Professional Documents
Culture Documents
Bartoli - 2007 - STRUCTURAL HEALTH MONITORING BY ULTRASONIC GUIDED WAVES
Bartoli - 2007 - STRUCTURAL HEALTH MONITORING BY ULTRASONIC GUIDED WAVES
Bartoli - 2007 - STRUCTURAL HEALTH MONITORING BY ULTRASONIC GUIDED WAVES
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
in
Structural Engineering
by
Ivan Bartoli
Committee in charge:
2007
Copyright
Chair
2007
iii
To my family
iv
TABLE OF CONTENTS
Signature Page..……………………………………………………………………..iii
Dedication…………………………………………………………………………….iv
List of Figures……………………………………………………………………...xii
Acknowledgements………………………………………………………………..xxiv
Vita……………………………………………………………………………….. xxvii
1 INTRODUCTION ....................................................................................1
1.1 Research motivation .............................................................................. 1
1.2 Outline of the dissertation.................................................................... 4
v
3.3 SAFE mathematical framework .......................................................... 40
vi
4 GUIDED WAVE STRUCTURAL MONITORING OF
COMPOSITE WING SKIN-TO-SPAR BONDED
JOINTS IN AEROSPACE STRUCTURES ......................................... 124
vii
6.2.1 Effect of Sensor Inclination ....................................................... 194
viii
7.3.2 Results – Test 1 ........................................................................ 261
APPENDIX A…………..…………………………………………………………297
REFERENCES…………..…………………………………………………………311
ix
LIST OF TABLES
Table 3.5: Elastic and viscous properties for the UAV wing
skin-to-spar joint (elastic constants in GPa.
*Neau et al., 2002)................................................................... 74
x
and material attenuations for the bond. ............................... 135
xi
LIST OF FIGURES
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
xii
three-node element.................................................................... 42
xiii
(c) attenuation. ......................................................................... 78
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 3.16 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
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 4.1: (a) Wing skin-to-spar bonded assembly. (b) Across bond
and within bond test configuration ....................................... 129
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
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 4.8: 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 5.1: ABAQUS model of the plate with meshes examined. .......... 171
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 5.7: Finite element mesh employed for the rail. .......................... 182
Figure 6.4: Formation of resonances in the lift-off air gap. .................... 197
xvii
crack (100% Head area reduction) as a function of
detection angle. ....................................................................... 198
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
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 7.7: (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
xix
and mode shapes at 100 kHz of the strand. ........................ 244
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
xx
Max load 41 kips = 70% U.T.S. Frequency range .............. 265
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
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
xxii
ACKNOWLEDGEMENTS
research under his guidance. His incessant help and academic passion
Prof. Charles Farrar, Prof. William Hodgkiss and Prof. Enrique Luco for
Phillips, Ankit Srivastava, Jake Finkler and Gaetano Restivo for their
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.
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
Mahmood Fateh, Program Manager and Mr. Gary Carr, Chief of the FRA
this paper.
xxiv
Francesco (2005). The title of this paper is “Ultrasonic guided wave
xxv
VITA
PUBBLICATIONS
Journals
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.
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.
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).
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
by
Ivan Bartoli
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
xxx
1 INTRODUCTION
Ultrasonic testing is one of the most widely used methods today for
(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)
research (Ditri and Rose 1992, Cawley and Alleyne, 1996). For example, in
allowing for long range inspection, are the complete coverage of the
1
2
errors in the interpretation of the guided wave signals and to fully exploit
Analytical Finite Element (SAFE) method for predicting modal and forced
to as Spectral Finite Element (SFE), the SAFE scheme describes the wave
element (FEM) approaches, the SAFE method allows reducing of one order
addition, the SAFE method does not suffer from modeling waveguides with
numerically challenging.
composites. Forced solutions are carried out to predict the arrival times of
the topic of ultrasonic guided waves is provided. The chapter begins with
longitudinal and shear waves are provided. The attention is then focused
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.
the thesis.
2.1 Introduction
boundaries for their existence. Guided waves can travel on the surface of
section beams, axial symmetric rods and cylinders. All the described
therefore their cross-section has the same physical and geometric properties
at all points along the waveguide’s axis. Ultrasonic guided waves are
reflections and mode conversions take place until their superposition form
6
7
and detect only specific modes that are sensitive to critical defects of the
structural component.
waveguides was laid at the end of the 19th century. Pochhammer (1876)
and later Chree (1889) developed the solution of mechanical wave equation
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.
experimentally. For instance Zemanek (1972) was one of the first authors
Lord Rayleigh derived the equation for waves traveling along the free
propagating without leakage that exist at the boundary between two solid
half spaces. The ranges of existence of free wave solutions (in which the
Scholte (1947). Pilant (1972) extended the study of Stoneley and examined
the leaky wave solutions representing waves that attenuate as they travel.
antisymmetric plate modes (see section 2.3). A plot of the roots in the
frequency domain gives the well known Lamb wave dispersion curves.
properties. His modes involve shearing motion in the plane of the layer.
describing guided plate waves (Lamb waves) are reported in the next
section.
where ux, uy and uz are the components of the displacement vector u and
∂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
The generalized Hook’s constitutive law relates the stress and the
Voigt:
11
⎡ ∂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
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
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
u = ∇ϕ + ∇ × ψ (2.10)
⎡ ∂ 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
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
satisfy the decoupled equations. The exponential terms, which are wholly
(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:
λ
λ
z T
t
y
x
λ spatial period wavelength T temporal period
χ=1/λ spatial frequency F=1/T temporal frequency
Vacuum
Plate x
2h
Vacuum
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
∂
uz = 0, ( )=0 (2.18)
∂z
∂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
where Φ (y) and Ψ (y) are functions of the position along the thickness
with a wave velocity equal to c=ω/k. The terms ω and k are the
wavelength λ (k=2π/λ).
obtained as:
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.
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)
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
obtained. Since the systems are homogeneous, non trivial solutions are
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 …,
eigencoefficients that are A2, B1 for the symmetric case and A1, B2 for
20
A2,...
the angular frequency ω, the eigenvalues kiS and kiA will change with the
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
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
corresponding Lamb wave mode. The plot of the Lamb wavespeeds against
(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
its cut-off frequency has infinite wavelength. For the same reason, the
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
packet as a whole. The speed at which the wave packet (or wave
envelope) propagates is the one measured experimentally and is called
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]
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]
u
x
u
y
λ=2π/k0Anti
u
x
u
y
λ=2π/k0Symm
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
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
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
waves. Not treated here, these shear waves induce a particle motion in the
and reflected shear horizontal bulk waves. In general, when guided waves
26
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]
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
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
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
When scattering occurs, part of the energy from the original wave is sent
randomly into different directions within the material. The last mechanism
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,
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
“Global Matrix method” for flat systems, is given by Lowe (1993, 1995).
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).
form the whole structure. From Floquet’s principle, the wave propagation
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
the superposition of bulk waves – SPBW and pure finite element methods.
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
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
highly efficient method for the non-destructive evaluation (NDE) and the
attenuation) are unveiled and managed for the given test structure. For
boundary conditions. For damped and/or leaky waves, the solutions are
wavenumber that satisfy the dispersive equation. The complex roots of the
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
where the order of the dispersive characteristic equation outgrows and the
the problem and calculate the dispersion relations. Built on this concept,
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
the evanescent modes do not transport any energy along the structure,
al., 2002). Other versions of the general SAFE method, again for the
(Volovoi et al., 1998), rods and rails (Hayashi et al., 2003). Both
were studied by SAFE methods for the first time by Dong and Huang
Datta (2003) for laminated plates of both finite and infinite widths. The
38
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
The present study extends the SAFE method for modeling dispersive
This section reviews the linear viscoelastic models that were used in
the SAFE formulation proposed in the present work. As well known, for
C = C′ − iC′′ (3.1)
39
where C′ contains the storage moduli and C′′ contains the loss moduli.
(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 ⎥⎥⎦
⎢⎣
material damping. In the Kelvin-Voigt model (Neau, 2003; Rose, 1999) the
ω f
C = C′ − i η = C′ − i η (3.3)
ω f
C = C′ − iη (3.4)
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
case of the hysteretic model. It is also evident that both models predict
strain field components at each point of the waveguide are expressed by:
u = ⎡⎣ux uy uz ⎤⎦
T
(3.5)
ε = ⎡⎣ εx εy εz γyz γxz γ xy ⎤⎦
T
(3.7)
⎡ ∂ ∂ ∂⎤
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)
Equations of motion for the cross section are formulated by inserting the
where Φ is the strain energy and K is the kinetic energy. The strain
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
energy.
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
discretize Ω. Matlab’s “pdetool” and the “GID” software were used for the
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
⎡ ∂ ∂ ∂⎤
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)
N,y and N,x are the derivatives of the shape function matrix with respect
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
(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
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
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
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
The matrix K2, instead, mixes ux with uy and uz but it does not mix uy
and uz with each other. It follows that:
1994; 1995).
eigenvalue problem in ω(k). All the solutions for this case correspond to
each wavenumber km, M propagating modes (km, ωm) are found along
[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
evaluated roughly one hundred times faster that the linearized version in
Eq. (3.31). In the undamped case, the waves which are nonoscillating
ωc. The cut-off frequencies can be computed by letting k=0 in Eq. (3.30)
Thus this matrix must be updated at each iteration over the frequency
simply be set at the beginning of the simulation for the entire frequency
domain.
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
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
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)
⎡∂ ∂ω ⎤
U LT ⎢ K(k) − 2ω M⎥ U R = 0 (3.36)
⎢⎣ ∂k ∂k ⎥⎦
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
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
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
3.4 Results
orthotropic layers stacked along the z direction (Figure 3.1 (a)). The
the layered plate and each layer lies parallel to the x-y plane. The plates
56
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.
For the SAFE modeling, the complex bulk velocities for the
−1
⎛ ⎞
cL, T = cL, T ⎜⎜1 + i κL, T ⎟⎟ (3.42)
⎜⎝ 2π ⎠⎟⎟
obtained as:
Eν E
λ = µ = (3.44)
(1 + ν )(1 − 2ν ) 2(1 + ν )
⎡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
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
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
Frequency [kHz] Frequency [kHz]
10000
10000
1500
1500
1000
1000 cT
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
Frequency [KHz] Frequency [KHz]
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.
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
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
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.
the principal directions of material symmetry are given in Table 3.1, where
62
C44=0.5×(C33-C12) (3.46)
Cθ = R1CR −2 1 (3.47)
⎡ 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
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
solutions exist in the entire frequency range. When damping is small, the
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
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
θ=0° θ=0°
15000 10000
(a) (b)
4000
5000
2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Frequency [kHz] Frequency [kHz]
θ=45° θ=45°
15000 10000
(c) (d)
8000
Phase Velocity [m/sec]
4000
5000
2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Frequency [kHz] Frequency [kHz]
θ=90° θ=90°
15000 10000
(e) (f)
8000
Phase Velocity [m/sec]
10000
6000
4000
5000
2000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Frequency [kHz] Frequency [kHz]
Symmetric Antisymmetric
10000 10000
(a) (c)
8000 8000
4000 4000
2000 2000
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
Frequency [kHz] Frequency [kHz]
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
Frequency [kHz] Frequency [kHz]
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
Kelvin-Voigt models
thickness. The material’s properties are shown in Table 3.4. The viscosities
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
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
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
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
Frequency [kHz] Frequency [kHz]
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]
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
Hysteretic Kelvin-Voigt
15000 15000
(a) (d)
5000 5000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Frequency [kHz] Frequency [kHz]
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
Frequency [kHz] Frequency [kHz]
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
Frequency [kHz] Frequency [kHz]
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
Frequency [kHz] Frequency [kHz]
5000 5000
(b) (e)
Energy Velocity [m/sec]
4000
3000 3000
2000 2000
1000 1000
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Frequency [kHz] Frequency [kHz]
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
Frequency [kHz] Frequency [kHz]
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)
layer.
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
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
Frequency [KHz] Frequency [KHz]
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
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
kHz) studies (Wu and Thompson, 1999; 1999). Knowledge of the high-
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
Frequency [kHz] Frequency [kHz]
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.
6000 m1
5000
Phase Velocity [m/sec]
4000
3000
2000
1000
0
0 2 4 6 8 10
Frequency [kHz]
m2 m3
m4 m5
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
(a) (b)
10
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
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
3.5.1 Formulation
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
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;
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
⎡∂ 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
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
where N (r) is the shape functions matrix and q (e) is the vector of nodal
⎡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
The stress within the element can now be written in terms of the
approximate displacement vector Eq. (3.51)-(3.52) as:
where:
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
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.
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
where:
ro( e )
m ( e)
= ∫ r ⎡⎣N T ρ (e)N ⎤⎦dr (3.64)
ri( e )
87
to:
nel
to obtain the n-th order axial symmetric modes and k is adopted as the
where
a root, then −km is also a root, where the overbar denotes complex
standing damped modes along the positive z-axis and the other half are
each guided wave the phase velocity can be evaluated by cph = ω / kRe
given by kIm . The attenuation may also be easily expressed as decibels per
For viscoelastic materials solving the Eq. (3) for a given input
solutions are defined as −km , where the overbar means complex conjugate.
can be extracted and the displacement uhm , strain εhm and stress σhm at a
point reconstructed. The time averaged strain and kinetic energy densities
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
= ∫ (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
−1
⎧⎪ Ro ⎫⎪
⎪ ⎤ rdr⎪⎬
= PF ⎨2π ∫ ⎡⎢⎣ S + K
m m m m
Ve (3.74)
⎪⎪ R T T ⎥⎦ ⎪⎪
⎩⎪ i ⎭⎪
off frequencies were calculated with a zero finding routine based on the
1 ⋅ 10−5 Hz . In Table 3.8 the SAFE and SPBW cut-off frequencies for the
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
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).
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
elements and ωmax = 2πfmax is the highest circular frequency for which the
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)
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
3.5.3 Elastic copper tube filled with elastic and viscoelastic bitumen
mode. In the 0-150 kHz frequency range the bitumen was found behaving
For the SAFE calculation, the elastic stiffness tensor C for the
for the bitumen were
copper and the constant complex stiffness matrix C
properties do not have an effect on the T(0,j) modes for this system were
elements, ten in the core and one in the pipe, were used to satisfy the
Torsional modes were analyzed by considering only the 23 uθ dof over the
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
roots with att > 130 dB m-1 were marked with a brighter color. The
97
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
Frequency [kHz] Frequency [kHz]
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).
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
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
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
The mode shapes were normalized with respect to their maximum value.
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
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)
5 5 5 5 5
Uθ Uθ Uθ Uθ Uθ
radius
(a)
norm.
0 0 0 0 0
-1 0 1 -1 0 1 -1 0 1 -1 0 1 -1 0 1
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
displacement and power flow in the tube wall slightly increases while the
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
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
the core with phase and energy velocities close to the shear speed of the
this feature must occur around f3 where the normalized strain energy is
minimum. A similar behaviour, not reported here, has been obtained for
ranges.
(complex wavenumber). Among these solutions, only the Td(0,1) mode has
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
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)
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
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
frequency, the mode attenuation grows linearly while the energy velocity
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
minima of Td(0,1) attenuation occur for those frequencies where the phase
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
mode power flow PF Td (0,1) is represented. It can be noted that this ratio
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
-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
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
shape. (c) normalized radial power flow density PrT (0,1) . Frequency for each d
peak, the normalized cross sectional distribution of the radial power flow
minimum of the normalized radial power flow takes place, the Td(0,1)
given in Table 3.7. The bulk wave attenuations for the viscoelastic
(2002) twelve finite elements across the steel pipe thickness and three
elements for the bitumen layer were used. For the Longitudinal modes,
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
Frequency [kHz] Frequency [kHz]
Here, in addition, the roots with att(m) > 10 dB m-1 are represented with
a brighter marker.
106
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
due to the fact that the bitumen coating is very thin in comparison with
(2002).
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
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
3.19(b) and (c) the normalized mode shapes and the normalized cross
the radial power flow distribution have a very small magnitude. These
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
(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
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
detecting defects and monitoring loads Pavlakovic et al. (2001). The aim
with external diameter equal to 152.40 mm . The present three layer axial
in Figure 3.20(a) and 8(b) which was the subject of several experimental
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
(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
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
Frequency [kHz] Frequency [kHz]
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
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
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
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
For example, highlighted in the insets of Figure 3.21 are the three
pulse dispersion and to reduce the risk of mode overlapping that generally
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
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,
displacement in the axial direction. Radial power flow and energies are
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
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
noted that the larger mesh adopted in the current example results in a
In this paragraph the steady state response for a time harmonic load
response must be periodic with the same frequency ω of the applied load.
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)
{⎡⎣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
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
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)-
⎡0⎤
p = ⎢ ⎥ (3.81)
⎢⎣ f ⎥⎦
2M
Q = ∑Q
m =1
mΦRm (3.82)
2M
(A − kB) ∑ QmΦRm = p (3.83)
m =1
2M
ΦLl (A − kB) ∑ QmΦRm = ΦlLp (3.84)
m =1
⎧⎪0 l ≠ m,
ΦLl AΦRm = ⎪⎨ L
⎪⎪⎩ΦmAΦRm = kΦLmBΦRm l = m
(3.85)
⎪⎧0 l ≠ m,
ΦLl (−kB) ΦRm = ⎪⎨
⎪⎪⎩−kΦLmBΦRm l = m
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
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
the surface at x=xs the forces can be represented by the Dirac delta
where p contains the amplitudes of the nodal loads applied in x=xs. The
+∞ +∞ 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,
2M
ΦLmp Rup i ⎡⎣ξm (x − xS )⎤⎦
U (x, ω ) = ∑ − Φm e (3.92)
m =1 Bm
∑α
i ⎡⎣ ξm (x − xS )⎤⎦
U (x, ω ) = mΦRup
m e (3.93)
m =1
ΦLmp
αm = − (3.94)
Bm
It should be noted that in the Eq. (3.93) the summation is extended only
fundamental a0 mode.
118
how the epoxy layer, that is assumed to have poor stiffness properties, is
(a)
x t
p(t)
z
Figure 3.23: (a) Deformation of a plate subjected to a uniform harmonic
vertical load (b), (c).
119
(a)
x t
(a)
xx
z Bond layer
(b) Bond
Bond
x
p(t)
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
Fourier Transform,
∞
1
V (x, t) =
2π ∫ V (x, ω )eiω tdω (3.97)
−∞
tone-burst pulses was predicted with the approach just discussed (Section
7.2.5).
121
sections. The main innovation over SAFE models proposed in the past is
values (for damped waveguides) and the group velocity values (for
frequency solution, without the need for tracking the modes and without
embedded rod).
of the problem for the highest frequency of interest. Obviously the number
122
plates and cylinders with quadratic elements smaller than 1/4 of the
the mesh should be refined further to compensate for the loss of accuracy
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
group velocity and the attenuation curves are calculated with the same
order of accuracy, while the energy velocity curves require a more refined
tool.
3.8 Acknowledgements
this paper.
4 GUIDED WAVE STRUCTURAL MONITORING
OF COMPOSITE WING SKIN-TO-SPAR BONDED
JOINTS IN AEROSPACE STRUCTURES
4.1 Introduction
have been examined for the wing skin and two different types of bond
simulated. The assessment of bond state has been based on monitoring the
in the aerospace industry due to their light weight and high strength. One
Vehicles (UAVs), that are employed for both military and civil purposes
124
125
this application.
predicted during the cure process of an epoxy resin layer between solid
semispaces.
In addition, through the use of built-in actuators and sensors, the guided
philosophy.
126
inspection of bonds. In the first approach, the waves are both generated
adherend on one side of the bond and received across the bond (“across
approach is the occurrence of mode conversion when the wave enters and
leaves the bond due to the transition from the single adherend geometry
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;
aerospace panels (Castaings and Hosten, 2003; Hay et al., 2003). The
“across the bond” configuration was also used successfully for the
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
has gained increasing interest only recently (Light et al., 2003). The use
and Chang, 1995). However, the preferred application has been the
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).
cured adhesive, poorly-cured adhesive and disbonds. The “within the bond”
configuration and the “across the bond” configuration have been examined
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
varying bond conditions. Both the predictions and the experiments show
that the degradation of the bond has a marked effect on the strength of
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
joint. The bond conditions that were examined included regions with
two testing configurations of “within the bond” and “across the bond,”
flexural rigidity and the ±45-deg fibers provide torsional rigidity to the
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
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
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
shapes of the excitation and the carrier modes found from the models,
dimensional finite elements with quadratic shape functions and three nodes,
with three degrees of freedom per node. Under the hypothesis of time-
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
isotropic layer with the following standard constitutive laws in the global
directions as defined in Jones (1975), and the stress vector σ and strain
manipulations:
(A − kB) Q = 0 (4.6)
where the matrices A and B have been defined in the previous Chapter.
described in Chapter 3 the energy velocity (see Eq. 3.38), rather than the
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
⎛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
mode over a unit period of time, and it was used to predict the wave
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-
introduced in Chapter 3 (see Eq. 3.4) was assumed where 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
the stiffness matrix were assumed equal to the values used by Neau et al.
(2002).
Ceq , for the equivalent spar layer were computed by averaging the
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
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
Table 4.2: Ultrasonic bulk longitudinal and shear velocities and material
attenuations for the bond.
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
this case the viscoelastic matrix, C*, depends only on the two elastic
calculated from the bulk longitudinal and shear wave velocities, cL* and
−1
⎛ ⎞
cL,S = cL,S
* ⎜⎜1 + i αL,S ⎟⎟ (4.10)
⎜⎝ 2π ⎠⎟⎟
where αL,S are the longitudinal and shear attenuation in the material,
for the properly-cured epoxy and for the poorly-cured epoxy were obtained
in the model by reducing the shear wave velocity (real part) of the
attenuations.
137
10000
00
10000
000
00 A1 000
(a) (b)
8000
00
8000
000
S0
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
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.
and five elements for the spar. The number of elements was doubled to
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
For the ease of the reader, throughout this paper the conventional
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-
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.
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
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
Figure 4.3(b) and (c) show that A1 is propagative only above 135
kHz whereas the other three modes are propagative throughout the
frequencies above 135 kHz, Figure 4.3(b), they are not considered further
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
range examined. Moreover, PZT disks used for built-in ultrasonic structural
thus coupled very effectively to the built-in PZTs. The opposite is true for
In this figure the phase velocity results are obtained from the SAFE
model of the bonded joint with and without damping losses. The
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
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),
three frequencies of 155 kHz, 205 kHz and 255 kHz, respectively. At the
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
transmission through the joint. Exciting modes with large energy transfer
[0/±45/0]S Skin
x
y
z
Spar
Properly-cured
Bond
(a)
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
the wave power flow between the two cases were predicted, and this will
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
Frequency [KHz] Frequency [KHz]
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
changes in the phase velocity, energy velocity and attenuation curves are
modes that did not exist for either of the two bond cases discussed
modes of the disbonded joint essentially coincide with the solutions of the
single wing skin plate, represented in the plots by the open dots.
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
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
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
Mode conversion effects as the wave travels through the joints were
energy transmission across the bond in the “across the bond” test
for the “within the bond” test configuration were based upon initial
the single-plate s0 mode. These assumptions were consistent with the in-
Figure 4.6(a) shows the displacement mode shapes for the [0/±45/0]S
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,
for the S0 and A1 carrier modes, and its symmetry within the upper plate
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
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.
both longitudinal and shear stiffness of the adhesive layer. Between the
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
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
that S0,plate is the predominant carrier mode through the disbonded region.
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
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
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
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
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
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
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).
frequency range 100 kHz – 300 kHz, for both plate lay-ups ([0/±45/0]S and
bond).
149
normalized power flow, averaged over the thickness of the upper plate and
⎡ ⎤
⎢ 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.
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
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
degraded bonds and the properly-cured bond. First, the relative excitability
considering the single incoming mode, s0, for the “across the bond”
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.
index. The following Excitation Factor was defined for an i-th carrier
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 )
displacement component of the incoming mode, uk,IM, and that of the i-th
frequency. For the subject tests, the excitability remains dominated by 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.
attenuation losses. This effect was only factored explicitly into the
Across Bond
(a) [0/±45/0]s (b) [0/±45/90]s
1 1
- - - Properly-cured - - - Properly-cured
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)
(Poorly-cured) – (Properly-cured)
Relative Transmission Strength
Disbond – (Properly-cured)
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
Frequency (KHz) Frequency (KHz)
(Poorly-cured) – (Properly-cured)
Relative Transmission Strength
Disbond – (Properly-cured)
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
Frequency (KHz) Frequency (KHz)
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
maintained for the Excitation Factors of S0 and A1 (with a zero value for
S0,plate mode, being the only one relevant for the disbonded case.
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
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
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).
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
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.
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
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
Bonding of the plates was done such that the 0-deg fiber direction ran
artificially created prior to assembling the joints (Figure 4.9). The majority
were 42% degradation for the longitudinal stiffness and 44% degradation
for the shear stiffness as reflected by the velocity values in Table 4.2).
bond. The release film was expected to severly degrade the shear stiffness
of the bond while degrading its longitudinal stiffness to a smaller extent.
25.4 mm × 25.4 mm, respectively. The total bond thickess was measured
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
Guided wave testing was performed in the joints with the objective
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
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
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
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
the energy of the signal) was computed following the known relation:
∑x
i =1
i
2
(4.13)
RMS =
Z
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
the RMS spectra difference between each defected bond and the properly-
cured bond, and then normalizing the results by the RMS spectrum of the
the power flow results from the model in Figure 4.2(c) through (f).
each test are shown in Figure 4.11. It can be seen that each of the
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
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
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
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
transmission will be somewhat reduced from the predicted values when the
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
Frequency (KHz) Frequency (KHz)
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
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
the majority of the wave energy within the same upper plate where the
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
defect cases, the [0/±45/90]S “across the bond” test remains the least
favorable since the three defects yield close results. The testing
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
and shear stiffness), and a disbond where the shear stiffness was nominally
lost.
damping to predict modal solutions for the joints when probed “across”
161
and “within” the bond. An excitation factor for the possible carrier modes
between the displacement mode shapes of the incoming mode and those of
the possible carrier modes. Given an incoming s0 mode which was known
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
the changes in the power flow (Poynting vector) along the wave
propagation direction relative to the properly-cured bond. For all the cases
properly-cured joint.
regions with isolated disbonds of two different sizes. The specimens were
162
tested in the “within the bond” and the “across the bond” configurations
increases in the presence of the two types of bond defects compared to the
carrier modes. In most of the cases, it was found that the poorly-cured
This result was opposite to what found by the SAFE power flow
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
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
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
(-40 °C to 74 °C) for high altitude flights and storage in closed hangers,
4.8 Acknowledgements
5.1 Introduction
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
notoriously the most dangerous flaws in rails. Safety statistics data from
indicate that train accidents caused by track failures including rail, joint
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
164
165
The next two chapters will discuss some of the aspects of the
railroad tracks.
vibrations are only accurate at frequencies below 0.5 kHz and 1.5 kHz,
Thompson (1993) the finite element method was used to model transient
displacement between the head and the foot of the rail was used by Wu
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
programs. These techniques are very efficient to calculate the wave modal
(Wilcox et al., 2003; Cawley et al., 2003) and adaptive mesh refinement
defects were the first leading cause of track failures in the US with $91M
defects of four different sizes and three different orientations were recently
impulsive excitation. The excitation used in the rail model simulates the
transverse-type flaws in the rail head of four different sizes and three
classical case of Lamb waves in plates. Results from the simulation could
learning. The simulations prove also useful towards the ultimate goal of
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
∆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 integration time step and the element size. Depending from the
1
∆t = (5.2)
20fmax
The size of the finite element, Le, is typically derived from the
nodes per wavelength are normally required (Alleyne and Cawley, 1991),
as
Le = λmin / 20 (5.3)
broad frequency domain. This work was also aimed at identifying criteria
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
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
wave velocity cL=5.98 km/sec; bulk shear wave velocity cT=3.20 km/sec.
plate based on the medium width of the web of the railroad track
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
the table of Figure 5.1, where a and b are the element sizes along 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.
(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
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
signal in the joint time-frequency domain, are shown in Figure 5.2 for the
single detection point. Group velocity dispersion curves for the plate model
the arrival times of the scalogram peaks for the two nodes following
L
Cg (f ) = (5.6)
t2 (f ) − t1(f )
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)
(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)
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
examining the effect of the element size both in the longitudinal direction
Figure 5.3: Group velocity dispersion curves extracted from the model for
varying mesh refinements. Plot also shows the Rayleigh-Lamb exact
solution.
174
from the plate edge. Both a perpendicular edge and a 35 deg inclined
transverse and inclined defects in the rail. The finite elements were
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
pulse at frequency f, and the subscripts 1 and 2 refer to the first arrival
The results are presented in Figure 5.4 for the different mesh
the perpendicular edge, Figure 5.4(a) and (b), the reflection coefficient for
5.4 (c)) decreases with increasing frequency because the remaining energy
A0 mode incident on the inclined edge (Figure 5.4 (d) and (f)). Contrarily
effect on the inclined reflection in all cases. The reason for this is that it
b=2mm) for the oblique reflections in Figure 5.4(c) through Figure 5.4(f).
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
km/sec. From Eq. (5.3), the corresponding limit on the element size is Le
modeled with the SAFE method. Since quadratic elements were used in
more stringent criteria (10-20 elements per wavelength) used here are
simulations.
that is satisfied with the value of 0.2 µsec used for the analysis.
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
Eq. (5.8).
(a) (b)
(c) (d)
(e) (f)
defects as far away as tens or even hundreds of feet from the sensoring
system.
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
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
the rail head. These are among the most dangerous flaws in rails as
section that was studied. The elastodynamic properties of the rail steel
material were the same as the ones assumed for the plate. Four defect
head section that was cut, i.e. ‘15% defect‘, ‘50% defect‘, ‘85% defect‘, and
‘100% defect‘. For all four sizes, three different orientations were
20 deg from the transverse direction, and one inclined at 35 deg from the
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
the top of the rail head at its left end along the vertical direction y. This
impulse is shown in Figure 5.6. The same time history was used as a
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.
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
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 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
5.8 shows two time signals for the vertical acceleration with the
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
bending wave and its reflection from the defect can be clearly identified.
split contained energy that was more than one order of magnitude smaller
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)
slight dispersion at the low frequencies. Also, the agreement between the
discrepancy of 4%.
183
Vertical Acceleration
Vertical Acceleration
(a) (b)
0 0.5 1 0 0.5 1
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).
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
these figures. It should be noted that the experimental data presented here
derived:
184
α (f ) (2L2 )
spectrum from Eq. (5.7), α(f) is the linear attenuation spectrum of the
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
general reflection coefficients that are not bound by the experimental setup
defect, resulting in stronger reflections than the raw data would indicate.
measured for the vertical bending mode (Lanza di Scalea and McNamara,
2003). In that study the attenuation was found highly varying with
between the real defects obtained by saw cutting and the flaws simulated
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
general decrease in reflection strength for the 20 deg oblique defects when
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
5.10 where the reflected lateral bending mode is now being monitored with
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
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
direction y. Since this node is on the plane of symmetry of the rail, the
computed from the usual Eq. (5.7), where now GWT2 refers to the
and GWT1 refers to the scalogram of the vertical acceleration for the first
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
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
this case the strongest reflections occur for the largest 100% defect. The
5.4 Conclusions
impulsive excitation, are examined in a broad frequency range with the aid
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
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
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
was also obtained for comparison with the finite element predictions. The
the rail head are found in the 20 kHz-45 kHz range. As expected, the
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
explaining why in some cases stronger mode conversions are predicted for
smaller defects.
type flaws and that can be operated at high speeds. Details of the
coefficient spectra from this variety of defect sizes and orientations can
Support Vector Machines for ADC of transverse defects in rails has been
6.1 Introduction
and are oriented at 0° from the surface of the rail head to detect
suffers from a limited inspection speed (typically 20 mph in the U.S.) and
the rail and the inspection wheel. More importantly, horizontal surface
cracks such as shelling can prevent the ultrasonic beams from reaching the
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
steel (on the order of thousands of miles per hour), resulting in the
techniques that do not require contact with the rail is being investigated
al., 2002; Kenderian et al., 2003; McNamara and Lanza di Scalea, 2004) as
are used for detection of the waves at distances much larger than those
the defect detection reliability, the inspection range and the inspection
speed.
(a)
(b) (c)
laboratory. The experimental setup used for the first system is shown in
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.
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.
detection angle from the normal to the rail surface, α, is given by Snell’s
⎛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
(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.
detect the wave incoming from the hammer that travels in an opposite
direction from the crack reflection. In this case the optimum detection
reflection from the defect are clearly visible. Figure 6.2(b) is the
defined as the ratio between the reflected energy, A2, and the incoming
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)
T
2
2
Air
Steel
1
T
1
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
reflection strength increases when the sensor is oriented towards the defect.
problem, such as when reflections are sought from small defects, the
sensors should be oriented away from the generation hammer. This point
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
Figure 6.5. The optimum detection angle depends on the phase velocity
beyond the 45 kHz limit examined here. Values of R larger than 1, are a
very small for sensor angles between – 10º and 10º. This implies that some
field.
Positioning the sensor as far away as possible from the top of the
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
stable reflection from the large transverse defect were identified between 76
that such frequencies are appropriate for detecting transverse head defects
exists for a truly “non-contact” inspection system that stays outside the
following:
Asensor # 2 (f )
T (f ) = (6.3)
Asensor # 1 (f )
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
40 First Arrival 40
Vertical displacement
Vertical displacement
(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
Time (msec) Time (msec)
40 40
Frequency (KHz)
Frequency (KHz)
(c) 30 30 (d)
25 Defect echo 25
(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
because much of the energy is reflected, rather than transmitted past the
sensor #1 besides the first arrival. This demonstrates that reflection data
can be monitored concurrently with transmission data. Figure 6.7 gives the
scanned over the defect (the distance from sensor #2 to the defect, L2, is
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.
10º, 20º and 35º from the transverse rail direction. These oblique defects
the results were more scattered than for the perfectly-transverse crack.
is too slow for field application. The Discrete Wavelet Transform (DWT)
203
given by:
∆×F
f (j) = (6.4)
2j
where ∆ is the sampling frequency of the original signal and F is the
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)
0.04
Detail Level 6 Thresholding
0.02
(a) 0
-0.02
-0.04
0 1000 2000 3000 4000
-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
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.
energy. The plot in Figure 6.8(a) resembles closely the raw time history
from the first arrival due to noise and reverberations. If only the largest
the defect reflection now appears clearly (Figure 6.8(b)). This is called
the threshold was set to 60% of the maximum value of the wavelet
to only 60 point at level 6 and much less after thresholding. The task of
transverse head cracks, and oblique head cracks at 10º, 20º and 35º
orientations.
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
as a few tens of Megahertz. As it will be shown later, the need for large
lift-off distances limit the operating frequency range to 100 kHz – 1 MHz.
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
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
mm. A 50 kHz high-pass filter was employed for the signals shown in the
following sections.
207
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.
reconstructed signal for sensor #1 in Figure 6.10 (e). In this case the de-
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
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
such a crack can be detected as far away as 500 mm (20”) from the
completely non-contact and the inspection probes are located far away
from the rail surface, these results are very encouraging for meeting the
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
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
sensor #1 prior to the defect (Figure 6.13(e)), and a decrease in the first
(a)
Vertical displacement
Vertical displacement
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
Time (msec) Time (msec)
First Arrival
800 Defect Echo 800
Frequency (KHz)
Frequency (KHz)
First Arrival
200 200
(e) 0 0 (f)
-0.5 -0.5
-1 -1
0 0.5 1 1.5 0 0.5 1 1.5
Time (msec) Time (msec)
0.4 0.4
Vertical displacement
Vertical displacement
0.2 0.2
0 0
(a) (b)
-0.2 -0.2
800 800
Frequency (KHz)
Frequency (KHz)
First Defect Echo
First
Arrival
Arrival
600 600
(c) (d)
400 400
Defect Echo
200 200
(e) 0 0 (f)
-0.2 -0.2
Defect
Echo Defect
-0.4 -0.4
Echo
T= 6°
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°
Vertical displacement
Vertical displacement
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
Time (msec) Time (msec)
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
Time (msec) Time (msec)
(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
Time (msec) Time (msec)
T= 6°
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,
Frequencies higher than 500 kHz did not produce an increase in defect
detection sensitivity in this case, even for the smallest flaws. The changes
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°
6.4 Conclusions
guided waves for the detection of transverse head cracks in rails that
measurements.
detection sensitivity. Transverse cracks larger than 15% of the head cross
frequencies between 100 kHz and 600 kHz that can be excited by a pulsed
laser.
(32’) for cracks larger than 15% head area and at least 500 mm (20”) for
large portion of the rail at once, thereby increasing inspection speed. This
the top of the rail head, satisfying the recommended clearance envelope for
7.1 Introduction
and Berlin (Schlaich et al. 1980). Monitoring of the applied loads and
Many techniques have been applied to the defect detection and load
degradation is visible at the surface of the structure; this may be too late
the highly scattering nature of concrete. Also, safety issues dissuade from
217
218
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
suspension bridges, the Magnetic Flux Leakage (MFL) method has gained
method cannot be used for continuous health monitoring and it does not
characterize and locate faults in suspended bridge cables (Liu et al., 1998)
Although the potential for void detection, the TDR cannot be used to
thickness; voids are detected by early wave echoes detected at the free
surface. The IE method cannot detect defects within the tendons, e.g.
tendons.
the strand for continuous structural monitoring, (2) the potential for
for the strands with the same sensing system, and (3) the possibility for
detecting both active defects and pre-existing defects toggling between the
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
and Teller, 1995; Kwun et al., 1998; Beard et al., 2003; Rizzo and Lanza
mode, for the real-time monitoring of active defects using acoustic emission
for both steel and composite cables (Kwun and Teller, 1995; Casey and
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
“Dsiamerica” http://www.dsiamerica.com/products/Multistrand_System.html)
are Grade 270, 15.2 mm (0.6”) diameter, seven-wire strands with a cross
dh diameters of core and helical wires respectively; the lay angle β of the
⎛ 2πR ⎞⎟
β = arctan ⎜⎜ ⎟ = 7.9 deg (7.1)
⎜⎝ p ⎠⎟⎟
222
strand to the centre of helical wire and p=230mm is the pitch of the
helical wire.
Axis of reference
of cylinder
Wire cross
ρ p
section s β
r
Axial line of
helical wire R
Figure 7.1: Geometry of the strand: helical wire wrapped around a core.
223
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
duration of ~8ns was used to generate at one end of the strand, while at
was used by Rizzo and Lanza di Scalea (2004) that used a Michelson
and Lanza di Scalea (2004), the following results cover a broader frequency
range (f<1MHz).
Two wires, the straight core and a peripheral helical wire, were
isolated by disassembling a 720mm long, 15.24mm (0.6”) diameter strand
5.08mm diameter steel bar was meshed using 720 triangular elements.
node of the circular section and inclined of 45deg with respect to the
224
response of the PICO sensors, the acceleration was computed as the sum
carried out adopting the Complex Morlet Wavelet (center frequency equal
noted the strong dispersive behavior of the first flexural mode F(1,1) at
cg = D / t(f ) (7.2)
where t(f) is the arrival time of the scalogram local peak corresponding to
dispersion group velocities computed with the SAFE. The curves obtained
setup are shown in Figure 7.3 for a single straight wire (central core) and
Both time histories include a reflection from the wire end of the
420µsec in Figure 7.4(a) and in Figure 7.5(a). The reflected L(0,1) mode is
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
summation of all the modes carrying energy and propagating in the right
by the coupling between PICO sensors and wire surface. Furthermore, the
wire end that are not predicted by the theoretical SAFE model and.
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
arrival times are slightly more significant for the case of the peripheral
Figure 7.4(b) it can be concluded that the helical waveguide has dispersion
analyzed.
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
couplers that could produce some additional effect on the behavior of the
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
(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
L(0,2)
(b)
F(1,2)
L(0,1) Reflected
F(1,1)
L(0,1)
Time (µsec)
Figure 7.4: (a) Measured time waveform from Laser Ultrasound testing of
single steel wire (helical wire, Length=720mm). (b) Normalized (Complex
Morlet) Continuous Wavelet Transform scalogram of time waveform.
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
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
modes are still visible in Figure 7.6(b) but longitudinal and flexural modes
Energy introduced in one of the wires can leak in the adjacent ones but it
load of 260.7kN. The diameters of the strand core wire and the individual
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)
Figure 7.5: (a) Measured time waveform from Laser Ultrasound 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
(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)
Figure 7.6: (a) Measured time waveform from Laser Ultrasound testing of
unloaded strand. Guided waves excited in central wire and received in
helical wire (Test C-P, Strand length=640mm). (b) Normalized (Complex
Morlet) Continuous Wavelet Transform scalogram of time waveform.
Theoretical arrival times are superimposed to the scalogram.
233
Following Machida and Durelli (1973), the length s of the axial line
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
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
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
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
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
possible between adjacent helical wires but with minor frictional forces
Figure 7.7(c).
compute P as
Nh
P = (7.8)
ρ
⎛ p ⎞⎟
2
1 ⎜⎜ ⎟
ρ = R+ (7.9)
R ⎜⎝ 2π ⎠⎟
Hertz theory that models the interaction between two frictionless elastic
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.
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.
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
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 stresses induced by the axial load applied to the strand change
contact stresses force the wires to remain “attached to each other”. In the
σr
C1
P
r σy
C2
(a) (b)
the contact between the cross section of the wires. This interaction is
changing the cross section of the strand i.e. increasing the length of the
peripheral wires.
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
where β is the uniform rate of pretwist in the axial direction. The above
vector (ε) and stress field (σ) with respect to the body coordinate system
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 ⎥⎦
⎡ 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)
motion:
⎛1 ⎪⎧⎪ ⎪⎫ ⎞
⎜⎜ εdξ dη⎪⎬ dζ ⎟⎟⎟dt = 0
t2
δ∫ ∫ ⎪⎪∫∫ ∫∫
⎨ ρ u T u dξ dη − εT C (7.18)
⎜ ⎪ ⎟
t1 ⎜⎝⎜ 2 ζ ⎩ Ω Ω ⎭⎪ ⎠⎟
the mass density and Ω is the area of the waveguide cross section.
⎡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 ⎥⎦⎥
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
the structural component shown in Figure 7.10(a). The material has the
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
frequency of 700 kHz. Notice the complexity of the modes. Mode shapes
al, 1998) first reported that in unloaded strands, the wave propagation
properties were the same as those seen in individual wires while under
a cutoff frequency slightly above 100 kHz that could justify the dispersive
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
Phase Velocity [m/sec]
7000
6000
5000
4000
3000
2000
1000
0
0 100 200 300 400 500 600 700
Frequency [kHz]
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.
Dispersion curves were found unchanged for lay angles typically found in
some of the experimental results that will be shown in the final part of
waves are generated using a PZT actuator bonded on the external surface
246
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
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
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
∑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.
force. This approach was followed to simulate the frequency sweep that
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]
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]
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]
simulation. The effect of the two pin forces representing the PZT actuator
was considered separately and the total displacement response was finally
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
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
experimental results shown in the next sections. Each curve in the plot
Ultimate tensile load). The curves do not show any appreciable difference
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
the cross section due to the radial contact forces P generated by the axial
These showed marginal effects on the SAFE results and therefore are not
presented here.
stress free bodies. Some studies attempted using the covariance effect to
monitor prestress level in strands. The works by Chen et al. (1998) and
and detect the waves. Washer (2001) applied the same concept for the
acoustoelastic change of ultrasonic bulk waves did not show robust changes
the predictions are ineffective to determine the exact value of applied load.
considering the strand anchorage, SAFE was able to capture the main
Lanza di Scalea et al. (2003) showed the effect of anchorages on the wave
the mentioned work, the authors concluded that for an ultrasonic guided
wave packet traveling in the loaded strand across one of the anchorages,
losses are expected and that a considerable portion of the ultrasonic energy
The SAFE method was here used to predict the effect of the
ultrasonic leakage into the anchorages (Figure 7.13). The loading condition
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
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
larger diameter was coupled to the seven wire strand waveguide (see
external diameter of 50mm (2”). Contact between the ring (anchorage) and
the strand was assumed equal to the length of the indentation induced by
load = 70% of U.T.S.) on each peripheral wire was 2mm long. The tendon
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
the simple application of the equivalent nodal forces vector FB’’(ω) to the
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.
(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.
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
Figure 7.14(a) and Figure 7.14(b) show the RMS spectra for the
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
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]
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]
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]
range considered, the RMS curve of the PICO peripheral drops in the
on the leakage of guided wave energy into the anchorage. The guided
wave generated by the PZT is unaffected until the wavefront reaches the
the peripheral wires is lost into the wedge reducing the strength of the
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.
tensile loading. The specimen tested was a Grade 270, 15.2 mm (0.6”)
yield stress of 1.67 GPa (243 ksi), and 1.82 m (72”) in length. Table 7.1
A 10° serrated wedge was placed on the tendon and inserted into a
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
Load-unload cycles were performed with 11 load steps in each cycle. The
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),
(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"
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
the sensor lay-out is shown in Figure 7.15(a). Two strain gages were
opposite peripheral wires. The following three different sensors were used
type) placed on the bottom end of the strand, one on the central wire
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)
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
PICO sensor.
generator card (PXI 5411) and two, 20GS/s 12-bit multi-channel digitizers
was used to control the sensors, acquire and process the data.
7.2. In Test 1 the PICO sensor on the central wire was used as the
and the MsS receiver. Seven load-unload cycles were completed under Test
three PZT sensors and the two PICO sensors were used as receivers. Two
261
at high frequency of the MsS coils, only the low frequency range of 50kHz
In Test 3 the PZT sensor at 35” from the strand end was used as
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
feature examined was the ultrasonic energy leakage between the central
262
wire and the peripheral wires. Figure 7.17 shows the energy leakage
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
With increasing load level, the interwire contact and the wedge
strand contact increase and, at the same time, the wave damping increases
The wave energy leakage in the anchorage and the increased wave
damping seem to dominate the Test 1 results, causing the observed trend
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
load and unload ramps can be attributed to the fact that interwire
contact stresses and contact interaction in the anchorages are not released
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]
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).
not require a baseline. Another feature which would not require a baseline
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
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]
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]
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
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.
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–
increasing load. The behavior is consistent during load and unload, and
50% between zero load and maximum load of 41 kips (70% U.T.S.). In
In Test 2 the MsS coil was used as a transmitter. This test set-up
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
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]
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)
receiving. Load and unload ramps. 100% load = 70% U.T.S. Low
frequency range (300 kHz – 700 kHz).
268
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
the need for a baseline measurement even in the energy leakage feature.
the strand and detecting at the strand’s bottom end. This result refers to
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
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
2 2
start0%
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]
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
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.
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
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,
to the SAFE numerical results shown in Figure 7.14(c). Hence the main
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
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
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
different load levels. Although the trend is consistent between load and
Figure 7.25 shows the relative frequency shift for the high-frequency
that the frequency decreases with increasing load. This result is due to 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
fully-unloaded strand.
272
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]
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]
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 (300 kHz – 700 kHz).
273
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]
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]
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
(50 kHz – 300 kHz).
274
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]
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]
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]
-0.10
(fload-f0)/f0
-0.15
Test3a Up
-0.20
Test3a Down
-0.25
-0.30
-0.35
-0.40
Load [kips]
-0.10
(fload-f0)/f0
-0.15
Test3b Up
-0.20
Test3b Down
-0.25
-0.30
-0.35
-0.40
Load [kips]
1854
d1=203 d 254
Load Load
detection results will be presented for the grade 270, seven-wire twisted
the helical wire (broken wire, b.w.), which was the largest defect
sectional area.
277
D1 D2
D3 D4
defect reflection
D5 D6
longitudinal guided waves in the strand (Figure 7.26). This frequency was
strands (Rizzo and Lanza di Scalea, 2004). The distance between the
203mm (8”) in all tests. By sliding the transmitter/receiver pair along the
(40”), and 1118mm (44”). The latter was the largest distance allowed by
278
stored after different number of digital averages, namely 500, 50, 10, 5, 2
Two time windows were selected for the direct signal and the defect
frequency, the 320 kHz frequency of interest was contained in the sixth
fj = ∆ × F / 2j (7.22)
the mother wavelet, the scale 2j, and the signal sampling frequency ∆.
Thus the sixth level was the only one considered in the further analysis
Figure 7.27, presenting the signals reconstructed from the first six DWT
detail decomposition levels (D1, D2,…, D6). The original signal was taken
140 µsec) the reflection from a 2.5mm-deep notch in the helical wire. Since
the defect sizing. An optimum threshold combination for the direct signal
and the defect reflection was searched based on obtaining the largest
was found that the larger sensitivities were obtained when setting more
and Lanza di Scalea (2006), optimum thresholds were fixed at 20% of the
maximum wavelet coefficient amplitude for the direct signal, and at 70%
ratios between certain features of the reflected signal, Freflection, and the
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)
D.I. components generally increasing for defects located further away from
280
wave attenuation effects, and its origin is probably associated with the
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
Dζ = (x ζ − x ) K−1 (x ζ − x )
T
(7.24)
transpose matrix. In the present study, since the potential outliers were
stored after averaging over ten acquisitions and corrupted by two different
levels of white Gaussian noise. The noise signals were created by the
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.
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 data were thus collected for each noise level. The added noise can
⎡ 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
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
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
Eight baseline samples are outliers, thus false positive indications. Clear
steps can be seen for increasing levels of damage. All damaged conditions
indications. The MSD values showed good discrimination between all defect
D.I. was still kept at a very low value by selecting only four features of
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
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
against eight) is the only improvement over the low noise level.
multivariate analyses for both levels of noise considered; the outliers are
false positive indications for the baseline data (Damage Size 0) and,
ultrasonic technique has the potential for providing both stress monitoring
coefficients)
0.01
0.00001
0 1 2 3 4 5
Notch depth (mm)
t
1
D.I. (RMS of wavelet
0.1
coefficients)
0.01
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.
strand with anchorages and (c) the case of a rod embedded in grout and
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
For the unloaded free strand, it was shown that each wire behaves
to the loaded case. Furthermore, the waves excited in a single wire remain
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
load to the strand was numerically predicted. The SAFE was successfully
interface.
individual wires near the strand’s end are proposed. The best transducer
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-
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.
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
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
of free strands. For embedded strands, the sensitive frequency ranges may
stress are expected to be similar to those found for the free waveguide
case.
transmitter, and access to the strand’s free end to install the piezoelectric
the entire strand surface; hence the defect detection method is more
installation.
8 CONCLUSIONS AND RECOMMENDATIONS FOR
FUTURE STUDIES
8.1 Conclusions
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
monitoring strategy that uses GUWs, can be successfully applied only after
general that has allowed the prediction of guided wave properties in flat
288
289
combinations.
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.
increases in the 100kHz - 300kHz range. The best sensitivity to the defects
point of the S0 and A1 carrier modes. When mode coupling occurs, in fact,
290
the plate. In Chapter 3 it was shown that the wave attenuation is directly
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
at the top of the rail head in the vertical direction. The defects examined
(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
waves at frequencies higher than 100 kHz becomes unfeasible due to the
smaller finite element dimension required and to the huge amount of finite
between finite element method and SAFE is possible and will be the
object of future studies. The need for modeling waves with frequency
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
transverse defects and, more directly, to the reduction of the rail head
guided waves and an array of air-coupled sensors to detect the waves. The
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
“non-contact” performance.
The SAFE method was used to model (a) the multimode and
(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
ultrasonic leakage from a rod into the surrounding media and to identify
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
individual wires near the strand’s end are proposed. The best transducer
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-
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.
strands. For embedded strands, the sensitive frequency ranges may change
transmitter, and access to the strand’s free end to install the piezoelectric
the entire strand surface; hence the defect detection method is more
predicted, the reflection from flaws is not captured by the present model.
Ongoing research at the NDE & SHM laboratory at UCSD has already
2007).
reflection from small defects in railroad tracks and (c) monitoring changing
finite elements, the resulting general eigenvalue problem has small matrices
and is solved with negligible computational times. However, when a large
to employ the solutions from the previous frequency step as initial guess
content. For several decades the frequency content was extracted by the
signal time domain into its frequency domain. If the waveform in the time
frequencies are localized in time or, in other words, when those frequencies
are detected.
Kim and Kim 2000), plates (Wooh and Veroy 2001, Jeong and Young-Su
297
298
the product of the resolution in time (∆t) and in frequency (∆f) by:
∆t ⋅ ∆f ≥ 0.5 (A.1)
position and the momentum of a free particle. Equation (A.1) implies that
window that is broader in time for observing low frequencies and shorter
that can be detrimental in the presence of multiple echoes (Jeon and Shin
ψ(t) , with:
1 ⎛ t − u ⎞⎟
ψu, s (t) = ⋅ ψ ⎜⎜
⎜⎝ s ⎠⎟⎟
(A.3)
s
299
where ψ(ω) is the Fourier transform of ψ(t) . An analytic wavelet can be
as follows
The sides of the Heisenberg boxes are equal to ∆f/s and s·∆t, with ∆f
between time and frequency resolution since it uses the smallest possible
η ⎛ − (η / G )2 t2 ⎞⎟
1 ⎜
g (t) = 4 ⋅ ⋅ exp ⎜⎜ ⎟⎟⎟
S
(A.6)
π GS ⎜⎝ 2 ⎠⎟
the standard deviation σ of the Gaussian window and the wavelet center
frequency η.
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
s=1/f.
time and in frequency: decreasing the value of the Gabor shaping factor
2003).
i.e. the energy density spectrum and it shows the signal energy with
attenuation.
A.2. Figure A.1(a) shows the strain waveform predicted by ABAQUS for a
specific case a different mother wavelet was used. Figure A.1(b) shows the
field. The arrival times of the two axisimmetric modes are captured by
1
Strain
-1
-2
-3
-4
0 100 200 300 400
Time (µsec)
(b)
-11
x 10
4
(a)
3
2
Strain
-1
-2
0 50 100 150 200 250 300
Time (µsec)
A1
A0
(b)
based on a set of filter banks (Mallat 1989, 1999). The DWT may be
into separate frequency bands (scales). The filtering outputs are then
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.
family
305
1 ⎛t ⎞
ψj, n (t) = ⋅ ψ ⎜⎜ j − n⎟⎟⎟ (A.8)
2j ⎜⎝ 2 ⎠
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
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
1 t
DWHF [n] = ψ( ), φ(t − n) (A.11)
2 2
At the first level, the output of filtering and downsampling steps is a set
closely related but not equal to those of the decomposition tree and are
zero:
f[n]
(a)
⊗ ⊗
DWLF DWHF
Decomposition Level 1
↓2 ↓2
cA1 cD1
⊗ ⊗
DWLF DWHF
Decomposition Level 2
↓2 ↓2
cA2 cD2
⊗ ⊗
DWLF DWHF
Decomposition Level 3
↓2 ↓2
cA3 cD3
cA1 ↑2 DWLF’
+ A1
zeros ↑2 DWHF’
(b)
↑2 DWLF’
zeros
+ D1
cD1 ↑2 DWHF’
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
single signal (no averages) is shown in Figure A.4(b) where the defect
reconstructed from the first six DWT decomposition levels are indicated in
only level 1 highpass filter coefficients, cD1, were used for the
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
fj = ∆ × F / 2j (A.18)
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
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.
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.
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.
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., 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.
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.
Casas, J.R. (1994) “A combined method for measuring cable forces: the
cable-stayed Alamillo bridge, Spain,” Structural Engineering International
4(4), 235-240.
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.
Chase S.B. (2001) “Smarter bridges, why and how?,” Smart Materials
Bulletin 2, 9-13.
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.
Fricker, S. and Vogel, T. (2007) “Site installation and testing of a continuous acoustic
monitoring,” Construction and Building Materials 21(3), 501-510.
Ghosh, T., Kundu, T. and Karpur, P. (1998) “Efficient use of Lamb modes
for detecting defects in large plates,” Ultrasonics 36, 791–801.
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.
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.
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
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.
Kohl, T., Datta, S.K. and Shah A.H. (1992) “Axially symmetric pulse
propagation in semi-infinite hollow cylinders,” AIAA Journal 30, 1617-1624.
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.
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., 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.
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.
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.
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.
Pilarski, A. and Rose, J.L. (1992) “Lamb wave mode selection concepts for
interfacial weakness analysis,” Journal of Nondestructive Evaluation 11,
237-249.
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.
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., Rajana, K.M. and Hansch, M.K.T. (1995) “Ultrasonic guided
waves for NDE of adhesively bonded structures,” Journal of Adhesion 50,
71-82.
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.
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.
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.
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.
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.
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.
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.
Worden K., Manson G. and Fieller, N.R.J. (2000) “Damage detection using
outlier analysis,” Journal of Sound and Vibration 229(3), 647-667.
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.