 and density ρ = 7700 kg/m3. The mesh, shown in Figure 7.10(b)
was generated by Matlab’s “pdetool” and consists of 323 nodes and 512
243
triangular elements with linear interpolation displacement functions. The
phase velocity dispersion curves are shown in Figure 7.10(c) up to a
frequency of 700 kHz. Notice the complexity of the modes. Mode shapes
computed by SAFE for the three fundamental modes (longitudinal,
torsional and flexural) are shown in the same plot.
It should be noted that the contact between wires was considered
by imposing the same displacement to the common node of neighbor
elements pertaining to adjacent wires. The interaction is obviously more
complicated and depends on the applied prestressing axial load. Several
papers have studied experimentally the influence of the applied load on
the properties of ultrasonic guided waves propagating in strands. (Kwun et
al, 1998) first reported that in unloaded strands, the wave propagation
properties were the same as those seen in individual wires while under
tensile loading a completely different behavior was observed. The time
frequency analysis of signals detected in loaded specimens showed a shaded

area around 100kHz. Such spot, absent in the unloaded case, was
characterized by a highly dispersive wave with a behavior similar to that
near a cutoff frequency. In the current simulation, by imposing the
displacement continuity in the adjacent wires the modeled strand behaves
as a global waveguide. As a result many modes are present in the
frequency range shown in Figure 7.10(c). In particular, some of them have
a cutoff frequency slightly above 100 kHz that could justify the dispersive
behavior described by Kwun et al (1998).

244
W3
U3 z,w
V3
W1
U1 W2
(b) V1 U2
2D-mesh
V2
x
αβ Harmonic motion
ξ
(a)
η
y x,u
y,v
(c)
10000
9000
8000
7000
6000
5000
4000
3000
2000
1000
0
0 100 200 300 400 500 600 700
Frequency [kHz]
Figure 7.10: (a) Pretwisted waveguide representing a seven-wire strand; (b)

generic element of the surface Ω; (c) Phase velocities and mode shapes at
100 kHz of the strand.
245
A secondary effect of the loading is an increased energy loss of the
guided waves. In fact, it was reported by Kwun et al. (1998) that stress
waves generated in a single wire of the strand leak energy into the
surrounding wires due to the increasing contact forces. Again this behavior
can be captured assuming the waveguide formed by all the seven wires.
The dispersion solutions theoretically depend on the uniform rate of
pretwist in the axial direction β . However, for the waveguide considered,

dispersion properties are only marginally affected by the lay angle.
Dispersion curves were found unchanged for lay angles typically found in
seven-wire strands and consequently, in the following sections, the
waveguide will be assumed straight rather than pretwisted.
The SAFE will be used to compute the forced response of the
waveguide excited by a Piezoelectric PZT transducer attached to the
surface of a helical wire. This was the condition adopted in experimental
tests that will be also presented.
7.2.5 Forced solution for the free strand
In Chapter 7.3 experimental results will be shown revealing the
change in the strength of transmission and energy leakage of ultrasonic
guided waves in progressively loaded seven-wire strands. The SAFE
simulations described in the present section were performed to explain
some of the experimental results that will be shown in the final part of
the present Chapter.
The system modeled is shown in Figure 7.11. The ultrasonic guided
waves are generated using a PZT actuator bonded on the external surface
246
of a peripheral wire. The incoming wave is registered by a pair of PICO
sensors located at the strand’s free end, 0.889m from the source.
The structural coupling between the strand and the PZT actuator
was modeled assuming a couple of pin forces, with opposite sign, applied
at the PZT ends (pin force model). The simplified model is justified by
the so called “shear lag” effect (Crawley et al. 1987, 1990; Giurgiutiu,
2005). The “shear lag” effect represents the shear stresses transferred to
the structure by the PZT. The shear stress distribution depends on the
thickness and stiffness of the bonding layer that couples structural
component and PZT actuator. In addition, the mechanical properties and
the thickness of the PZT affect the interfacial shear stress. In the case of
ideal bonding (when the bond is thin and relatively stiff), the load transfer
takes place over an infinitesimal region at the PZT ends (Giurgiutiu,
2005). As a consequence, two pin forces can be applied to represent the
forces induced by the PZT actuator.

The response is computed by averaging the longitudinal
displacement components of the peripheral and central wire cross sections
located at 0.889m from the PZT actuator. Response in each node was
computed by using the SAFE expression for the forced solution derived in
Chapter 3 and here reported for completeness:
∑α
i ⎡⎣km (z − zS )⎤⎦
U (z, ω ) = ΦRup
mm (ω ) e (7.21)
m =1
247
Generated Toneburst
(a)
V(t) PZT
t actuator
Pico
Peripheral
Pico
Central
(b) x
Figure 7.11: (a) Guided waves excited in seven wire strand by a PZT
actuator and detected with PICO sensors at the end. (b) Mesh of the
strand cross section employed in SAFE model.
The displacement expressed in Eq. (7.21) provides the response of the

system in the frequency domain. The vector U (z, ω ) obtained can be
multiplied for the Fourier transform F (ω ) of the load time history (Eq.
3.96). An inverse Fourier transform finally provides the displacement time
history in the generic node of the strand cross-section due to an arbitrary
force. This approach was followed to simulate the frequency sweep that
was performed experimentally where a total of 66 narrowband tonebursts

248
(a) PICO Central

2.0E-08
1.8E-08
1.6E-08
1.4E-08
20%
1.2E-08
RMS
1.0E-08 40%
8.0E-09 60%
6.0E-09 80%
4.0E-09 100%
2.0E-09
0.0E+00
50 150 250 350 450 550 650
Freq [kHz]
(b) PICO Peripheral

4.0E-08
3.2E-08
20%
2.4E-08
RMS
40%
1.6E-08 60%
80%
8.0E-09 100%
0.0E+00
50 150 250 350 450 550 650
Freq [kHz]
RATIO PICO(C) / PICO(P)

(c)
1.4
1.2
1 20%
Ratio C/P (RMS)
0.8 40%
60%
0.6
80%
0.4 100%
0.2
0
50 250 450 650
Freq [kHz]
Figure 7.12: SAFE predictions of the RMS spectrum of the waveforms

recorded by (a) the Pico Central sensor, (b) the Pico Peripheral sensor
and generated by the PZT actuator. (c) Ratio between RMS spectra of
Pico Central and Pico Peripheral. 100% Load=70% of Ultimate load.
249
with central frequencies varying from 50 to 700kHz were generated. A
typical 3-cycle narrowband toneburst is shown in Figure 7.11(a). A total of
66 signals in the frequency domain was obtained for each SAFE
simulation. The effect of the two pin forces representing the PZT actuator
was considered separately and the total displacement response was finally
achieved by superimposing the two separate contributions.
The approach described was repeated for 4 different cases
corresponding to 4 axial loading conditions 5%, 20%, 60% and 100% of the
maximum load (70% of the Ultimate load). Eq. (7.11) was employed to
compute the contact arc between the wires and the cross-section of the
strand. The mesh adopted by the SAFE was updated for each loading
case, maintaining, however, the same number of elements.
Figure 7.12(a) shows the predicted Root Mean Square (RMS)
spectrum for the PICO sensor attached to the central wire. RMS was
computed on a gated portion of the time histories containing only the first
arrival of the wave packets. The gates assumed were consistent to the
time windows (tmin=150µsec, tmax=250µsec) employed to analyze the
experimental results shown in the next sections. Each curve in the plot
refers to a different percentage of the maximum load applied (70% of the
Ultimate tensile load). The curves do not show any appreciable difference
among different loading conditions. Figure 7.12(b) represents the Root
Mean Square spectrum for the PICO sensor attached to the peripheral
wire. Again, only small variations of the RMS spectra curves can be
observed. Figure 7.12(c) shows the Ratio between RMS spectra of the
signals from the Pico Central and the Pico Peripheral. This feature seems
also unaffected by the change of the contact arc between the wires.
250
However, experimental tests have shown that the RMS ratio is dependent
on the axial load applied to the strand, indicating that inter-wire contact
is not the only parameter that should be considered influencing wave
propagation in loaded strands.
It should be noted that SAFE model considered the deformation of
the cross section due to the radial contact forces P generated by the axial
load. Other effects as elongation of the strand were evaluated as well.
These showed marginal effects on the SAFE results and therefore are not
presented here.
A few SAFE simulations were attempted to consider the effect of
the stress on ultrasonic wave speed (Acoustoelastic effect). According to
Acoustoelastic theory, the velocities of ultrasonic waves in prestressed
solids show measurable variations with respect to their counterparts in
stress free bodies. Some studies attempted using the covariance effect to
monitor prestress level in strands. The works by Chen et al. (1998) and
Chen and Wissawapaisal (2001) employed conventional piezoelectric
transducers positioned at the end of the prestressing tendons to generate
and detect the waves. Washer (2001) applied the same concept for the
evaluation of stress levels in prestressed rods and strands. Finally, Lanza
di Scalea et al. (2003) employed magnetostrictive transducers to estimate
the variation of group velocities of ultrasonic guided waves traveling in
strands subjected to varying stresses.
All these studies observed that the sensitivity of the ultrasonic
speed to a preexistent stress field is very small and difficult to distinguish
from other effects such as temperature variations and elastic elongation. As
a confirmation, the SAFE numerical simulations performed including the

251
acoustoelastic change of ultrasonic bulk waves did not show robust changes
in the dispersion behavior of the loaded strand modeled.
In conclusion, SAFE models of loaded free strands can be
successfully used to separate completely-unloaded and loaded strands, but
the predictions are ineffective to determine the exact value of applied load.
In the next section, however, it will be demonstrated that by
considering the strand anchorage, SAFE was able to capture the main
propagation behavior observed in experimental tests of loaded strands.
7.2.6 Forced solution for the free strand: effect of anchorages
In the present section the effect of the anchorage will be considered
in the SAFE model of the loaded strand. Experimental results obtained by
Lanza di Scalea et al. (2003) showed the effect of anchorages on the wave
attenuation of ultrasonic guided waves in prestressed seven-wire strands. In
the mentioned work, the authors concluded that for an ultrasonic guided
wave packet traveling in the loaded strand across one of the anchorages,
the attenuation would increase substantially. This means that acoustic
losses are expected and that a considerable portion of the ultrasonic energy
is generally lost in the anchorages. Lanza di Scalea et al. (2003) exploited
successfully this acoustic leakage, using sensors mounted on the external
surface of the anchorages to perform acoustic emission (AE) monitoring of
damage evolution in strands progressively loaded to failure.
The SAFE method was here used to predict the effect of the
ultrasonic leakage into the anchorages (Figure 7.13). The loading condition
considered is the same as in Section 7.2.5. A couple of opposite forces

252
located at the ends of the PZT actuator generates the ultrasonic pulse.
The forced solution is initially obtained for all the nodes of the strand
cross section immediately before the anchorage using Eq. (7.21). The
distance between PZT actuator and anchorage is 0.56m (22”). The nodal
displacement components uB’(ω) computed on the cross section that
precedes the anchorage, are used to compute the nodal forces FB’’(ω) that
the incoming wave applies to the neighbor waveguide (see Figure 7.13 (a)).
Obviously, if two waveguides with the same cross section and physical
properties are connected in series, their global behavior has to be the same
of the equivalent waveguide.
In order to consider the effect of the anchorage, a waveguide of
larger diameter was coupled to the seven wire strand waveguide (see
Figure 7.15(a)). The larger waveguide is represented by a strand
surrounded by a 60mm long cylinder, shown in Figure 7.15, that has an
external diameter of 50mm (2”). Contact between the ring (anchorage) and
the strand was assumed equal to the length of the indentation induced by
the tendon couplers during the experimental tests. As an example, the
maximum indentation created by the wedges (due to the maximum axial
load = 70% of U.T.S.) on each peripheral wire was 2mm long. The tendon
couplers consisted of three-piece steel wedges (Coupler Type D, Dywidag
System international).
253
PZT
(a) actuator
Anchorage
Pico
Peripheral
FPZT(ω)
Pico
Central
(ω)
uB'(ω) FPZT(ω)
uA'(ω)
FB''(ω)
(c)
(b)
Figure 7.13: SAFE simulation of leakage of guided waves into the strand
anchorage: (a) guided wave excited by the PZT actuator and received by
the PICO sensors. (b) Mesh of the strand cross section coupled to the
anchorage and (c) mesh of the seven wire strand employed in SAFE
method.
254
When the waveguides combined together have different cross section,
the simple application of the equivalent nodal forces vector FB’’(ω) to the
right end side of the second waveguide introduces in general an
approximation. Part of the wave energy, in fact, is reflected back due to
the discontinuity of the waveguide cross section. For the scope of the
present study, the reflected waves are neglected. This approximation does
not affect substantially the response on the left end side cross section of
the anchored area uA’(ω), because, in the actual structure, severe mode
conversion and energy leakage in the loading machine jags prevent such
reflections.
The simulations used a total of 538 elements to discretize the cross
section of the strand (Figure 7.13(b)) and a total of 538+320=858
triangular elements to divide the cross section of the constrained strand
(Figure 7.13(c)). For obvious reasons, the geometry and the discretization
of the seven wire strand are exactly the same in both the cross sections.
Once the displacement uA’(ω) in section A’ was computed, the inverse
Fast Fourier Transform was applied to compute the time histories of each
node of the cross section A’. The time history of the PICO peripheral
(central) sensor was numerically predicted by summing the response of the
nodes of the peripheral (central) wire cross section.
Figure 7.14(a) and Figure 7.14(b) show the RMS spectra for the
PICO sensors bonded to the central and peripheral wires, respectively. It
can be noted that the plots contain also the RMS curve for the case of a
free strand (“No Wedge”). The latest curve refers to a fully loaded
(70%U.T.S.) strand but in the absence of anchorage.

255
(a) PICO Central

8.0E-09
6.0E-09
NoWedge
20%
RMS
4.0E-09 40%
60%
2.0E-09 80%
100%
0.0E+00
50 150 250 350 450 550 650
Freq [kHz]
(b) PICO Peripheral

8.0E-09
6.0E-09
NoWedge
20%
RMS
4.0E-09 40%
60%
2.0E-09 80%
100%
0.0E+00
50 150 250 350 450 550 650
Freq [kHz]
RATIO PICO(C) / PICO(P)

(c)
0.7
0.6
NoWedge
0.5
Ratio C/P (RMS)
20%
0.4 40%
0.3 60%
80%
0.2
100%
0.1
0
300 400 500 600 700
Freq [kHz]
Figure 7.14: Strand with anchorage: SAFE predictions of the RMS

spectrum of the waveforms recorded by (a) the Pico Central sensor, (b)
the Pico Peripheral sensor and generated by the PZT actuator. (c) Ratio
between RMS spectra of Pico Central and Pico Peripheral. Waveforms are
recorded on the left hand side of the anchorage. 100% Load=70% of
Ultimate load.
256
It can be noted immediately in Figure 7.14(a) that the wedge is
responsible for a substantial variation of the RMS spectra. While the
PICO central spectra presents a non uniform behavior in the frequency
range considered, the RMS curve of the PICO peripheral drops in the
entire frequency range as observed in Figure 7.14(b). This result depends
on the leakage of guided wave energy into the anchorage. The guided
wave generated by the PZT is unaffected until the wavefront reaches the
wedge. From this point, a considerable amount of wave energy carried in
the peripheral wires is lost into the wedge reducing the strength of the
signal (RMS) recorded in the peripheral wire.
Figure 7.14(c), which shows the Ratio between RMS spectra of
signals from the Pico Central and the Pico Peripheral sensors, confirms
this result. This ratio is now sensitive to the applied load. For example,
at f=500kHz, the RMS Ratio varies between 0.4 and 0.55 for loads
varying between 20% and 100% of the maximum load (70% U.T.S.). A
large jump is observed between the case of the free strand (“No wedge”),
where the RMS Ratio is equal to 0.15, and the 20% case.
The ratio between signal strengths of the central and peripheral
sensors will be confirmed as a suitable feature for load monitoring in the
strands by the experimental tests presented in the next section.
7.3 Experimental Study of Prestress Level Monitoring in Free Strands

(Selection of Optimum Wave Features and Sensor Lay-out)
Tests were performed at UCSD’s Powell Labs on the SATEC
M600XWHVL, 600 kip capacity, pneumatic test apparatus configured for

257
tensile loading. The specimen tested was a Grade 270, 15.2 mm (0.6”)
diameter, seven-wire strand having an U.T.S. of 1.86 GPa (270 ksi), an
yield stress of 1.67 GPa (243 ksi), and 1.82 m (72”) in length. Table 7.1
summarizes the properties of the strand.
A 10° serrated wedge was placed on the tendon and inserted into a
corresponding collar that was clamped pneumatically into the machine to
transfer loads. The tested length was 1.4 m (56”) with 0.33 m (13”)
extending on one end and 76 mm (3”) at the other end to allow for the
wedge, collar and sensor placement.
Figure 7.15(a) shows a picture of the strand on the testing machine.
Load-unload cycles were performed with 11 load steps in each cycle. The
steps were based on a percentage of 70% of ultimate load 182.4 kN (41
kip), consisting of a 0%, 20% (8.2 kip), 40%(16.4kip), 60% (24.6 kip), 80%
(32.8 kip), 100% (41.0 kip), and down to 80% (32.8 kip), 60% (24.6 kip),
40% (16.4 kip), 20% (8.2 kip), and 0%.
Table 7.1: Properties of the seven-wire strand tested on the SATEC

machine (http://www.dsiamerica.com/products/MultistrandSystem.html)
Yield strength fy 243 (ksi) 1,670 (N/mm2)
Ultimate strength 270 (ksi) 1,860 (N/mm2)

fu
Nom. diameter 0.6 (in) 15.24 (mm)
Cross sectional 0.217 (in2) 140 (mm2)

area
Weight 0.74 (lbs/ft) 1.102 (kg/mm)
Ultimate load 58.6 (kips) 260.7 (kN)
Modulus of 28,000 (ksi) 195,000 (N/mm2)

elasticity
258
(a) (b)
PZT3(same wire as
PZT2 and Pico(P))
Ms Transm
(375 turns 38AWG)
L2=15"
L4=14"
PZT1 PZT2
Ms Receiv
(450 turns 38AWG)
L1=20"
13"
L3=15"
PICO (P) PICO (C)

PICO sensors attached with thermal adhesive
PZT 3
PICO (P)
(c)
PICO (C)
Figure 7.15: (a) The 1.82-m, seven-wire strand installed in the SATEC
testing machine for stress monitoring tests; (b) ultrasonic sensor lay-out;
(c) pictures of the piezoelectric transmitter (PZT 3) on the peripheral wire
and the two piezoelectric receivers on the strand’s bottom end probing the
central wire, PICO(C), and the peripheral wire, PICO(P).
259
Figure 7.16: Magnetostrictive coil sensor probing entire strand.
A variety of sensors were installed on the specimen. A schematic of
the sensor lay-out is shown in Figure 7.15(a). Two strain gages were
placed at the strand’s midpoint to measure the axial strain on two
opposite peripheral wires. The following three different sensors were used
to transmit and receive ultrasonic waves in the strand:
• Two ultra-mini broadband sensors (Physical Acoustic Corporation PICO
type) placed on the bottom end of the strand, one on the central wire
and the other one on a peripheral wire – Figure 7.15(b), (c).
• Two magnetostrictive (MsS) coil sensors, designed at UCSD, made from
38 AWG enamel-coated copper wire – Figure 7.16. The coils were made
with 375 turns for the ultrasonic transmitter, and 450 turns for the
ultrasonic receiver. The position of the MsS’s varied slightly with the test
performed. The coils were protected by a Faraday cage (copper wire mesh)
to prevent electromagnetic interference.
• Three, 6.35 mm x 3.2 mm piezoelectric (PZT) sensors (APC
International) – Figure 7.15(b), (c). Attached by a heat activated adhesive,

260
the PZT sensors were placed on the peripheral wires, one at 0.889 m (35”)
and two at 0.5 m (20”) from the strand bottom end. Two of the sensors
were placed on the same peripheral wire as the previously mentioned
PICO sensor.
The measurement equipment used for signal generation and data
acquisition was mainly constituted by a National Instruments (NI),
modular PXI 1042 unit. The unit included an arbitrary waveform
generator card (PXI 5411) and two, 20GS/s 12-bit multi-channel digitizers
(PXI 5105). In addition, a gated RF amplifier (RITEC GA 2500) was
used to amplify the excitation to the ultrasonic transmitters. The amplified
signal was monitored by a LeCroy LT262 Waverunner oscilloscope to
ensure a 600 V peak-to-peak excitation. A frequency sweep of the
transmitter of either 50kHz – 700kHz or 700 kHz - 2MHz was performed
at each load step in the cycles. LabVIEW software developed at UCSD
was used to control the sensors, acquire and process the data.
7.3.1 Test Protocols
Three different test protocols were adopted, as summarized in Table
7.2. In Test 1 the PICO sensor on the central wire was used as the
ultrasound transmitter. Signals were acquired by the three PZT sensors
and the MsS receiver. Seven load-unload cycles were completed under Test
1. The full frequency sweep was performed in two separate bands of
50kHz-700kHz and 700kHz-2MHz.
In Test 2 the MsS transmitter was used as the excitation. The
three PZT sensors and the two PICO sensors were used as receivers. Two
261
load-unload cycles were performed in Test 2. Because of the poor efficiency
at high frequency of the MsS coils, only the low frequency range of 50kHz
– 700kHz was examined.
Table 7.2: Test protocols for prestress level monitoring in seven-wire

strand (designation of ultrasonic sensors refers to Figure 7.15(b)-(c)).
# load
PICO PICO Ms Ms Freq.
PZT1 PZT 2 PZT 3 unload
(C) (P) Trans Receiv range
cycles**
50kHz-
Test 1 Trans n/a Receiv Receiv Receiv n/a Receiv 7
2MHz
50kHz-
Test 2 Receiv Receiv Receiv Receiv Receiv Trans Receiv 2
700kHz
50kHz-
Test 3* Receiv Receiv Receiv Receiv Trans n/a Receiv 7
2MHz
* best testing configuration
** 0% to 70% of U.T.S. in five steps, each cycle including load and unload ramps.
Trans = ultrasound transmitter, Receiv = ultrasound receiver, n/a = not used.
In Test 3 the PZT sensor at 35” from the strand end was used as
the transmitter. Signals were acquired by the MsS receiver, by the
remaining two PZT sensors, and by the two PICO sensors at the strand
end. Seven load-unload cycles were performed in Test 3 for the full
frequency range of 50kHz-700kHz and 700kHz-2MHz. Test 3 proved to be
the optimum testing configuration as discussed in the next Section.
7.3.2 Results – Test 1
The Test 1 configuration – PICO (C) transmitter – showed
promising results for prestress level monitoring. The first load-sensitive
feature examined was the ultrasonic energy leakage between the central
262
wire and the peripheral wires. Figure 7.17 shows the energy leakage
measured by transmitting on the central wire by PICO (C) and receiving
on the peripheral wire by PZT 3. In this figure the low frequency range of
50kHz – 700kHz is shown, for both load and unload ramps. Figure 7.18
shows the same feature for the high frequency range of 700 kHz – 2 MHz.
In the two figures the strength of the signal decreases with increasing load
level, in both loading and unloading.
With increasing load level, the interwire contact and the wedge
strand contact increase and, at the same time, the wave damping increases
due to the larger cross-section of the waveguide.
The wave energy leakage in the anchorage and the increased wave
damping seem to dominate the Test 1 results, causing the observed trend
of decreasing signals with increasing load.
Comparing load and unload results, it appears that the load ramp
provides good discrimination among the different load levels, in both the
low and the high frequency ranges. Hence the potential for determining
the exact prestress level. The unloading ramp shows a distinct difference
between the unloaded strand and all of the loaded cases, indicting the
potential for detecting a complete loss of prestress but failing to determine
the exact level of applied prestress. The difference in behavior between
load and unload ramps can be attributed to the fact that interwire
contact stresses and contact interaction in the anchorages are not released
at once during unloading, until the load is completely removed. Tests
conducted at slower unloading rates may eliminate this phenomenon and
provide consistent results in both ramps.

263
PZT3 – LOAD RAMP

0.25
0.2
start0%
up20%
0.15
up40%
RMS
up60%
0.1
up80%
up100%
0.05
0
50 150 250 350 450 550 650
Freq [kHz]
PZT3 – UNLOAD RAMP

0.16
0.14
0.12 up100%
0.1 down80%
down60%
RMS
0.08
down40%
0.06 down20%
0.04 end0%
0.02
0
50 150 250 350 450 550 650
Freq [kHz]
Figure 7.17: Energy leakage between central wire and peripheral wire as a
function of applied prestress. Test 1: PICO (C) transmitting and PZT 3
receiving. Load and unload ramps. 100% load = 70% U.T.S. Low
frequency range (50 kHz – 700 kHz).
In conclusion, the RMS of the PZT 3 signal emerges as a promising
feature for detecting loss of prestress, although this technique requires a
baseline. The transmissibility PZT 3 / PZT 2, on the other hand, does

264
not require a baseline. Another feature which would not require a baseline
is the wave frequency shift, which was considered next.
PZT3 – LOAD RAMP

0.025
0.02
start0%
up20%
0.015
up40%
RMS
up60%
0.01
up80%
up100%
0.005
0
700 900 1100 1300 1500 1700 1900
Freq [kHz]
PZT3–UNLOAD RAMP
0.03
0.025
up100%
0.02 down80%
down60%
RMS
0.015
down40%
0.01 down20%
end0%
0.005
0
700 900 1100 1300 1500 1700 1900
Freq [kHz]
Figure 7.18: Energy leakage between central wire and peripheral wire as a
function of applied prestress. Test 1: PICO (C) transmitting and PZT 3
receiving. Load and unload ramps. 100% load = 70% U.T.S. High
frequency range (700 kHz – 2 MHz).
265
Test 1 – Peak frequency shift (50 kHz – 130 kHz)

MsS Rec (1st cycle)
0.6
0.5
0.4
(fload-f0)/f0
Test1a Up
0.3
Test1a Down
0.2
0.1
0
0 4 8 12 16 20 24 28 32 36 40
Load [kips]
MsS Rec (2nd cycle)
0.6
0.5
0.4
(fload-f0)/f0
Test1b Up
0.3
Test1b Down
0.2
0.1
0
0 4 8 12 16 20 24 28 32 36 40
Load [kips]
Figure 7.19: Shift in peak frequency as a function of applied prestress.

Test 1: PICO (C) transmitting, MsS coil receiving. Load and unload
ramps for two cycles. Max load 41 kips = 70% U.T.S. Frequency range 50
kHz – 130 kHz.
A second feature monitored under the Test 1 configuration was the
shift in maximum transmission frequency, which does not require a

266
baseline measurement. It can be noticed in Figure 7.17 that a considerable
shift in peak frequency exists with changing load level in the range 50
kHz – 130 kHz. It was anticipated that increasing load levels will produce
an increase in wave frequency in analogy with modal frequencies and the
vibrating cord theory. In order to best capture this phenomenon, the
signal recorded by the Ms Receiver coil was analyzed, since this sensor is
sensitive to the global strand dynamics, rather than the dynamics of any
individual wire.
Figure 7.19 shows the shift in maximum transmission frequency
relative to the unloaded case for both load and unload ramps and for two
separate load cycles. The peak frequency is that measured in the 50kHz–
130 kHz frequency range. As expected, the frequency increases with
increasing load. The behavior is consistent during load and unload, and
also between different cycles. The relative frequency shift is as high as
50% between zero load and maximum load of 41 kips (70% U.T.S.). In
conclusion, the frequency shift under Test 1 configuration provides another
means for monitoring prestress levels in the strand.
In Test 2 the MsS coil was used as a transmitter. This test set-up
showed poor results due to a large source of electromagnetic interference
at the location of the SATEC machine in the Powell Labs. Interestingly,
the coil behaved appropriately in prior tests not conducted in the Powell
Labs. Thus results from the Test 2 configuration will not be shown here.
267
Ongoing studies are being aimed at providing further shielding to the coil
against strong electromagnetic interference.
RATIO PICO(C)/PICO(P) – LOAD RAMP – 1st cycle
2.5
2
start0%
Ratio P/C (RMS)
up20%
1.5
up40%
up60%
1
up80%
up100%
0.5
0
300 400 500 600 700
Freq [kHz]
RATIO PICO(C)/PICO(P) – UNLOAD RAMP – 1st cycle

2.5
2
Ratio P/C (RMS)
up100%
1.5 dow n80%
dow n60%
1 dow n40%
dow n20%
0.5 end0%
0
300 400 500 600 700
Freq [kHz]
Figure 7.20: Energy leakage between peripheral wire and central wire as a
function of prestress. Test 3: PZT 3 transmitting, PICOs (C) and (P)
268
In Test 3 the wave was generated on the peripheral wire by PZT 3.
This proved to be the optimum testing configuration. The same features
monitored in Test 1, namely ultrasonic energy leakage and frequency shift,
were examined. However, the ability of Test 3 to monitor both the signal
propagating in the central wire – PICO (C), and that propagating in the
peripheral wire – PICO (P), provided a unique advantage which eliminates
the need for a baseline measurement even in the energy leakage feature.
Figure 7.20 shows the baseline-free energy leakage from peripheral to
central wire computed by transmitting with PZT 3 in the free portion of
the strand and detecting at the strand’s bottom end. This result refers to
the low frequency range of 300 kHz – 700 kHz.
The elimination of the baseline is achieved by taking the ratio of
the RMS detected by PICO (C) and that detected by PICO (P). It can
be seen that the interwire leakage increases with increasing load level as a
result of the increasing interwire contact and leakage in the anchorage.
This intuitive result, which is opposite to what found in Test 1 (Figure
7.17 and Figure 7.18), only arises once PICO (C) is normalized by PICO
(P). The behavior is consistent between load and unloading ramps. Also,
the trend is such that individual load levels can be well discriminated, in
addition to allowing a clear distinction between loaded and unloaded
strand. Maximum sensitivity to load levels is observed between 400 kHz
and 500 kHz.

269
RATIO PICO(C)/PICO(P) –LOAD RAMP
2nd cycle 3rd cycle

2.5 2.5
2 2
start0%
Ratio C/P (RMS)

Ratio C/P (RMS)
up20%
1.5 1.5
up40%
up60%
1 1
up80%
up100%
0.5 0.5
0 0
300 400 500 600 700 300 400 500 600 700
Freq [kHz] Freq [kHz]
RATIO PICO(C)/PICO(P) –UNLOAD RAMP
2nd cycle 3rd cycle

2.5 2.5
2 2
up100%
Ratio C/P (RMS)
Ratio C/P (RMS)
dow n80%
1.5 1.5
dow n60%
dow n40%
1 1
dow n20%
end0%
0.5 0.5
0 0
300 400 500 600 700 300 400 500 600 700
Freq [kHz] Freq [kHz]
Figure 7.21: Test 3 – Peripheral wire - central wire energy leakage (low
frequencies). Same feature as in Figure 7.20 for two additional load-unload
cycles.
Because of the baseline-free character and large sensitivity to load level,
the normalized energy leakage feature [RMS PICO(C)/RMS PICO(P)]
under Test 3 configuration was found to be the optimum one for prestress
level monitoring.
Figure 7.21 shows the same energy leakage features measured for
two additional load-unload cycles. Comparing these results to Figure 7.20,

270
it can be seen that the repeatability of the measurements was very
satisfactory. A fourth cycle was performed after replacing the PICO
sensors. A smaller amplification (34dB instead of 40dB) was provided to
the recorded signals to avoid the saturation experienced. The results
obtained are represented in Figure 7.22. It can be readily observed that
the RMS ratio was still robust in discriminating between different load
levels but the feature value had a smaller range of variation. This is due
to the changed coupling conditions between PICO sensors and strand end,
as well as the different frequency response of the PICO sensors. As a
consequence, although the RMS ratio is, potentially, a base-line free
feature, particular attention has to be taken in the experimental set-up.
For example, bonding conditions of the PICO sensors and misalignment
could affect the RMS ratio.
It should be noted that results illustrated in Figure 7.22 are similar
to the SAFE numerical results shown in Figure 7.14(c). Hence the main
source of the observed change in the ultrasonic strength of transmission is
the leakage into the anchorage wedges.
The frequency shift feature, which also does not require a baseline,
is better discussed in light of Figure 7.23 which shows the RMS of the
PICO (C) under PZT 3 excitation. Two distinct peaks of maximum
ultrasound transmission are seen in the 50 kHz – 130 kHz range and the
130 kHz – 250 kHz range, respectively. The two peaks have opposite
trends with increasing load, with the first one shifting towards higher
frequencies and the second one shifting towards lower frequencies. Hence
the need for examining the two frequency peaks independently.

271
The peak frequency shift relative to the unloaded peak frequency is
shown in Figure 7.24 for the low-frequency range of 50kHz–130kHz. Three
distinct cycles are shown to demonstrate the repeatability of this feature.
All plots show the expected increase in peak frequency with increasing
load level, which is consistent with vibrating cord theory and previous
Test 1 results (Figure 7.19). The shift is as high as 50% between fully
loaded and unloaded strand. It is also possible to discriminate among
different load levels. Although the trend is consistent between load and
unload ramps, differences exist in absolute frequency values in the two
cases. These results demonstrate that frequency shift in the 50kHz –
130kHz range under Test 3 configuration is another promising feature for
prestress level monitoring.
Figure 7.25 shows the relative frequency shift for the high-frequency
range of 130kHz–250kHz in the same load-unload cycle. It can be seen
that the frequency decreases with increasing load. This result is due to the
fact that higher frequencies capture the vibrational behavior of the
individual wires rather than that of the entire strand. The difference
between load and unload ramps also confirms the sensitivity to residual
inter-wire stresses. The high-frequency range of Figure 7.25 does not offer
the same load discriminating capability as the low-frequency range of
Figure 7.24. However, it still allows to distinguish between a loaded and a
fully-unloaded strand.
272
RATIO PICO(C)/PICO(P) – LOAD RAMP – 4th cycle
1
0.9
0.8
start0%
Ratio C/P (RMS)
0.7
up20%
0.6
up40%
0.5
up60%
0.4
up80%
0.3
up100%
0.2
0.1
0
300 400 500 600 700
Freq [kHz]
RATIO PICO(C)/PICO(P) – UNLOAD RAMP – 4th cycle
1
0.9
0.8
up100%
Ratio C/P (RMS)
0.7
down80%
0.6
down60%
0.5
down40%
0.4
down20%
0.3
end0%
0.2
0.1
0
300 400 500 600 700
Freq [kHz]
function of prestress. Test 3: PZT 3 transmitting, PICOs (C) and (P)
273
PICO (C) – LOAD RAMP
0.3
0.25
start0%
0.2
up20%
up40%
RMS
0.15
up60%
up80%
0.1
up100%
0.05
0
50 100 150 200 250 300
Freq [kHz]
PICO (C) – UNLOAD RAMP
0.25
0.2
up100%
dow n80%
0.15
dow n60%
RMS
dow n40%
0.1
dow n20%
end0%
0.05
0
50 100 150 200 250 300
Freq [kHz]
function of prestress. Test 3: PZT 3 transmitting and PICO (C) receiving.
Load and unload ramps. 100% load = 70% U.T.S. Low frequency range
(50 kHz – 300 kHz).
274
PICO(C) – 1st cycle

0.7
0.6
0.5
(fload-f0)/f0
0.4 Test3a Up
0.3 Test3a Down
0.2
0.1
0.0
0 4 8 12 16 20 24 28 32 36 40
Load [kips]
PICO(C) – 2nd cycle

0.7
0.6
0.5
(fload-f0)/f0
0.4 Test3b Up
0.3 Test3b Down
0.2
0.1
0.0
0 4 8 12 16 20 24 28 32 36 40
Load [kips]
PICO(C) – 3rd cycle

0.6
0.5
0.4
(fload-f0)/f0
Test3c Up
0.3
Test3c Down
0.2
0.1
0.0
0 4 8 12 16 20 24 28 32 36 40
Load [kips]

Test 3: PZT 3 transmitting, PICO (C) receiving. Load and unload ramps
for three cycles. Max load 41 kips = 70% U.T.S. Frequency range 50 kHz
– 130 kHz.
275
PICO(C) – 1st cycle

0.00
0 4 8 12 16 20 24 28 32 36 40
-0.05
-0.10
(fload-f0)/f0
-0.15
Test3a Up
-0.20
Test3a Down
-0.25
-0.30
-0.35
-0.40
Load [kips]
PICO(C) – 2nd cycle

0.00
0 4 8 12 16 20 24 28 32 36 40
-0.05
-0.10
(fload-f0)/f0
-0.15
Test3b Up
-0.20
Test3b Down
-0.25
-0.30
-0.35
-0.40
Load [kips]

Test 3: PZT 3 transmitting, PICO (C) receiving. Load and unload ramps
for two cycles. Max load 41 kips = 70% U.T.S. Frequency range (130 kHz
– 250 kHz).
276
1854
d1=203 d 254
Load Load
MsS MsS Notch defect

transmitter receiver
Figure 7.26: Experimental setup for defect detection in a strand using
reflections of guided waves excited and detected by magnetostrictive
transducers (dimensions in mm).
7.4 Defect detection in strands
7.4.1 Experimental setup and procedure
In addition to load monitoring, the technique of guided wave was
also employed for defect (notch-like) detection in the strands. Defect
detection results will be presented for the grade 270, seven-wire twisted
strand with a diameter of 15.24 mm (0.6 in). A notch was machined,
perpendicular to the strand axis, in one of the six peripheral wires by
saw-cutting with depths increasing by 0.5-mm steps to a maximum depth
of 3 mm (Figure 7.26). A final cut resulted in the complete fracture of
the helical wire (broken wire, b.w.), which was the largest defect
examined. The smallest notch depth of 0.5 mm corresponded to a 0.7%
reduction in the strand’s cross-sectional area. The largest notch depth of
broken wire corresponded to a 15.6% reduction in the strand’s cross-
sectional area.
277
D1 D2
D3 D4
defect reflection
D5 D6
35 95 155 215 275 35 95 155 215 275

Time (Psec) Time (Psec)
Figure 7.27: Signals reconstructed after pruning the DWT coefficients at
the first six decomposition levels.
The strand was subjected to a 120 kN tensile load, corresponding to
45% of the material’s Ultimate Tensile Strength (U.T.S.). Magnetostrictive
transducers resonant at 320 kHz, were used to excite and detect
longitudinal guided waves in the strand (Figure 7.26). This frequency was
chosen since it is known to propagate with little losses in loaded, free
strands (Rizzo and Lanza di Scalea, 2004). The distance between the
transmitting and the receiving transducers, d1 in Figure 7.26, was fixed at
203mm (8”) in all tests. By sliding the transmitter/receiver pair along the
strand, tests were conducted at the five different notch-receiver distances,
d in Figure 7.26, of 203mm (8”), 406mm (16”), 812mm (32”), 1016mm
(40”), and 1118mm (44”). The latter was the largest distance allowed by
278
the rigid frame of the hydraulic loading. Five-cycle tonebursts centered at
320 kHz, modulated with a triangular window, were used as generation
signals. Signals were acquired at a sampling rate equal to 33 MHz and
stored after different number of digital averages, namely 500, 50, 10, 5, 2
and 1 (single generation).
Two time windows were selected for the direct signal and the defect
reflection measured by the receiver. The gated waveforms were then
processed through the Discrete Wavelet Transform (DWT) using the
Daubechies of order 40 (db40) mother wavelet. For a 33 MHz sampling
frequency, the 320 kHz frequency of interest was contained in the sixth
level of DWT decomposition, according to
fj = ∆ × F / 2j (7.22)
relating the reconstructed frequency fj at level j to the center frequency F of
the mother wavelet, the scale 2j, and the signal sampling frequency ∆.
Thus the sixth level was the only one considered in the further analysis
(pruning). Representative results of the pruning process are shown in
Figure 7.27, presenting the signals reconstructed from the first six DWT
detail decomposition levels (D1, D2,…, D6). The original signal was taken
without any averages. The D6 reconstruction correctly identifies (at around
140 µsec) the reflection from a 2.5mm-deep notch in the helical wire. Since
levels 1 to 5 will merely reconstruct noise, they were eliminated in the
DWT analysis process.
Subsequently to pruning, the sixth decomposition level was subjected
to the thresholding process. The threshold chosen to select the relevant
wavelet coefficients is an important variable that affects the sensitivity of

279
the defect sizing. An optimum threshold combination for the direct signal
and the defect reflection was searched based on obtaining the largest
sensitivity to defect size through a variance-based reflection coefficient. It
was found that the larger sensitivities were obtained when setting more
severe thresholds on the defect-reflected signals, with little effect of the
thresholds imposed on the direct signal. Based on the findings of Rizzo
and Lanza di Scalea (2006), optimum thresholds were fixed at 20% of the
maximum wavelet coefficient amplitude for the direct signal, and at 70%
of the same quantity for the defect reflection.
A “reflection” Damage Index vector (D.I.) was constructed from the
ratios between certain features of the reflected signal, Freflection, and the
same features of the direct signal, Fdirect :

⎡ Freflection, i ⎤
D.I. = ⎢⎢ ⎥
⎥ (7.23)
F
⎢⎣ direct, i ⎥⎦
After parametric studies, the following four features were used to compute
a four-dimensional D.I.: variance, root mean square, peak amplitude and
peak-to-peak amplitude of the thresholded wavelet coefficients at level 6.
All D.I. components showed a quite linear dependence in a semi-
logarithmic scale on the notch depth, and a relatively negligible
dependence on the defect position for notches between 1.5mm and 3mm in
depth. The experimental data for two of these components are shown in
Figure 7.28. The results for very small notches, below 1 mm in depth,
were less stable against varying distances due to the poorer SNRs of the
defect reflections. The results for the broken wire case (5mm-deep notch)
also showed an increased dependence on the notch-receiver distance, with
D.I. components generally increasing for defects located further away from
280
the receiver. This trend is opposite to what would be expected considering
wave attenuation effects, and its origin is probably associated with the
interference of multiple propagating modes that is distance dependent. It
was also found that the D.I. component based on the variance of the
wavelet coefficient vector (Figure 7.28) had the largest sensitivity to notch
depth compared to all other components.
7.4.2 Statistical defect classification
A multivariate statistical analysis was performed to discriminate the
defect indications from random noise which may be present in the
measurements. The Mahalanobis Squared Distance (MSD), Dζ, was used as
the discordancy test (Worden et al., 2000):
Dζ = (x ζ − x ) K−1 (x ζ − x )
T
(7.24)
where xζ is the potential outlier vector, x is the mean vector of the
baseline, K is the covariance matrix of the baseline and T represents a
transpose matrix. In the present study, since the potential outliers were
always known “a priori”, Dζ was calculated exclusively without
contaminating the statistics of the baseline data.
The baseline distribution was obtained from the ultrasonic signals
stored after averaging over ten acquisitions and corrupted by two different
levels of white Gaussian noise. The noise signals were created by the
MATLAB randn function. The random noise increased the sample
population and simulated possible variations in SNR of the measurements
that can be originated, in practice, by a number of factors including
changing sensor/structure ultrasonic transduction efficiency, and changing

281
environmental temperature affecting ultrasonic damping losses. The randn
function generates arrays of random numbers whose elements are normally
distributed with zero mean and standard deviation equal to 1. The
function was pre-multiplied by a factor that determines the noise level.
Factors equal to 0.01 and 0.1 were considered as “low noise” and “high
noise”, respectively. For each noise level, 300 baseline samples were
created.
The same approach was taken to generate a large number of data
for the damaged conditions. Six of the seven total notch sizes discussed in
the previous section were considered. The ten-average signals acquired for
each of the six defects were corrupted by the low noise level and the high
noise level, generating a total of 300 samples for each damage size. These
samples represented the testing data of the algorithm. A total 2100
samples data were thus collected for each noise level. The added noise can
be quantified in terms of SNR by the following expression:
⎡ N 2 ⎤
⎢ ∑ si / N ⎥
⎢ ⎥
SNR[dB] = 10Log ⎢ iN=1 ⎥ (7.25)
⎢ ⎥
⎢ ∑ ui / N ⎥
2
⎢⎣ i =1 ⎥⎦
where si and ui are the amplitudes of the ultrasonic signal and of the
noise signal, respectively, and N is the number of points. The SNR
between the direct signal and the two 0.01 and 0.1 noise levels was about
43 dB and 23 dB, respectively. The SNR between the reflection from the
3 mm-deep notch and the two 0.01 and 0.1 noise levels was about 32 dB
and 12 dB, respectively. Clearly, the latter two values decreased with
decreasing notch depth.

282
7.4.3 Defect detection results – “low” noise
The MSD computed from the four-dimensional D.I. of all samples,
including the baseline data and the damage data, calculated for the low
noise level of 0.01 are summarized in Figure 7.29(a). The mean vector and
the covariance matrix were determined from the 300 D.I. vectors
associated with the undamaged condition of the strand. The horizontal line
in this figure represents the 99.73% confidence threshold value of 21.579.
Eight baseline samples are outliers, thus false positive indications. Clear
steps can be seen for increasing levels of damage. All damaged conditions
were properly classified as outliers, thus there were no false negative
indications. The MSD values showed good discrimination between all defect
sizes, including the smallest notch depths, confirming that it is
advantageous to combine multiple GUW features to provide a large
sensitivity to the defects. Nevertheless, compared to previous multivariate
outlier analyses in structural monitoring applications, the dimension of the
D.I. was still kept at a very low value by selecting only four features of
the DWT-processed wave signals.
7.4.4 Defect detection results – “high” noise
The MSD results of the D.I. corrupted with the high noise level of
0.1 are shown in Figure 7.29(b). The 99.73% confidence threshold was now
computed as 18.137. Compared to the low noise results of Figure 7.29(a),
it is clear that the heavier noise corruption compromises the ability to
detect the notch depths below 2.0mm, corresponding to a 5% reduction in
strand’s cross-sectional area. The ratios of correctly classified outliers below

283
5% area reduction were only 12/300, 7/300 and 1/300 for notch depths of
0.5mm, 1.0mm and 1.5mm, respectively. Above the 5% area reduction, the
sensitivity to defect detection was also degraded with the increasing noise
level; for example, the MSD values for the 2mm notch depth in Figure
7.29(b) are four orders of magnitude smaller than the corresponding values
in Figure 7.29(a). The reduced number of false positive indications (three
against eight) is the only improvement over the low noise level.
Table 7.3 summarizes the number of outliers detected in the
multivariate analyses for both levels of noise considered; the outliers are
false positive indications for the baseline data (Damage Size 0) and,
instead, correct indications of anomalies for the defect data.
7.5 Discussion and Conclusions
This chapter presents a technique based on ultrasonic guided waves
to monitor the structural condition of multi-wire strands used in
prestressed concrete structures and cable-stayed or suspension bridges. The
ultrasonic technique has the potential for providing both stress monitoring
and defect detection in the strands.
Table 7.3: Results of defect detection from outlier analysis: number of

outliers n/300 for the various damage sizes and two levels of noise.
Damage Size (notch depth – mm)
Noise level
0 0.5 1 1.5 2 3 5.0 (b.w.)
0.01 8/300 300/300 300/300 300/300 300/300 300/300 300/300
0.1 3/300 12/300 7/300 1/300 300/300 300/300 300/300
284
D.I. (Variance of wavelet

0.1
coefficients)
0.01
0.001 203 mm 406 mm

812 mm 1016 mm
0.0001 1118 mm
0.00001
0 1 2 3 4 5
Notch depth (mm)
t
1
D.I. (RMS of wavelet
0.1
coefficients)
0.01
0.001 203 mm 406 mm

812 mm 1016 mm
0.0001 1118 mm
0.00001
0 1 2 3 4 5
Notch depth (mm)
f
Figure 7.28: Components of the Damage Index vector measured from the
variance and from the root-mean-square of the thresholded wavelet
coefficients at the sixth decomposition level.
The semi-analytical finite element (SAFE) method was used to
model (a) the multimode and dispersive behavior in a free, seven-wire
strand (a pre-twisted waveguide), (b) the ultrasonic leakage in the loaded
strand with anchorages and (c) the case of a rod embedded in grout and
concrete (an axis-symmetric multilayer waveguide). The last case was
treated at the end of Chapter 3 and not repeated here. It was studied to
285
investigate the ultrasonic leakage from a rod into the surrounding media
and to identify mode-frequency combinations which propagate within the
rod with minimum attenuation losses for long-range monitoring.
For the unloaded free strand, it was shown that each wire behaves
as an independent waveguide and the wave attenuation is small compared
to the loaded case. Furthermore, the waves excited in a single wire remain
predominantly confined within that wire.
1.0E+11
1.0E+10 (a)
1.0E+09
Mahalanobis distance
1.0E+08
1.0E+07
0 mm 0.5 mm 1.0 mm 1.5 mm 2.0 mm 3.0 mm b.w
1.0E+06
1.0E+05
1.0E+04
1.0E+03
1.0E+02
1.0E+01
1.0E+00
1.0E-01
0 300 600 900 1200 1500 1800 2100
Sample number
1.0E+11
1.0E+10 (b)
1.0E+09
Mahalanobis distance
0 mm 0.5 mm 1.0 mm 1.5 mm 2.0 mm 3.0 mm b.w

1.0E+08
1.0E+07
1.0E+06
1.0E+05
1.0E+04
1.0E+03
1.0E+02
1.0E+01
1.0E+00
1.0E-01
0 300 600 900 1200 1500 1800 2100
Sample number
Figure 7.29: Mahalanobis squared distance for the baseline (undamaged)

and damaged strand data corrupted with the low-level noise (a) and the
high-level noise (b).
286
Finally, the leakage induced by the anchorages that transfer the
load to the strand was numerically predicted. The SAFE was successfully
used to model the strand-wedge interaction, by coupling two waveguides of
different cross section, while neglecting waves reflected I the strand-wedge
interface.
For stress monitoring purposes, piezoelectric transducers probing the
individual wires near the strand’s end are proposed. The best transducer
lay-out was found where ultrasound excitation is performed on a peripheral
wire and ultrasound detection is performed on the central wire and on the
peripheral wire at the strand’s end. Two features proved suitable for stress
monitoring. The first feature, the inter-wire leakage between the peripheral
and the central wire, does not require a baseline once normalized, and in
the 400 kHz – 500 kHz range it appears effective not only to detect a
complete loss of stress, but also to quantify the level of applied stress.
The second feature was the shift in peak frequency of the peripheral-to-
central wire transmissibility spectrum. The frequency shift proved most
sensitive in the 50 kHz – 130 kHz range; it does not require a baseline
and it can detect both a complete loss of stress and the level of applied
stress.
For defect detection purposes, an array of magnetostrictive
transducers probing the strand as a whole is proposed. Proof-of-principle
results for the detection of notch-like defects were shown based on a
reflection Damage Index vector containing four components. A multivariate
statistical analysis based on an Outlier Analysis showed the ability to
distinguish the notches, which were located as far away as 1,100 mm from
287
the transducers, from simulated digital noise. The algorithm was able to
properly flag notches as small as 0.5 mm (0.7% strand’s area reduction)
for SNRs on the order of 32 dB. For higher noise level, corresponding to
SNRs on the order of 12 dB, the properly flagged notches were as small
as 2 mm (5% strand’s area reduction).
Most of the experimental results presented were obtained in the case
of free strands. For embedded strands, the sensitive frequency ranges may
change as a result of the different waveguide problem. However, the
general trends of ultrasonic features as a function of defects and applied
stress are expected to be similar to those found for the free waveguide
case.
In an actual post-tensioned structure, the stress monitoring
technique simply requires access to a peripheral wire of the strand (at a
location ~1 m into the embedded portion) to install the piezoelectric
transmitter, and access to the strand’s free end to install the piezoelectric
sensors. If these access points are granted, existing structures can be
instrumented. For new structures, the strands would clearly be
instrumented before prestressing. Because of their geometry, the installation
of the magnetostrictive transducers for defect detection requires access to
the entire strand surface; hence the defect detection method is more
applicable to new structures where the strands are instrumented before
installation.
8 CONCLUSIONS AND RECOMMENDATIONS FOR
FUTURE STUDIES
8.1 Conclusions
In the present dissertation, three distinct structural components that
will benefit from Structural Health Monitoring systems have been shown.
The aspect in common between the three cases is the use of guided
ultrasonic waves (GUWs). It was stated that guided waves have many
advantages that make them ideal candidates for the inspection of
structural components. GUWs provide long range inspection capabilities,
carry the energy in the entire cross-section and consequently, can be
sensitive to defects located anywhere in the waveguide.
It is well known, however, that GUWs have a fairly complex
behavior that depends on the geometry of the structural component in
which they travel. As a consequence, the potential structural health
monitoring strategy that uses GUWs, can be successfully applied only after
a careful theoretical study of the dispersion properties (speeds, attenuation,
mode shapes) of the propagating waves.
The SAFE method developed during this study is a tool quite
general that has allowed the prediction of guided wave properties in flat
composite systems, cylindrical layered waveguides, railroad tracks and
pretwisted structural components. The approach accounts for material
damping and consequently, attenuation of guided modes can be computed.
The knowledge of the dispersive wave properties, and in particular the
attenuation, is relevant for NDE/SHM testing for identifying propagating
288
289
modes, locating defects, as well as exciting low-loss mode-frequency
combinations.
For example, in composite structures the inherently high material
attenuation might impose a large number of transducers to cover large
areas. Since a structural health monitoring system has to be economically
sustainable, the knowledge of low-loss mode combinations is crucial to
minimize the number of sensors required.
The focus of Chapter 4 was the propagation of ultrasonic guided
waves in composite adhesively-bonded joints. The specific component
investigated was representative of a skin-to-spar joint of UAVs. Two lay-
ups for the composite skin were investigated experimentally and
numerically and two types of bond defects were considered, namely a
poorly-cured bond and a disbond where the shear stiffness was nominally
lost. In this study the SAFE method was used to predict modal solutions,
stress and power flow profiles for the joints modeled as composite
waveguides.
Experimentally the root mean square of the signals captured by
PZT transducers in the across the bond configuration was used to
compute the strength of transmission across the bond. Numerically the
change in the Power flow of the Poynting vector was monitored as a
function of the bond conditions.
Both experimental and numerical results confirmed that the strength
of transmission of selected guided waves across the damaged bond
increases in the 100kHz - 300kHz range. The best sensitivity to the defects
was measured at around 200 kHz corresponding to the mode coupling
point of the S0 and A1 carrier modes. When mode coupling occurs, in fact,
290
a substantial amount of energy flow occurs in the thickness direction of
the plate. In Chapter 3 it was shown that the wave attenuation is directly
connected to the power flow in the thickness direction of multilayered
coated pipes. Such flow is clearly affected by a defected bond. It should
be noted that mode coupling phenomena can be captured only when
material damping is taken into account as in the SAFE formulation
proposed in this dissertation.
In Chapter 5 the use of a commercial finite element package,
ABAQUS EXPLICIT, was demonstrated to model guided ultrasonic waves
propagating at frequencies of tens of kilohertz. The waves, generated by
impulsive excitation, were examined in a broad frequency range with the
aid of joint time-frequency (wavelet) analysis. The case of Lamb waves
propagating in a 20mm-thick free plate was first examined with the
purpose of checking the validity of existing recommendations for the
spatial and temporal resolution of the finite element analysis as applied to
ABAQUS. The guidelines were then applied to modeling guided waves
propagating in 115lb A.R.E.M.A. rails in the context of rail defect
detection by long-range ultrasonic inspection. Modeshapes and dispersion
properties of the railroad track were computed in Chapter 3 by using
SAFE. Among the modes that exist in the low frequency range (f<50kHz),
the vertical bending mode was examined for its ease of generation and
detection in the field. This mode was generated by an impulse excitation
at the top of the rail head in the vertical direction. The defects examined
included four different sizes of transverse head flaws at three different
orientations, for a total of twelve cases. The model extracted same-mode
reflections (vertical incident and reflected) and mode-converted reflections

291
(vertical incident and lateral reflected). The study showed that appreciable
reflections from defects as small as 15% of the rail head are found in the
20kHz - 45kHz range.
Finite element modeling has shown to be an important tool to
predict the wave interactions with a variety of defects that would be
impractical to replicate experimentally. However the analysis of propagating
waves at frequencies higher than 100 kHz becomes unfeasible due to the
smaller finite element dimension required and to the huge amount of finite
elements necessary to accurately represent the wave motion. The SAFE
can be used to represent the response also at frequencies above 100kHz.
The approach, currently, is not used to consider defects but coupling
between finite element method and SAFE is possible and will be the
object of future studies. The need for modeling waves with frequency
higher than 100kHz becomes clear in the following paragraph.
In Chapter 6, an inspection system based on ultrasonic guided
waves excited and detected non-contact in the high frequency range
(100kHz-1MHz) was demonstrated for the detection of transverse head
cracks in rails. This system has been tested in the field twice and it
shows promise for actual use by railroad inspection industry. The main
advantages of the proposed technology include 1) an inherent sensitivity to
transverse defects and, more directly, to the reduction of the rail head
cross-sectional area, 2) an increased inspection speed and large inspection
ranges due to non-contact and signal processing techniques, and 3) an
increased defect detection reliability due to a dual detection scheme based
on both reflection and transmission measurements.

292
The current prototype uses a pulsed laser to generate surface-type
guided waves and an array of air-coupled sensors to detect the waves. The
air-coupled sensors were used to provide both reflection and transmission
information on a crack. It was observed that the reflection coefficient
increases with increasing crack depth, and, complementarily, the
transmission coefficient decreases with increasing crack depth. A Discrete
Wavelet Transform processing of the air-coupled signals was performed.
The wavelet processing increases robustness to noise and can be performed
in real-time. Successful detection of transverse cracks as small as 1mm in
depth was shown for lift-off distances of the detecting sensors as large as
76 mm (3”) from the top of the rail head, satisfying the recommended
clearance envelope for new rail inspection systems in the US claiming
“non-contact” performance.
In Chapter 7, a technique based on ultrasonic guided waves to
monitor the structural condition (applied load and presence of defects) of
multi-wire strands used in prestressed concrete structures and cable-stayed
or suspension bridges was presented.
The SAFE method was used to model (a) the multimode and
dispersive behavior in a free, seven-wire strand (a pre-twisted waveguide),
(b) the ultrasonic leakage in the loaded strand with anchorages and (c)
the case of a rod embedded in grout and concrete (an axis-symmetric
multilayer waveguide). The last case was studied to investigate the
ultrasonic leakage from a rod into the surrounding media and to identify
mode-frequency combinations which propagate within the rod with
minimum attenuation losses for long-range monitoring.

293
For the unloaded free strand, it was shown numerically and
experimentally that each wire behaves as an independent waveguide and
the wave attenuation is small compared to the loaded case. For loaded
strands, the leakage induced by the anchorages that transfer the load to
the strand, was numerically predicted. The SAFE was successfully used to
model the interaction strand-wedge, by coupling two waveguides of
different cross section.
For stress monitoring purposes, piezoelectric transducers probing the
individual wires near the strand’s end are proposed. The best transducer
lay-out was found where ultrasound excitation is performed on a peripheral
wire and ultrasound detection is performed on the central wire and on the
peripheral wire at the strand’s end. Two features proved suitable for stress
monitoring. The first feature, the inter-wire leakage between the peripheral
and the central wire, does not require a baseline once normalized, and in
the 400 kHz – 500 kHz range it appears effective not only to detect a
complete loss of stress, but also to quantify the level of applied stress.
The second feature was the shift in peak frequency of the peripheral-to-
central wire transmissibility spectrum. The frequency shift proved most
sensitive in the 50 kHz – 130 kHz range; it does not require a baseline
and it can detect both a complete loss of stress and the level of applied
stress.
For defect detection purposes, an array of magnetostrictive
transducers probing the strand as a whole was proposed. A multivariate
statistical analysis based on an Outlier Analysis showed the ability to
distinguish artificial notches, which were located as far away as 1,100 mm
from the transducers, from simulated digital noise.

294
The experimental results presented were obtained in the case of free
strands. For embedded strands, the sensitive frequency ranges may change
as a result of the different waveguide problem. However, the general
trends of ultrasonic features as a function of defects and applied stress are
expected to be similar to those found for the free waveguide case.
In an actual post-tensioned structure, the stress monitoring
technique simply requires access to a peripheral wire of the strand (at a
location ~1 m into the embedded portion) to install the piezoelectric
transmitter, and access to the strand’s free end to install the piezoelectric
sensors. If these access points are granted, existing structures can be
instrumented. For new structures, the strands would clearly be
instrumented before prestressing. Because of their geometry, the installation
of the magnetostrictive transducers for defect detection requires access to
the entire strand surface; hence the defect detection method is more
applicable to new structures where the strands are instrumented before

installation.
8.2 Recommendations for future studies
As discussed in the precedent section, the SAFE is currently used
without considering the presence of defects or waveguide cross-section
discontinuities. Although the response in a point of the waveguide can be
predicted, the reflection from flaws is not captured by the present model.
A possible development of the SAFE method is represented by the Global
Local method. This approach considers a complete finite element

295
discretization of the local region where a discontinuity is located while the
SAFE models the semi-infinite remaining portions of the waveguide.
Ongoing research at the NDE & SHM laboratory at UCSD has already
proved the ability of the SAFE-FE approach to predict numerically the
scattering (reflection and transmission) of guided waves in aluminum plates
with geometry discontinuities such as corrosion defects (Srivastava et al.,
2007).
Potential developments will be (a) modeling the wave scattering in

the wing skin to spar joint due to small disbonds; (b) predicting the wave
reflection from small defects in railroad tracks and (c) monitoring changing
loads in the strand by performing a complete finite element discretization
of the strand anchorage.
The current SAFE could become computationally more efficient by
performing a specific study on the eigenvalue problem. At the current
stage, standard Matlab subroutines are used to compute eigenvalues and
eigenvectors at each frequency. The problem could be easily divided into
N problems where N is the number of frequencies at which the user
wants to find dispersion solutions. Therefore SAFE could be adapted for
parallel computation. For waveguides discretized with a small number of
finite elements, the resulting general eigenvalue problem has small matrices
and is solved with negligible computational times. However, when a large
mesh is required, the eigenvalue problem becomes the major obstacle in
the SAFE solution. It should be noted that at each frequency, the
eigenvalues and eigenvectors are, in general, relatively similar to the
corresponding solutions computed at the previous frequency. As an
alternative to parallel computation, the use of eigenvalue subroutines able

296
to employ the solutions from the previous frequency step as initial guess
for eigenvalues and eigenvectors could further decrease the computational
effort. Implicitly restarted Arnoldi methods (Lehoucq et al., 1998) are
possible candidates for the calculation of few eigenvalues and eigenvectors
of large scale generalized eigenvalue problems.
Finally, Superposition of Partial Bulk Waves based formulations can
account for attenuation induced by wave energy leakage into a surrounding
semi-medium. Currently, SAFE frameworks cannot accurately capture this

source of energy dissipation. The implementation of absorbing or infinite
elements (Thompson, 2000; Bessel and Zienkiewicz, 1977; Astley, 2000),
generally present in commercial finite element packages, connected to the
outer boundaries of the structure would allow to estimate the wave
propagation. However, infinite elements are unable to model and accurately
capture the guided wave attenuation related to the leakage. The
combination of the Boundary Element Method and SAFE Method should
be considered as a possible alternative.

APPENDIX A
An ultrasonic wave is characterized by its speed and its frequency
content. For several decades the frequency content was extracted by the
Fast Fourier Transform (FFT), a well-known algorithm that transforms the
signal time domain into its frequency domain. If the waveform in the time
domain of a certain phenomenon is, for example, the result of three
sinusoids of different frequencies, the correspondent FFT shows three peaks
at those frequencies but no information are revealed on where those
frequencies are localized in time or, in other words, when those frequencies
are detected.
To remedy this lack of information, a joint time-frequency analysis
is necessary. The continuous wavelet transform (CWT) is a powerful
method for time-frequency analysis that in recent years has gained
increased attention in ultrasonic NDE. Recent applications cover the study
of wave propagation in beams (Kishimoto et al. 1995, Inoue et al. 1996,
Kim and Kim 2000), plates (Wooh and Veroy 2001, Jeong and Young-Su
2000), railroads (Lanza di Scalea and McNamara 2004a,b) and localization
of acoustic emission sources (Gaul et al. 2001).
It is known that sufficiently broad time windows are necessary to
characterize low-frequency components, while narrow windows are wide
enough to extract information regarding high frequency components. This
means that a constant time window does not guarantee an adequate
resolution in both high- and low-frequency component simultaneously. This
297
298
is established by the Heisenberg uncertainty principle that lower bounds
the product of the resolution in time (∆t) and in frequency (∆f) by:
∆t ⋅ ∆f ≥ 0.5 (A.1)
This principle in quantum mechanics reflects the uncertainty on the
position and the momentum of a free particle. Equation (A.1) implies that
the area of the Heisenberg box is greater or equal to 0.5.
The CWT has a multi-resolution capability deriving from a flexible
window that is broader in time for observing low frequencies and shorter
in time for observing high frequencies, as required by the Heisenberg
uncertainty principle (Kishimoto et al. 1995).
Multi-resolution is absent in the short-time Fourier transform,
another joint time-frequency method. When compared to the pseudo-
Wigner-Ville distribution, the WT does not generate spurious cross terms
that can be detrimental in the presence of multiple echoes (Jeon and Shin
1993). The wavelet transforms decompose the original signal by computing
its correlation with a short-duration wave, the mother wavelet, that is
flexible in time and in frequency.
The CWT of a function f(t) is defined as (Mallat 1999):

∞
1 ⎛ t − u ⎞⎟
Wf (u, s) = ∫ f (t) ⋅ ⋅ ψ * ⎜⎜ ⋅ dt
⎜⎝ s ⎠⎟⎟
(A.2)
−∞
s
where ψ * is the complex conjugate of a function, the mother wavelet
ψ(t) , with:
1 ⎛ t − u ⎞⎟
ψu, s (t) = ⋅ ψ ⎜⎜
⎜⎝ s ⎠⎟⎟
(A.3)
s
299
The mother wavelet can be viewed as a windowing function in time
and in frequency domains where u is called the translation parameter, and
s is the scaling parameter. The parameter u shifts the wavelet in time
and s controls the wavelet frequency bandwidth, hence the time-frequency
resolution of the analysis.
The mother wavelet function must satisfy the admissibility condition

2
ψ(ω)
∫ ω
dω < ∞ (A.4)

where ψ(ω) is the Fourier transform of ψ(t) . An analytic wavelet can be
created by modulating the frequency of a real and symmetric window g(t)
as follows
ψ(t) = g(t)* eiη t (A.5)
where η is the wavelet center frequency.
An analytic wavelet transform defines a local time-frequency energy
density Pwf, which measure the energy of f in the Heisenberg box
centered at u on the time axis and at s = η/s on the frequency axis.
The sides of the Heisenberg boxes are equal to ∆f/s and s·∆t, with ∆f
and ∆t, wavelet window in frequency and time, respectively.
Several mother wavelets have been proposed. Gabor Wavelet
Transform (GWT) is frequently used because it provides the best balance
between time and frequency resolution since it uses the smallest possible
Heisenberg uncertainty box (Mallat, 1999). The GWT modulates the
analyzing wavelet with a Gaussian window g(t) defined as:

300
η ⎛ − (η / G )2 t2 ⎞⎟
1 ⎜
g (t) = 4 ⋅ ⋅ exp ⎜⎜ ⎟⎟⎟
S
(A.6)
π GS ⎜⎝ 2 ⎠⎟
where GS is known as the Gabor shaping factor equal to the product of
the standard deviation σ of the Gaussian window and the wavelet center
frequency η.
The choice of η and GS affects the time-frequency resolution of the
analysis. The GWT with η = 2π and GS = π(2/ln2)1/2 = 5.336 has been
used in Chapter 5. These values are typical for the study of dispersive
wave propagation (Kishimoto et al. 1995, Inoue et al. 1996, Suzuki et al.
1996, Gaul et al. 2001, Lanza di Scalea and McNamara 2004a,b). With the
choice of η = 2π, the scaling parameter s, the frequency f and the
angular frequency ω are related by ω = η/s = 2π/s = 2πf and therefore
s=1/f.
The choice of GS influences the relative resolution achievable in
time and in frequency: decreasing the value of the Gabor shaping factor
increases the achievable time resolution but decreases the frequency
resolution. The choice of Gs = 5.336 was also corroborated by a pilot
study on the flexural rod vibrations conducted at UCSD (McNamara
2003).
The information from the joint time-frequency analysis is
extrapolated from the energy density Pwf, called scalogram, analytically

expressed by:
2
2 η
Pwf (u, s) = Wf (u, s) = Wf (u, ) (A.7)
ω
301
The scalogram represents the magnitude of the wavelet transformed signal,
i.e. the energy density spectrum and it shows the signal energy with
different frequencies ω = η/s at various times t = u. The scalogram may
be displayed in 3-D plots of time-frequency-amplitude. Because the
scalogram provides the time–frequency information of the energy
components of a function, it is possible to extract the dispersion curves in
term of the group/energy velocity and the frequency-dependent
attenuation.
An example of a CWT scalogram is given in Figure A.1 and Figure
A.2. Figure A.1(a) shows the strain waveform predicted by ABAQUS for a
800 mm long alluminum 1.5875 mm thick plate. The incoming wave is
an S0 mode excited by the imposed symmetric displacement field applied

on the left end side of the aluminum plate as shown in Figure A.1. In the
specific case a different mother wavelet was used. Figure A.1(b) shows the
Complex Morlet Wavelet scalogram. The scalogram amplitudes are plotted
versus frequency (y-axis) and time (x-axis).
Figure A.2(a) shows the strain waveform predicted by ABAQUS due
to the incoming A0 and A1 modes excited by an axisimmetric displacement
field. The arrival times of the two axisimmetric modes are captured by
the scalogram in Figure A.2(b).

302
-3 Zoomed Time Signal 1

x 10
4
(a)
3
1
Strain
-1
-2
-3
-4
0 100 200 300 400
Time (µsec)
(b)
Figure A.1: (a) Strain Waveform predicted in a 800 mm-long aluminum

plate showing the first arrival of the dispersive S0 wave. (b) Complex
Morlet Wavelet transform scalogram of time waveform identifying arrival
times and frequency components of the signals.
303
-11
x 10
4
(a)
3
2
Strain
-1
-2
0 50 100 150 200 250 300
Time (µsec)
A1
A0
(b)
Figure A.2: (a) Strain Waveform predicted in a 800 mm-long aluminum

plate showing the first arrival of the dispersive A0 and A1 modes. (b)
Complex Morlet Wavelet transform scalogram of time waveform identifying
arrival times and frequency components of the signals.
304
For high-speed applications, the Discrete Wavelet Transform (DWT)
implemented with parallel filter banks is a better choice compared to the
CWT because of the DWT’s computational efficiency. The efficiency
results from the existence of a fast orthogonal wavelet transform algorithm
based on a set of filter banks (Mallat 1989, 1999). The DWT may be
intuitively considered as a decomposition of a function (signal) following
hierarchical steps (levels) of different resolutions (Figure A.3). At the first
step the function is decomposed into wavelet coefficients; low-frequency
components (low-pass filtering) and high-frequency components (high-pass
filtering) of the function are retained. The signal is therefore decomposed
into separate frequency bands (scales). The filtering outputs are then
downsampled. The number of wavelet coefficients for each branch is thus
reduced by a factor of 2 such that the total number of points at a given
level is that of the original signal. Each level j corresponds to a dyadic
scale 2j at the resolution 2-j. According to the causality property, an
approximation at resolution 2-j contains all necessary information to
compute the next approximation at resolution 2-j-1 (scale 2j+1).
Furthermore, the decomposition means increasing the scale that
corresponds to zooming into the low-frequency portions of the spectrum.
Because of the downsampling, a signal of 2u points can be decomposed
into u levels, which will produce a total of 2u+1 sets of coefficients. The
final level u has 2 coefficients and each branch has only one coefficient.
The DWT of the discrete signal f(t) is computed at scales s = 2j. It is

possible to construct wavelets such that the dilated and the translated
family
305
1 ⎛t ⎞
ψj, n (t) = ⋅ ψ ⎜⎜ j − n⎟⎟⎟ (A.8)
2j ⎜⎝ 2 ⎠
is an orthonormal basis of L2(R), the space of the finite energy functions
g(t)
+∞
∫
2
g(t) dt < +∞ (A.9)
−∞
The filter bank tree used for the wavelet decomposition of the signals
presented in this paper can be seen in Figure A.3 where DWLF is the
discrete wavelet lowpass filter and DWHF is the discrete wavelet highpass
filter. The low-pass filter is given by
1 t
DWLF [n] = φ( ), φ(t − n) (A.10)
2 2
where the symbol indicates the inner product operator, φ(t) is the
mother scaling function associated with the mother wavelet ψ(t) and n is
the coefficient number of the lowpass and highpass filters. In a similar
fashion the highpass filter is given by
1 t
DWHF [n] = ψ( ), φ(t − n) (A.11)
2 2
Low- and high-pass filters are related by the formula:
DWHF[n] = (-1)1-n DWLF[1-n] (A.12)
At the first level, the output of filtering and downsampling steps is a set
of lowpass, cA1, and highpass, cD1, filter coefficients given by
cA1 = [f(n) ⊗ DWLF] (↓2) (A.13)

306
cD1 = [f(n) ⊗ DWHF] (↓2) (A.14)
where ⊗ is the convolution operator and (↓2) is the downsampling
operator. The decomposition then proceeds to the higher levels.
De-nosing and compression of the original signal can be achieved if

only a few wavelet coefficients representative of the signal are retained and
the remaining coefficients, related to noise, are discarded. This task is
known as pruning (Abbate et al. 1997). The process of reconstructing the
time signal from a set of wavelet coefficients is illustrated in Figure
A.3(b). The coefficients are upsampled to regain their original number of
points and then passed through a reconstruction lowpass filter, DWLF’,
and reconstruction highpass filter, DWHF’. The reconstruction filters are
closely related but not equal to those of the decomposition tree and are
given by (Strang and Nguyen 1996)
DWLF’[n] = DWHF[-n] (A.15)
DWHF’[n] = - DWLF[-n] (A.16)
Reconstruction by using the decomposition level k (scale 2k), for example,
is achieved by setting the wavelet coefficients from other scales equal to
zero:
Wj,n = 0 for j = 1, 2,…, k-1,

(A.17)
k+1,…, u
Finally, the linear combination of the reconstructions from various
decomposition levels results in the reconstruction of the original time signal
as indicated in Figure A.3(c).

307
f[n]
(a)
⊗ ⊗
DWLF DWHF
Decomposition Level 1
↓2 ↓2
cA1 cD1
⊗ ⊗
DWLF DWHF
↓2 ↓2
cA2 cD2
⊗ ⊗
DWLF DWHF
↓2 ↓2
cA3 cD3
cA1 ↑2 DWLF’
+ A1
zeros ↑2 DWHF’
(b)
↑2 DWLF’
zeros
+ D1
cD1 ↑2 DWHF’
f [n] = Ai +Di +Di-1 + … + D2 +D1

(c)
where i is the largest level of the DWT decomposition
Figure A.3: (a) Wavelet decomposition by filter bank tree; (b) signal
reconstruction from wavelet coefficients; (c) reconstruction of original
signal..
308
3
2 (a)
Amplitude (V)
1
0
-1
-2 defect reflection
-3
3
(b)
2
Amplitude (V)
1
0
-1
-2
-3
35 65 95 125 155 185 215 245 275
Time (µsec)
(c)
D1 D2
D3
D4
defect reflection
D5 D6
35 95 155 215 275 35 95 155 215 275
Time (µsec) Time (µsec)
Figure A.4: (a) Signal in seven-wire strand after 500 averages; (b) signal
with no averages; (c) reconstructed signal after pruning the DWT
coefficients at the first six decomposition levels.
309
Figure A.4 illustrates the pruning concept as applied to ultrasonic
stress waves propagating in strands. Figure A.4(a) shows a typical
ultrasonic signal detected in a seven-wire, 15.24mm (0.6in) - toneburst
centered at 320 kHz. The acquisition sampling frequency was 33 MHz.
The trace detected at around 140 microseconds is the reflection from a
2.5mm-deep indentation in one of the helical wires of the strand. A
single signal (no averages) is shown in Figure A.4(b) where the defect
reflection is completely buried in noise. The no-averaged signals
reconstructed from the first six DWT decomposition levels are indicated in
Figure A.4(c) as D1, D2,…,D6. The Daubechies wavelet of order 40 (db40)
was used as the mother wavelet for this particular decomposition.
Reconstruction D1 corresponds to what the time signal would look like if
only level 1 highpass filter coefficients, cD1, were used for the
reconstruction. D2 corresponds to what the reconstructed time signal
would look like if only level 2 highpass filter coefficients, cD2, were used
to reconstruct the signal, and so on. When choosing which filter levels
should be selected to reconstruct the time signal, the criterion used was to
select the filter level producing a synthetic time signal that most closely
resembled the actual signal being analyzed. The result of averaging the
raw ultrasonic measurements 500 times was the comparison signal. The
filter associated with D6 in Figure A.4(c) appears to yield the best

reconstruction given the close resemblance with the averaged result in
Figure A.4(a). Levels 1 to 5 will merely reconstruct noise, so they are
eliminated in the pruning process. In summary, the DWT reconstruction
D6 successfully de-noises the raw time signal to the point where it

310
eliminates the need for signal averaging. The appropriateness of level 6
can be confirmed by considering the formula
fj = ∆ × F / 2j (A.18)
relating the reconstructed frequency fj at level j to the center frequency
F of the mother wavelet, the scale 2j, and the signal sampling frequency
∆. For the results in Figure A.4, F = 0.671 rad for the db40 wavelet and
∆ = 33×106 Hz, yielding f6 = 346 kHz that is close to the frequency of
the generated toneburst.
A thresholding step can be used after the pruning process to further
increase the SNR (Abbate et al. 1997). In this case, a threshold is applied
to the magnitude of the coefficients that are retained. This step assumes
that the smaller coefficients represent noise, and can be safely omitted.
Hence the data compression outcome of the DWT processing.

REFERENCES
Aalami, B. (1973) “Waves in prismatic guides of arbitrary cross section,”

Journal of Applied Mechanics 40, 1067-1072.
Abaqus User’s Manual Version 6.6.
Abbate, A., Koay, J., Frankel, J., Schroeder, S.C. and Das, P., (1997)
“Signal Detection and Noise Suppression Using a Wavelet Transform Signal
Processor: Application to Ultrasonic Flaw Detection,” IEEE Transactions
on Ultrasonics, Ferroelectrics, and Frequency Control 44(1), 14-26.
Achenbach, J.D. (1975) “Wave propagation in elastic solids,” American

Elsevier Publishing Company, New York.
Alers, G. (1998) “Railroad Rail Flaw Detection System Based on

Electromagnetic Acoustic Transducers,” Report DOT/FRA/ORD-88/09,
U.S. Department of Transportation.
Alleyne D. and Cawley P.A. (1991) “Two-dimensional Fourier transform

method for the measurement of propagating multimode signals,” Journal of
the Acoustical Society of America 89(3), 1159-1168.
Ansari, F. (2005) “Fiber optic health monitoring of civil structures using

long gage and acoustic sensors,” Smart Materials and Structures 14, S1-S7.
Astley, R.J. (2000) “Infinite elements for wave problems: a review of

current formulations and an assessment of accuracy,” International Journal
for Numerical Methods in Engineering 49, 951-976.
Auld, B. (1990) “Acoustic fields and waves in solids” New York: John
Wiley and Sons.
Barshinger, J., Rose, J.L. and Avioli, M.J. Jr. (2002) “Guided wave
resonance tuning for pipe inspection,” Journal of Pressure Vessel
Technology 124, 303-310.
Barshinger, J. and Rose, J.L. (2004) “Guided wave propagation in an

elastic hollow cylinder coated with a viscoelastic material,” IEEE
Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 51(11),
1547-1556.
311
312
Bartoli, I., Lanza di Scalea, F., Fateh, M. and Viola, E. (2005) “Modeling
Guided Wave Propagation with Application to the Long-range Defect
Detection in Railroad Tracks,” NDT&E International 38(5), 325-334.
Bartoli, I., Marzani, A., Lanza di Scalea, F. and Viola, E. (2006)

“Modeling wave propagation in damped waveguides of arbitrary cross-
section,” Journal of Sound and Vibration 295, 685-707.
Bartoli, I., Marzani, A., Lanza di Scalea, F., Rizzo, P., Viola, E., Sorrivi
E. and Phillips, R. (2007) “SAFE modeling of waves for the structural
health monitoring of prestressing tendons,” Proceedings of SPIE’s 14th
Annual International Symposium on Smart Structures and Materials,
65320D.
Beard, M.D., Lowe, M.J.S. and Cawley, P. (2003) “Ultrasonic guided

waves for inspection of grouted tendons and bolts,” ASCE Journal of
Materials in Civil Engineering 15(3), 212-218.
Bernard, A., Deschamps, M. and Lowe, M.J.S. (1999) “Energy velocity and
group velocity for guided waves propagating within an absorbing or non-
absorbing plate in vacuum,” Review of Progress in Quantitative NDE 18,
183-190.
Bernard, A., Lowe, M.J.S. and Deschamps M. (2001) “Guided waves

energy velocity in absorbing and non-absorbing plates,” Journal of the
Acoustical Society of America 110, 186-196.
Bettess, P. and Zienkiewicz, O.C. (1977) “Diffraction and Refraction of

surface waves using finite and infinite elements,” International Journal for
Numerical Methods in Engineering 11, 1271-1290.
Birgersson, F., Finnveden, S. and Nilsson, C-M. (2005) “A spectral super

element for modelling of plate vibration. Part 1: general theory,” Journal
of Sound and Vibration 287, 297-314.
Bolotin, V.V. (1964) The dynamic stability of elastic systems. San

Francisco: Holden Day.
Bouchilloux, P., Lhermet, N., and Claeyssen, F. (1999) “Electromagnetic

Stress Sensor for Bridge Cables and Prestressed Concrete Structures,”
Journal of Intelligent Material Systems and Structures 10, 397-401.
313
Casas, J.R. (1994) “A combined method for measuring cable forces: the
cable-stayed Alamillo bridge, Spain,” Structural Engineering International
4(4), 235-240.
Casey, N.F. and Laura P.A.A. (1997) “A review of the acoustic-emission

monitoring of wire ropes,” Ocean Engineering 24, 935-947.
Castaings, M. and Hosten, B. (2003) “Guided waves propagating in

sandwich structures made of anisotropic, viscoelastic, composite materials,”
Journal of the Acoustical Society of America 113, 2622-2634.
Cawley P., Alleyne D. (1996) “The use of Lamb waves for the long range
inspection of large structures,” Ultrasonics 34(2), 287-290.
Cawley, P., Lowe, M., Alleyne, D., Pavlakovic, B. and Wilcox, P. (2003)
“Practical long range guided wave testing: applications to pipes and rail,”
Materials Evaluation 61(1), 66-74.
Chajes, M., Hunsperger, R., Liu, W., Li, J. and Kunz, E. (2003) “Void
detection in grouted post-tensioned bridges using time domain
reflectometry,” Proceedings of the Transportation Research Board, Vol.
3853.
Chang, Z. and Mal, A.K. (1995) “A global-local method for wave

propagation across a lap joint,” Numerical Methods in Structural
Mechanics 204, 1-11.
Chase S.B. (2001) “Smarter bridges, why and how?,” Smart Materials
Bulletin 2, 9-13.
Chen, H-L. and Wissawapaisal, K. (2002) “Application of Wigner-Ville

Transform to Evaluate Tensile Forces in Seven-wire Prestressing Strands,”
ASCE Journal of Engineering Mechanics 128, 1206-1214.
Cheng, A., Murray, T.W. and Achenbach, J.D. (2001) “Simulation of laser-
generated ultrasonic waves in layered plates,” Journal of the Acoustical
Society of America 110, 848-855.
Cheung, Y.K. (1968) “The finite strip method in the analysis of elastic
plates with two opposite simply supported ends,” Proceedings of the
Instituition of Civil Engineers 40, 1-7.
Chimenti, D.E. (1997) “Guided waves in plates and their use in materials
characterization,” Applied Mechanics Reviews 50, 247-284.
314
Coccia, S., Bartoli, I., Lanza di Scalea, F. and Rizzo, P. (2007) “Non-
contact rail defect detection: first and second field tests,” Technical Report
submitted to the Federal Railroad Administration GRANT No. DTFR53-
02-G-00011- Phase V.
Crawley E.F. and J. de Luis. (1987) “Use of piezoelectric actuators as

elements of intelligent structures,” AIAA Journal 25, 1373-1385.
Cullington, D.W., MacNeil, D., Paulson, P. and Elliott, J. (2001)

“Continuous acoustic monitoring of grouted post-tensioned concrete
bridges,” NDT&E International 34, 95-105.
Cunha, A., Caetano, E. and Delgado, R. (2001) “Dynamic Tests on Large

Cable-Stayed Bridge,” Journal of Bridge Engineering, January/February,
54-62.
Deschamps, M. and Hosten, B. (1992) “The effects of viscoelasticity on the

reflection and transmission of ultrasonic guided waves by orthotropic
plate,” Journal of the Acoustic Society of America 91, 2007-2015.
Disperse User’s Manual, (2003) “Imperial College London, London, United

Kingdom.
Dong, S.B. and Huang, K.H. (1985) “Edge vibrations in laminated

composite plates,” Journal of Applied Mechanics 52, 433-438.
Ervin, B.L., Bernhard J.T., Kuchma, D.A. and Reis, H. (2006)

“Estimation of general corrosion damage to steel reinforced mortar using
frequency sweeps of guided mechanical waves,” Insight-NDT and Condition
Monitoring 48(11), 682-692.
Federal Railroad Administration. (1992-2002) Safety Statistics Data. FRA

Office of Safety Analysis.
Finnveden, S. (2004) “Evaluation of modal density and group velocity by

a finite element method,” Journal of Sound and Vibration 273, 51-75.
Fricker, S. and Vogel, T. (2007) “Site installation and testing of a continuous acoustic
monitoring,” Construction and Building Materials 21(3), 501-510.
Gabor, D. (1946) “Theory of communication,” Journal of Institution of

Electrical Engineers 93(26), 429–457.
315
Galan, J.M., Abascal, R. (2002) “Numerical simulation of Lamb wave

scattering in semi-finite plates,” International Journal for Numerical
Methods in Engineering 53, 1145-1173.
Gaul, L., Hurlebaus, S. and Jacobs, L.J. (2001) “Localization Of A

"Synthetic" Acoustic Emission Source On The Surface Of A Fatigue
Specimen,” Research in Nondestructive Evaluation 13, 105-117.
Gavrić, L. (1994) “Finite element computation of dispersion properties of

thin-walled waveguides,” Journal of Sound and Vibration 173, 113-124.
Gavrić, L. (1995) “Computation of propagating waves in free rail using a

finite element technique,” Journal of Sound and Vibration 185, 531-543.
Ghosh, T., Kundu, T. and Karpur, P. (1998) “Efficient use of Lamb modes
for detecting defects in large plates,” Ultrasonics 36, 791–801.
Giurgiutiu, V., Bao, J. and Zhao, W. (2003) “Piezoelectric wafer active

sensor embedded ultrasonics in beams and plates,” Experimental mechanics
43, 428-44.
Giurgiutiu, V. and Zagrai, A. (2002) “Embedded self-sensing piezoelectric

active sensors for on-line structural identification,” ASME Journal of
Vibration and Acoustics 124, 116-125.
Giurgiutiu V. (2005) “Tuned Lamb wave excitation and detection with

piezoelectric wafer active sensors for structural health monitoring,” Journal
of Intelligent Material Systems and Structures 16, 291-305.
Guyott, C.C.H. and Cawley, P. (1988a) “The ultrasonic vibration

characteristics of adhesive joints,” Journal of the Acoustical Society of
America 83, 632-640.
Guyott, C.C.H. and Cawley, P. (1988b) “Evaluation of the cohesive

properties of adhesive joints using ultrasonic spectroscopy,” NDT&E
International 21, 233-240.
Han, X., Liu, G.R., Xi, Z.C. and Lam, K.Y. (2002) “Characteristics of
waves in a functionally graded cylinder,” International Journal for
Numerical Methods in Engineering 53, 653-676.
Hay, T.R., Wei, L. and Rose, J.L. (2003) “Rapid inspection of composite
skin-honeycomb core structures with ultrasonic guided waves,” Journal of
Composite Materials 37, 929-939.
316
Hayashi, T., Song, W.J. and Rose, J.L. (2003) “Guided wave dispersion
curves for a bar with an arbitrary cross-section, a rod and rail example,”
Ultrasonics 41, 175-183.
Heller, K., Jacobs, L.J. and Qu, J. (2000) “Characterization of adhesive

bond properties using Lamb waves,” NDT&E International 33, 555-563.
Hladky-Hennion, A.C. (1996a) “Finite element analysis of the propagation

of acoustic waves in waveguides,” Journal of Sound and Vibration 194,
119-136.
Hladky-Hennion, A.C., Langlet, P. and De Billy, M. (1996b) “Finite

element analysis of the propagation of acoustic waves along waveguides
immersed in water,” Journal of Sound and Vibration 200, 519-530.
Huang, K.H. and Dong S.B. (1984) “Propagating waves and edge
vibrations in anisotropic composite cylinders,” Journal of Sound and
Vibration 96, 363-379.
Ihn, J., Chang, F.K. and Speckmann, H. (2001) “Built-in diagnostics for
monitoring crack growth in aircraft structures,” Proceedings of the 4th
International Conference on Damage Assessment on Structures, Cardiff,
204-205, 299-308.
Inoue, H, Kishimoto, K. and Shibuya, T. (1996) “Experimental Wavelet

Analysis of Flexural Waves in Beams,” Experimental Mechanics 36, 212-
217.
Jeon, J. and Shin, Y.S., (1993) “Pseudo Wigner-Ville Distribution,

Computer Program and its Applications to Time-Frequency Domain
Problems”, Naval Postgraduate School Report NPS-ME-93-002.
Jeong, H. and Young-Su, J. (2000) “Fracture Source Location in Thin

Plates Using the Wavelet Transform of Dispersive Waves,” IEEE
Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 47, 612-
619.
Jones, R.M. (1975) Mechanics of composite materials. New York: McGraw-

Hill.
Karunasena, W., Shah, A.H. and Datta, S.K. (1991) “Wave propagation in
a multilayered cross-ply composite plate,’’ Journal of applied mechanics 58,
1028–1032.
317
Keilers, C.H. and Chang, F.K. (1995) “Identifying delaminations in

composite beams using built-in piezoelectrics: part 1 – Experiments and
analysis; part 2 – An identification method,” Journal of intelligent material
systems and structures 6, 649-672.
Kenderian, S., Cerniglia, D., Djordjevic, B.B., Garcia, G., Sun, J. and
Snell, M. (2003) “Rail Track Field Testing Using Laser/Air Hybrid
Ultrasonic Technique,” Materials Evaluation 61(10) 1129-1133.
Kenderian, S., Djordjevic, B.B. and Green, R.E. jr. (2002) “Laser Based
and Air Coupled Ultrasound as Noncontact and Remote Techniques for
Testing of Railroad Tracks,” Materials Evaluation 60, 65-70.
Kim, Y.Y. and Kim, E.H. (2000) “A New Damage Detection Method
Based on a Wavelet Transform,” Proceedings of SPIE – IMAC-XVIII, San
Antonio, TX, 4062, 1207-1212.
Kishimoto, K, Inoue, H., Hamada, M. and Shibuya, T. (1995) “Time

Frequency Analysis of Dispersive Waves by Means of Wavelet Transform,”
ASME Journal of Applied Mechanics 62, 841-846.
Kohl, T., Datta, S.K. and Shah A.H. (1992) “Axially symmetric pulse
propagation in semi-infinite hollow cylinders,” AIAA Journal 30, 1617-1624.
Krautkramer, J. and Krautkramer, H. (1990) “Ultrasonic testing of

materials,” 4th ed. Berlin: Springer-Verlag.
Kundu, T., Maji, A., Ghosh, T. and Maslov, K. (1998) “Detection of

kissing disbonds by Lamb waves,” Ultrasonics 35, 573-580.
Kundu, T. and Maslov, K. (1997) “Material interface inspection by Lamb

waves,” International Journal of Solids and Structures 34, 3885-3901.
Kwun, H. and Teller, C.M. (1994) “Detection of fractured wires in steel

cables using magnetostrictive sensors,” Materials Evaluation 52, 503-507.
Kwun, H. and Teller, C.M. (1995) “Nondestructive Evaluation of Steel

Cables and Ropes Using Magnetostrictively Induced Ultrasonic Waves and
Magnetostrictively Detected Acoustic Emissions,” U.S. Patent 5,456,113.
Lagasse, P.E. (1973) “Higher-order finite element analysis of topographic

guides supporting elastic surface waves,” Journal of the Acoustical Society
of America 53, 1116-1122.
318
Lakshmanan, K.A. and Pipes, D.J. (1997) “Modeling damage in composite

rotocraft flexbeams using wave mechanics,” Smart materials & structures 6,
383-392.
Lanza di Scalea, F., Bonomo, M. and Tuzzeo, D. (2001) “Ultrasonic

guided wave inspection of bonded lap joints: noncontact method and
photoelastic visualization,” Research in nondestructive evaluation 13, 153-
171.
Lanza di Scalea, F. and McNamara, J. (2003) “Ultrasonic NDE of railroad

tracks: air-coupled cross-sectional inspection and long-range inspection,”
Insight – NDT& Condition Monitoring 45, 394-401.
Lanza di Scalea F., McNamara J. (2004) “Measuring high-frequency wave

propagation in railroad tracks by joint time–frequency analysis,” Journal of
Sound and Vibration 273(3), 637-651.
Lanza di Scalea, F. (2000) “Advances in Non-contact Ultrasonic Inspection

of Railroad Tracks,” Experimental Techniques 24(5), 23-26.
Lanza di Scalea, F., Rizzo, P. and Marzani, A. (2004) “Propagation of

ultrasonic guided waves in lap-shear adhesive joints: case of incident a0
Lamb wave,” Journal of the Acoustical Society of America 115, 146-156.
Lanza di Scalea, F., Rizzo, P., and Seible, F. (2003) “Stress Measurement
and Defect Detection in Steel Strands by Guided Stress Waves,” ASCE
Journal of Materials in Civil Engineering 15(3), 219-227.
Lehoucq, R.B., Sorensen, D.C. and Yang, C. (1998) “ARPACK Users'

Guide: Solution of Large-Scale Eigenvalue Problems with Implicitly
Restarted Arnoldi Methods,” Philadelphia: SIAM Publications.
Lemistre, M.B. and Balageas, D.L. (2001) “Structural health monitoring

system based on diffracted Lamb wave analysis by multiresolution
processing,” Smart materials & structures 10, 504-511.
Light, G.M., Kwun, H., Kim, S-Y. and Spinks, R.L. (2003) “Health
monitoring for aircraft structures,” Materials evaluation 61, 844-847.
Liu, W., Hunsperger, R.G., Folliard, K., Chajes, M.J., Barot, J., Jhaveri,
D. and Kunz, E. (1998) “Detection and characterization of corrosion of
bridge cables by time domain reflectometry,” Nondestructive Evaluation of
Bridges and Highways III, SPIE Vol. 3587, 28-39.
319
Lowe, M.J.S. and Cawley, P. (1994) “Applicability of plate wave

techniques for the inspection of adhesive and diffusion bonded joints,”
Journal of Nondestructive Evaluation 13, 185-200.
Lowe, M.J.S. (1995) “Matrix techniques for modeling ultrasonic waves in

multilayered media,” IEEE Transactions on Ultrasonics, Ferroelectrics, and
Frequency Control 42, 525-542.
Lowe, M.J.S., Challis, R.E. and Chan, C.W. (2000) “The transmission of
Lamb waves across adhesively bonded lap joints,” Journal of the
Acoustical Society of America 107, 1333-1345.
Machida, S. and Durelli, A.J. (1973) “Response of a Strand to Axial and

Torsional Displacement,” Journal of Mechanical Engineering Science 15(4),
241- 251.
Matlab 7.04: (2000) User’s Guide, The Mathworks, Natick, MA.
Mal, A.K. (1988) “Guided waves in layered solids with interface zones,”
International Journal of Engineering Science 26, 873-881.
Mal, A.K., Chang, Z. and Guo, D. (1996) “Lap joint inspection using
plate waves,” Proceedings SPIE 2945, 128-137.
Mal, A.K., Xu, P. and Bar-Cohen, Y. (1990) “Leaky Lamb waves for the
ultrasonic nondestructive evaluation of adhesive bonds,” ASME The
Journal of Engineering Materials and Technology 112, 255-259.
Mallat, S.G. (1989) “A Theory for Multiresolution Signal Decompostion:

The Wavelet Representation,” IEEE Transactions on Pattern Analysis and
Machine Intelligence 11, 674-93.
Mallat, S.G. A Wavelet Tour of Signal Processing, 2nd ed.,

Academic Press, New York, 1999.
Matt, H., Bartoli, I. and Lanza di Scalea, F. (2005) “Ultrasonic guided

wave monitoring of composite wing skin-to-spar bonded joints in aerospace
structures,” Journal of the Acoustic Society of America, 118, 2240-2252.
McNamara, J. (2003) “Health monitoring of railroad tracks by elastic-wave

based non-destructive testing,” Ph.D. Dissertation, University of California,
San Diego.
320
McNamara, J. and Lanza di Scalea, F. (2004) “Improvements in Non-

contact Ultrasonic Testing of Rails by the Discrete Wavelet Transform,”
McNamara J., Lanza di Scalea F., Fateh M. (2004) “Support Vector

Machine based smart system for automatic defect classification in long
range ultrasonic rail inspection,” Insight 46(6), 331-337.
Moser F., Jacobs L.J., Qu J. (1999) “Modeling elastic wave propagation in

waveguides with the finite element method,” NDT&E International 32(4),
225-234.
Mukdadi, O.M. and Datta, S.K. (2003) “Transient ultrasonic guided waves
in layered plates with rectangular cross section,” Journal of Applied
Physics 93, 9360-9370.
Mukdadi, O.M., Desai, Y.M., Datta, S.K., Shah, A.H. and Niklasson, A.J.
(2002) “Elastic guided waves in a layered plate with rectangular cross
section,” Journal of the Acoustical Society of America 112, 1766-1779.
Nagy, P.B. and Adler, L. (1989) “Nondestructive evaluation of adhesive

joints by guided waves,” Journal of Applied Physics 66, 4658-4663.
Neau, G. Lowe, M.J.S. and Deschamps, M. (2002) “Propagation of lamb

waves in anisotropic and absorbing plates: theoretical derivation and
experiments,” Review of Progress in Quantitative NDE, 21, 1062-1069.
Neau, G. (2003) “Lamb waves in anisotropic viscoelastic plates. Study of

the wave fronts and attenuation,” Ph.D Thesis, L’Université Bordeaux I.
Nelson, R.B. Dong, S.B. and Kalra, R.D. (1971) “Vibrations and waves in
laminate orthotropic circular cylinders,” Journal of Sound and Vibration
18(3) 429-444.
Onipede, O. and Dong, S.B. (1996) “Propagating waves and end modes in
pretwisted beams,” Journal of Sound and Vibration 195, 313-330.
Parker, D. (1996) “Pacific bridge collapse throws up doubt on repair

method,” New Civil Engineering, Oct 17, 3-4.
Pavlakovic, B.N., Lowe, M.J.S., Alleyne, D.N. and Cawley, P. (1997)

“Disperse: a general purpose program for creating dispersion curves,”
Review of Progress in Quantitative NDE, 16, 185-192.
321
Pavlakovic, B.N., Lowe, M.J.S. and Cawley, P. (1999) “The inspection of

tendons in post-tensioned concrete using guided ultrasonic waves,” Insight-
NDT&Condition Monitoring 41, 446-452.
Pavlakovic, B.N., Lowe, M.J.S., Cawley, P. (2001) “High-frequency low-loss

ultrasonic modes in imbedded bars,” Journal of Applied Mechanics 68, 67-
75.
Pilarski, A. and Rose, J.L. (1992) “Lamb wave mode selection concepts for
interfacial weakness analysis,” Journal of Nondestructive Evaluation 11,
237-249.
Reddy, J.N. Mechanics of laminated composite plates. CRC Press, New

York, 1996.
Reddy, J.N. (1984) Energy and variational methods in applied mechanics.

New York: Wiley Interscience.
Reis, H., Ervin, B.L., Kuchma, D.A. and Bernhard, J.T. (2005)
“Estimation of corrosion damage in steel reinforced mortar using guided
waves,” ASME Journal of Pressure Vessel Technology 127, 255-261.
Rizzo, P. and Lanza di Scalea, F. (2001) “Acoustic emission monitoring of

carbon-fiber-reinforced-polymer bridge stay cables in large-scale testing,”
Experimental Mechanics 41(3), 282-290.
Rizzo, P. and Lanza di Scalea, F. (2004) “Load measurement and health

monitoring in cable stays via guided wave magnetostrictive ultrasonics,”
Rizzo, P. and Lanza di Scalea, F. (2005) “Ultrasonic inspection of multi-

wire steel strands with the aid of the wavelet transform,” Smart Materials
and Structures 14(4), 685-695.
Rizzo, P. and Lanza di Scalea, F. (2006) “Feature extraction for defect

detection in strands by guided ultrasonic waves,” Journal of Structural
Health Monitoring 5(3), 297-308.
Rokhlin, S.I. (1991) “Lamb wave interaction with lap-shear adhesive joints:
theory and experiment,” Journal of the Acoustical Society of America 89,
2758-2765.
322
Rokhlin, S.I. and Wang, Y.J. (1991) “Analysis of boundary conditions for
elastic wave interaction with an interface between two solids,” Journal of
the Acoustical Society of America 89, 503-515.
Rose, J.L. (1999) Ultrasonic waves in solid media. Cambridge University

Press, Cambridge.
Rose, J.L., (2002) “Guided Wave Ultrasonic Pipe Inspection—The Next

Generation,” 8th European Conference on Non-Destructive Testing,
Barcelona, Spain, June 17-21.
Rose, J.L., Rajana, K.M. and Hansch, M.K.T. (1995) “Ultrasonic guided
waves for NDE of adhesively bonded structures,” Journal of Adhesion 50,
71-82.
Rose, J.L., Zhu, W. and Zaidi, M. (1998) “Ultrasonic NDT of titanium

diffusion bonding with guided waves,” Materials evaluation 56, 535-539.
Sansalone, M., Jaeger, B.J., and Randall, W.P. (1996) “Detecting Voids in
Grouted Tendons of Post-Tensioned Concrete Structures Using Impact-
Echo Method,” ACI Structural Journal 93(4), 462-472.
Scheel, H. and Hillemeier, B. (2003) “Location of prestressing steel

fractures in concrete,” ASCE Journal of Materials in Civil Engineering
15(3), 228-234.
Seco, F., Martín, J.M., Jiménez, A.R., Pons, J.L., Calderón, L. and Ceres,
R., (2002) “PCDISP: a tool for the simulation of wave propagation in
cylindrical waveguides,” Proceedings of the Ninth International Congress on
Sound and Vibration, Orlando, Florida.
Seifried, R., Jacobs, L.J. and Qu, J. (2002) “Propagation of guided waves
in adhesive bonded components,” NDT&E International 35, 317-328.
Shull, J. (2002) “Nondestructive Evaluation: Theory, Techniques and

Applications,” New York: Marcel Dekker, Inc.
Simonetti, F. and Cawley, P. (2003) “A guided wave technique for the

characterization of highly attenuative viscoelastic materials,” Journal of the
Acoustical Society of America 114(1), 158-165.
Shorter, P.J. (2004) “Wave propagation and damping in linear viscoelastic

laminates,” Journal of the Acoustical Society of America 115, 1917-1925.
323
Schlaich, J., Kordina, K. and Engell, H.J. (1980) “Teileinsturz der

kongresshalle Berlin-schadensursachen zusammenfassendes gutachten (Partial
collapse of the Berlin congress hall-summary of the investigation into the
causes of the collapse),” Beton und Stahlbetonbau 75(12), 281-294.
Sohn, H., Park, G., Wait, J., Limback, N. and Farrar, C.R. (2003)
“Wavelet-based active sensing for delamination detection in composite
structures,” Smart Materials and Structures 13, 153-160.
Staszewski, W., Boller, C. and Tomlinson, G. (2004) “Health monitoring of

aerospace structures,” John Wiley & Sons.
Sun, Z., Rose, J.L. and Zaidi, M. (2002) “A phased array guided wave
approach to adhesive bonding structural integrity analysis,” Materials
evaluation 61, 941-946.
Suzuki, H., Kinjo, T., Hayashi, Y., Takemoto, M., Ono, K. and Hayashi,
Y. (1996) “Wavelet Transform of Acoustic Emission Signals,” Journal of
Acoustic Emission 14(2), 69-82.
Tabatabai, H., Mehrabi, A.B., and Yen W.P. (1998) “Bridge Stay Cable
Condition Assessment Using Vibration Measurement Techniques,”
Structural Materials Technology III, SPIE Vol. 3400, Medlock, R.D., and
Laffrey, D.C. Eds., San Antonio, TX, pp. 194-204.
Taweel, H., Dong, S.B. and Kazic, M. (2000) “Wave reflection from the
free end of a cylinder with an arbitraty cross-section,” International
Journal of Solids and Structures 37, 1701-1726.
Thompson, D.J. (1997) “Experimental analysis of wave propagation in

railway tracks,” Journal of Sound and Vibration 203(5), 867-888.
Thompson, D.J. (2003) “Wheel-rail noise generation, part III: rail

vibration,” Journal of Sound and Vibration 161, 421-446.
Thompson, L.L. and Huan, H. (2000) “Implementation of exact non-

reflecting boundary conditions in the finite element method for the time-
dependent wave equation,” Computer Methods in Applied Mechanics and
Engineering 187, 137-159.
Torvik P.J. (1967) “Reflection of wave trains in semi infinite plates,”

Journal of the Acoustical Society of America 41(2), 346-353.
324
Trivedi A., Krysl P., McNamara J., Lanza di Scalea F. (2004) “Adaptive
simulation of ultrasonic guided waves in railroad tracks using CHARMS,”
In: Proceedings of the 6th World Congress of Computational Mechanics,
Beijing, China, in press.
Volovoi, V.V., Hodges, D.H., Berdichevsky, V.L. and Sutyrin, V.G. (1998)
“Dynamic dispersion curves for non-homogeneous, anisotropic beams with
cross-section of arbitrary geometry,” Journal of Sound and Vibration 215,
1101-1120.
Victorov, I.A. (1967) “Rayleigh and Lamb waves” New York: Plenum
Press.
Wang, C.S. and Chang, F.K. (1999) “Built-in diagnostics for impact
damage identification in composite structures,” Proceedings of the 2nd
International Workshop on Structural Health Monitoring, Stanford, 612-621.
Wang, M., Lloyd, G.M., and Hovorka, O. (2001) “Development of a

Remote Coil Magnetoelastic Stress Sensor for Steel Cables,” Health
Monitoring and Management of Civil Infrastructure Systems, SPIE Vol.
4337, 122-128.
Washer, G., Green, R.E., and Pond, R.B. (2002) “Velocity Constants for
Ultrasonic Stress Measurements in Prestressing Tendons,” Research in
Nondestructive Evaluation 14(3), 81-94.
Watanabe, T., Ohtsu, M., and Nakayama, Y. (1999) “Impact-Echo NDT

for Grouting Performance in Post-Tensioning Tendon Duct,” CD-ROM
Proceedings of the 8th International Conference on Structural Faults and
Repair, M.C. Forsem, Ed., Engineering Technics Press.
Watson, S.C. and Stafford, D. (1988) “Cables in trouble,” Civil

Engineering 58, 38-41.
Weischedel, H. R., (1995) “Quantitative Nondestructive In-Service

Evaluation of Stay Cables of Cable-Stayed Bridges: Methods and Practical
Experience,” Nondestructive Evaluation of Aging Bridges and Highways,
Proceedings of SPIE, Vol. 2456, Oakland, pp. 122-128.
Wilcox, P., Evans, M., Diligent, O., Lowe, M.J.S. and Cawley, P. (2002)
“Dispersion and excitability of guided acoustic waves in isotropic beams
with arbitrary cross section,” Review of Progress in quantitative NDE, 21,
203-210.
325
Wilcox, P., Evans, M., Pavlakovic, B., Alleyne D., Vine, K., Cawley, P.
and Lowe, M.J.S. (2003) “Guided wave testing of rail,” Insight 45(6), 413-
420.
Williams, H.T. and Hulse, M.E. (1995) “From theory to experience with
inspection of post-tensioned bridges,” Proceedings of the Sixth International
Conference on Structural Faults and Repairs, M.C. Forde, Ed. 1,
Engineering Techniques Press, 199-202.
(Kear and Leeming, 1994)
Wooh, S.C. and Veroy, K. (2001) “Spectrotemporal Analysis of Guided-

wave Pulse-echo Signals: Part 2. Numerical and Experimental
Investigations,” Experimental Mechanics 41, 332-342.
Worden K., Manson G. and Fieller, N.R.J. (2000) “Damage detection using
outlier analysis,” Journal of Sound and Vibration 229(3), 647-667.
Wu, T.X. and Thompson, D.J. (1999a) “Analysis of lateral vibration

behavior of railway track at high frequencies using a continuously
supported multiple beam model,” Journal of the Acoustic Society of
America 106, 1369-1376.
Wu, T.X. and Thompson, D.J. (1999b) “A double Timoshenko beam

model for vertical vibration analysis of railway track at high frequencies,”
Journal of Sound and Vibration 224, 329-348.
Xu, P.C., Mal, A.K. and Bar-Cohen, Y. (1990) “Inversion of leaky Lamb
wave data to determine cohesive properties of bonds,” International journal
of engineering science 28, 331-346.

Bartoli - 2007 - STRUCTURAL HEALTH MONITORING BY ULTRASONIC GUIDED WAVES

Uploaded by

Copyright:

Available Formats

You might also like

Bartoli - 2007 - STRUCTURAL HEALTH MONITORING BY ULTRASONIC GUIDED WAVES

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bartoli - 2007 - STRUCTURAL HEALTH MONITORING BY ULTRASONIC GUIDED WAVES

Uploaded by

Copyright:

Available Formats

Electronic Theses and Dissertations

eScholarship provides open access, scholarly publishing

STRUCTURAL HEALTH MONITORING BY

A dissertation submitted in partial satisfaction of the

Professor Francesco Lanza di Scalea, Chair

Ivan Bartoli, 2007

All rights reserved.

quality and form for publication on microfilm:

University of California, San Diego

Michela, Gabriella e Domenico

Table of Contents …………………………………………………………………..v

Abstract of the dissertation……………………………………………………..xxx

2 INTRODUCTION TO ULTRASONIC GUIDED WAVES......................6

2.1 Introduction ........................................................................................... 6

2.2 Equations of motion in Unbounded Medium....................................... 9

2.3 Guided plate waves: Lamb waves ...................................................... 15

2.4 Analytical solutions of Guided Waves for Multilayered systems. ..... 27

2.5 Numerical solutions of guided waves in Multilayered systems. ........ 31

3 MODELING WAVE PROPAGATION IN DAMPED

3.1 Introduction ......................................................................................... 35

3.3.1 Problem definition ........................................................................ 40

3.3.3 Semi Analytical Finite element method ...................................... 44

3.3.6 Group and energy velocity .......................................................... 52

3.4.1 Plate systems................................................................................ 55

3.5 SAFE for axial symmetric systems .................................................... 81

3.5.3 Elastic copper tube filled with elastic and

3.5.4 Elastic steel pipe coated by a thin layer

3.6 Forced responces................................................................................ 113

3.6.2 Response to arbitrary force ....................................................... 120

4.1 Introduction ....................................................................................... 124

4.3 Dispersion results for different bond states ..................................... 137

4.5 Strength of Transmission as a Function of Bond Condition .......... 147

4.6.1 Test Specimens ........................................................................... 154

4.6.3 Experimental results................................................................... 157

4.8 Acknowledgements ............................................................................. 163

5 GUIDED WAVE STRUCTURAL MONITORING OF RAILROAD

5.1 Introduction ....................................................................................... 164

5.4 Conclusions ........................................................................................ 188

6 GUIDED WAVE STRUCTURAL MONITORING OF RAILROAD

6.2 Defect detection using hammer generation/air-coupled detection ... 194

6.2.2 Effect of Sensor Lift-off Distance .............................................. 199

6.3.1 Defect detection using laser generation/air-coupled detection.. 206

6.4 Conclusions ........................................................................................ 215

7 GUIDED WAVE STRUCTURAL MONITORING

7.2.1 Geometry of the strand ............................................................. 221

loaded seven-wire strands.......................................................... 230

7.2.5 Forced solution for the free strand ........................................... 245

in Free Strands (Selection of Optimum Wave

7.3.1 Test Protocols ............................................................................ 260

7.3.3 Results – Test 2 ........................................................................ 266

7.4.2 Statistical defect classification ................................................... 280

7.5 Discussion and Conclusions............................................................... 283

8 CONCLUSIONS AND RECOMMENDATIONS

8.1 Conclusions ........................................................................................ 288

8.2 Recommendations for future studies................................................. 294

Table 3.1: Elastic properties for the T300/914 laminate

Table 3.2: Elastic and viscous properties of the orthotropic plate

Table 3.3: Elastic properties of the T800/924 lamina (elastic

Table 3.4: Elastic and viscous properties of the orthotropic plate