Академический Документы
Профессиональный Документы
Культура Документы
by
Queen’s University
Kingston, Ontario, Canada
August 2016
Copyright
c Shaddy Samir Zaki Shokralla, 2016
Abstract
i
multi-frequency eddy current systems with hardware limitations, and is generalizable
to other applications where real time acquisition of large data sets is prohibitive.
ii
Co-Authorship
iii
density eddy current non-destructive examination data,” NDT & E International, vol.
62, March 2014, pp. 153-159.
iv
Acknowledgments
I would like to sincerely thank Dr. Thomas W. Krause for his dedication to supporting
this thesis work. Dr. Krause provided steadfast guidance at all critical junctures in
this endeavor and gave both the appropriate academic and industrial context to make
this work impactful in both academic and industrial spheres.
I would also like to thank Dr. Jordan E. Morelli for providing exemplary counseling
and direction to ensure successful completion of doctoral requirements at Queen’s
University. Dr. Morelli’s commitment to rigour ensured manuscripts published as a
product of this thesis were prepared with quality.
Managers of the IMS NDE Projects group at Ontario Power Generation were
instrumental in not only their support of derivation and publication of this work, but
also its field implementation.
My parents, the late Samir Shokralla and Wasima Shokralla instilled an intellec-
tual curiosity in me at a young age, and provided encouragement in pursuit of this
endeavor.
I am most grateful to my loving wife Sally Shehata for her steadfast support and
sacrifice during the period that this work was undertaken. During this period, we
married and were blessed with the birth of our daughter Alexandra. Without Sally’s
continued patience and constancy, this would not have been realized.
v
Contents
Abstract i
Co-Authorship iii
Acknowledgments v
Contents vi
List of Tables ix
List of Figures x
Chapter 1: Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Need for Comprehensive Characterization of Measurement Data Gath-
ered by Gap Probe . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Presence of Pressure Tube Artefacts and Relationship to Gap
Probe Lift-off . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Identifying Redundancies in Multifrequency Eddy Current Data 12
1.2.3 Employment of an Analytical Model of Responses to Factors
Affecting PT to CT Gap Measurement . . . . . . . . . . . . . 14
1.3 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
vi
Chapter 3: Theory 24
3.1 Mathematical Gradient and Directional Derivative . . . . . . . . . . . 25
3.2 Equivalent Circuit and Impedance Plane Representation . . . . . . . 26
3.3 Maxwell’s Equations and the Skin Depth Equation . . . . . . . . . . 30
3.4 Dodd and Deeds Solutions . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5 Principal Components Analysis . . . . . . . . . . . . . . . . . . . . . 38
3.5.1 Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Bibliography 120
viii
List of Tables
ix
List of Figures
1.1 CANDU reactor face with array of fuel channel end-fittings [1]. . . . . 2
1.2 Heat transport system with primary components shown [1]. . . . . . . 3
1.3 CANDU reactor assembly with heat transport sub-assemblies [1]. . . 4
1.4 CANDU reactor assembly with fuel channel components magnified [1]. 5
1.5 Fuel channel components - garter spring spacer separates PT and CT
[1]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 Principle of eddy current testing. Primary magnetic field induces eddy
currents in conductive surface secondary magnetic field that opposes
primary field. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.7 PT to CT gap measurement configuration showing minimum and max-
imum gap locations and eddy current (EC) probe that traverses the
circumference of the PT ID. . . . . . . . . . . . . . . . . . . . . . . . 8
1.8 Method for generation of estimated pressure tube to calandria tube gap. 9
1.9 Eddy current (EC) probe riding over pressure tube surface protrusion
with probe degrees of motion indicated. . . . . . . . . . . . . . . . . . 11
1.10 Impedance plane response of probe lift off from pressure tube surface.
Direction of ˆl gives direction of impedance plane probe lift-off response
starting from contact. . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
x
3.1 Model of a transmit and receive coil with a test object. . . . . . . . . 27
3.2 Equivalent parallel circuit of coil and test sample represented without
and with equivalent impedance. Zp is the equivalent impedance of the
parallel circuit with Np2 Rs and ωL0 components. . . . . . . . . . . . . 28
3.3 Equivalent series circuit of coil and test sample. . . . . . . . . . . . . 29
3.4 Impedance (reactance vs. resistance) diagram with effects of lift-off,
test-object resistivity (ρ) change and test-object equivalent resistance
(Rs ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 δ-function coil between conductor plates k and k 0 . Conductor plates
are enumerated n = 1, 2, . . . , k − 1, k, 10 , 20 , . . . , k 0 − 1, k 0 . . . . . . . . 34
3.6 Coaxial transmitter and detector coils above a series of conductors. . 37
3.7 PCA transformation, reducing the dimensionality of a data set from 3
to 2 (modified from [2]). . . . . . . . . . . . . . . . . . . . . . . . . . 38
xi
5.8 8kHz gy identifying the positions of three constrictions and one spacer. 60
5.9 a) gy computed from 8kHz eddy current data, averaged at the bottom of
the pressure tube. b) dP T −ID computed from ultrasonic measurement
data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.1 Material components in the vicinity of the gap probe, where pressure
tube and calandria tube are approximated as flat plates. . . . . . . . 91
7.2 PCA Scores 1, 2, and 3, plotted together. Analytically modelled data
is clustered in five groups, corresponding to the different modelled re-
sistivity values as labeled. As indicated in Table 7.1, each cluster has
a 0 to 16 mm gap change for three different wall thicknesses. . . . . . 96
xii
7.3 Score 1 peak-to-peak variation (due 0 to 16 mm change in gap) for
model data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
7.4 Data Acquisition Flowchart. . . . . . . . . . . . . . . . . . . . . . . . 99
7.5 Reconstructed 16 kHz ET data on Lissajous plot. . . . . . . . . . . . 101
7.6 Component 1 score with regression surfaces, for variations in PT wall
thickness and PT to CT gap. Experimental data is shown in blue while
analytical model data is shown in yellow. . . . . . . . . . . . . . . . . 105
7.7 Component 1 score with regression surfaces, for variations in PT resis-
tivity and PT to CT gap. Experimental data is shown in blue while
analytical model data is shown in yellow. . . . . . . . . . . . . . . . . 106
xiii
List of Acronyms
AC - Alternating current
AECL - Atomic Energy Canada Limited
CANDU - Canada deuterium uranium
CT - Calandria tube
EC - Eddy current
ET - Eddy current testing
ID - Inner diameter
IMS - Inspection and Maintenance Services
Nb - Niobium
NDE - Non-destructive evaluation
Ni - Nickel
OD - Outer diameter
OPG - Ontario Power Generation
PCA - Principal components analysis
PT - Pressure Tube
Zr - Zirconium
xiv
1
Chapter 1
Introduction
1.1 Background
CANDU (CANada Deuterium Uranium) reactors are pressurized heavy water nuclear
reactors, developed in Canada for electricity generation. Fission reactions are gener-
ated in the reactor core, where uranium bundles serve as fuel. Figure 1.1 depicts the
reactor face, where a lattice of fuel channel end-fittings are shown. These end fittings
are removed prior to uranium fuel bundle deposit.
Figure 1.2 depicts the heat transport system of a CANDU reactor. Fuel bundles
are pushed through the reactor core, and provide the energy source necessary to
keep heavy water surrounding the fuel at elevated temperatures (greater than 250◦
Celsius).
Fuel channels are a critical component of CANDU power plants. They carry the
fuel bundles as they are transported across the core of the reactor. Positioning of
fuel channels relative to other reactor assembly components is shown in Figure 1.3,
while fuel channel components are shown in Figure 1.4. Each fuel channel is made
up of a six metre long pressure tube (which directly houses twelve 0.5 m long fuel
1.1. BACKGROUND 2
Figure 1.1: CANDU reactor face with array of fuel channel end-fittings [1].
bundles) and a surrounding calandria tube. The pressure tube and calandria tube
are separated by four annulus garter spring spacers, which are manufactured from
a square cross-section (either Zr-alloy or Ni-alloy) wire formed into a tight helix.
Further details of these fuel channel components are shown as a magnification in
Figure 1.4, as well as Figure 1.5.
CANDU reactors contain upwards of 400 fuel channels, enclosed within a calan-
dria, a tank containing heavy water neutron moderator (see Figure 1.3). Each fuel
channel consists of a Zr 2.5% Nb pressure tube (PT) with 104 mm inner diameter (ID)
lying within a larger 129 mm ID Zircaloy-2 calandria tube (CT). Nominal resistivities
1.1. BACKGROUND 3
Figure 1.2: Heat transport system with primary components shown [1].
for pressure tubes and calandria tubes are 52 µΩ · cm and 72 µΩ · cm, respectively.
In addition to the fuel bundles, the pressure tube contains heavy water used for
heat transport. Between the PT and the CT is a gas annulus that insulates the hot
(greater than 250◦ Celsius) pressure tube from the colder (50 ◦ C) moderator cooled
CT. Pressure, heat and irradiation induced creep produce diametral creep (increase of
pressure tube diameter and simultaneous reduction of pressure tube wall thickness)
of the PT ID to a maximum of 111 mm and a gradual change in the nominal 4.2
mm wall thickness to an allowable minimum of 3.7 mm [3]. This results in a PT to
CT gap that starts with an average of 8.3 mm, which then decreases with increasing
reactor life. Contact between pressure tube and calandria tube is initially avoided by
1.1. BACKGROUND 4
Figure 1.3: CANDU reactor assembly with heat transport sub-assemblies [1].
4 garter spring spacers that have either a 5.7 mm (older reactors) or 4.8 mm thick-
ness. However sag of the hotter PT within the cooler CT introduces the potential
for contact between PT and CT. Contact for a prolonged period of time introduces
1.1. BACKGROUND 5
Figure 1.4: CANDU reactor assembly with fuel channel components magnified [1].
a risk of hydride blister formation on the PT outer diameter with a consequent risk
of PT cracking [3]. Therefore, PT to CT contact is to be avoided, and monitoring of
PT to CT gap is required [4]. Eddy current testing is employed in monitoring PT to
CT gap [3].
Eddy current testing of conductive materials employs electromagnetic induction
for identification of material properties, including material classification, detection of
flaws, and characterization of defects [5]. Eddy current testing is typically realized
by voltage excitation of a coil, resulting in a time-varying magnetic field that is
produced around the coil. This coil can act as an eddy current test probe [5]. If the
1.1. BACKGROUND 6
Figure 1.5: Fuel channel components - garter spring spacer separates PT and CT [1].
Figure 1.6: Principle of eddy current testing. Primary magnetic field induces eddy
currents in conductive surface secondary magnetic field that opposes pri-
mary field.
measures the gap between PT and CT [3]. An example of the parameters of interest
and the basic measurement configuration are shown in Figure 1.7. The technique
compensates for both wall thickness and PT inner diameter variations using normal
beam ultrasonic measurements. Since eddy current response to change in PT to
CT gap is highly sensitive to PT wall thickness, calibrated eddy current responses,
including changes in PT-CT gap and PT wall thickness are employed. The objective
of the inspection system is to measure the PT-CT gap with sufficient accuracy so that
time-to-contact between PT and CT can first be predicted and second, be avoided
[3].
Figure 1.8 depicts the general format for generating the estimate of gap. Inputs
are the calibrated multi-frequency eddy current responses that account for PT wall
1.2. NEED FOR COMPREHENSIVE CHARACTERIZATION OF
MEASUREMENT DATA GATHERED BY GAP PROBE 8
Figure 1.8: Method for generation of estimated pressure tube to calandria tube gap.
In this section, the mechanics of the gap probe will be introduced, and the relationship
of acquired eddy current signals to pressure tube artefacts (described by depressions
and protrusions on the pressure tube surface) will be discussed.
Although the specifics of the probe geometry and design are proprietary, general
aspects of the probe structure may be considered. The gap probe is secured on a
fuel channel inspection head. Given an inspection head installed in the fuel channel,
transmit and receive coils are aligned in the axial direction. The area of the eddy
current probe body is approximately 50 mm x 25 mm. A spring loads the probe
against the pressure tube surface, such that it maintains contact and its area rides
the pressure tube inner diameter surface. The probe is allowed two degree-of-freedom
motion. The first degree of freedom is away from the pressure tube surface - the
direction of spring force. The second degree of motion is about the inspection head’s
rotary direction of motion. This can be described as forward and backward probe
tilt. See Figure 1.9 for an illustration of gap probe directions of motion.
Interactions with pressure tube artefacts [7, 8, 9], which can be characterized as
1.2. NEED FOR COMPREHENSIVE CHARACTERIZATION OF
MEASUREMENT DATA GATHERED BY GAP PROBE 11
Figure 1.9: Eddy current (EC) probe riding over pressure tube surface protrusion
with probe degrees of motion indicated.
protrusions on the pressure tube surface, will induce lift-off of the gap probe, as
shown in Figure 1.9. The presence of lift-off, can therefore, be used as a marker
of pressure tube artefacts, since a lift-off response will be acquired by the gap mea-
surement system. The identification of coincidence of acquisition of lift-off signals
with the locations of pressure tube artefacts is a prime goal for the application of
a signal processing algorithm to extract the location and features of pressure tube
artefacts. Employment of raw eddy current data gathered by the gap measurement
system, information on probe frequency lift-off response (which is near constant for
a given PT/CT configuration) in addition to locations of acquired eddy current ac-
quisition data can all be used as inputs into an algorithm characterizing pressure
tube artefacts. Figure 1.10 gives the direction of impedance plane probe lift-off re-
sponse. Impedance plane representation of eddy current measurements, including
1.2. NEED FOR COMPREHENSIVE CHARACTERIZATION OF
MEASUREMENT DATA GATHERED BY GAP PROBE 12
Figure 1.10: Impedance plane response of probe lift off from pressure tube surface.
Direction of ˆl gives direction of impedance plane probe lift-off response
starting from contact.
Given ZrNb pressure tube resistivity of 52 µΩ · cm, for 4 kHz, 8 kHz, and 16 kHz
frequencies, the skin depth is 5.7 mm, 4.1 mm, and 2.9 mm, respectively. Despite
differences in penetration depth and sensitivity, there is nonetheless a significant
intersection of useful penetration depth data and responses to lift-off and defects, for
different inspection frequencies (4 kHz, 8 kHz and 16 kHz) employed for PT to CT
gap inspection at Ontario Power Generation.
With respect to employment of multi-frequency eddy current data for NDE appli-
cations, a characterization of the unique (or non-redundant) data gathered by using
multiple frequencies is not presently available in the literature. As multi-frequency
eddy current is used for estimation of PT to CT gap, principal components analysis
(PCA) was investigated to assess what, if any, independent information is gathered
by the gap probe due to changes in critical parameters: pressure tube wall thickness,
and pressure tube to calandria tube gap. This will form an essential component of de-
veloping a comprehensive understanding of relevant inspection information gathered
by the gap probe, as redundancies in information gathered by multifrequency eddy
current measurements can be identified and excluded from a comprehensive model.
PCA involves employing orthogonal transformations to convert a set of observa-
tions, which are correlated, into values belonging to linearly uncorrelated (orthogonal)
variables, called principal components. The main advantage of PCA is that complex
data sets can be reduced to data sets of lower dimension to reveal unapparent and/or
simplified trends in the data [11, 12].
Multi-frequency responses to changes in gap, PT wall thickness, and PT resistiv-
ity can be represented in PCA space, where a smaller number of variables (compared
to the dimension of the original multi-frequency dataset) can be used to investigate
1.2. NEED FOR COMPREHENSIVE CHARACTERIZATION OF
MEASUREMENT DATA GATHERED BY GAP PROBE 14
• effects of instrumentation and setup, which can vary across different instruments
1.2. NEED FOR COMPREHENSIVE CHARACTERIZATION OF
MEASUREMENT DATA GATHERED BY GAP PROBE 15
(e.g. Olympus MS5800 [14], Zetec TC7700 [15]) and experimental setup (e.g.
cable length) employed.
The limitations to employing only experimental data as input into PCA on gap
eddy current data, suggest potential application of an analytical model, which ac-
curately emulates probe response to experimental conditions. Fortunately, variables
which affect gap probe response such as PT wall thickness, PT to CT gap, and PT
resistivity can be modelled via Dodd and Deeds solutions [16, 17, 18], by approx-
imating the pressure tube and calandria tube, in the vicinity of the gap probe, as
parallel plate conductors (introduced in Section 3.4). Data from an analytical model
can be employed towards a PCA analysis, without the complications and limitations
imposed by experimental data. Such an examination will facilitate the development
of methods for extracting parameters of interest from real data.
Physical measurements involving changes in PT wall thickness, PT resistivity and
PT to CT gap can be used to validate PCA computations made using data sets
derived from the analytical model. This will involve using the linear transformations
derived from the PCA computation, on both the analytical model and physical data,
and comparing the transformed data of both data sets.
Finally, using an analytical model (and validating its approach) has benefit to the
inspection qualification process, which is integral to the nuclear industry’s efforts to
qualify its inspection procedures to the regulator in Canada [19]. Showing that it is
possible to analytically describe the effect of variation of an essential parameter (a
parameter whose change in value can affect an inspection procedure’s ability to meet
its intended objectives [20]) can facilitate determination of an inspection system’s
performance as compared to its intended requirements. This can reduce the reliance
1.3. OBJECTIVE 16
1.3 Objective
Chapter 2
Literature Review
First, available works that identify the topology of pressure tube artefacts are dis-
cussed. Localized loading of a pressure tube in the vicinity of spacers can produce
deformation on the pressure tube interior surface at these locations [7, 8, 9] (see
Figure 1.4). Mechanical wear marks are caused by fuel bundles dragging along the
bottom surface of pressure tubes as they are moved along the fuel channel during
refuelling operations. These are typically axially oriented and uniform in depth [7].
Pressure tube diametral creep refers to the increase of pressure tube diameter and
simultaneous reduction of pressure tube wall thickness due to prolonged exposure to
radiation and heat. Pressure tube constrictions located near fuel bundle ends result
2.1. PRESENCE OF PRESSURE TUBE ARTEFACTS AND
RELATIONSHIP TO GAP PROBE LIFT-OFF 19
from relatively less diametral creep (characterized by less PT inner diameter increase
resulting from increased radiation and heat) compared to areas close to fuel bundle
centres. As these pressure tube constrictions have relatively less inner diameter [9],
these locations have a protruded pressure tube inner diameter surface. As shown in
Figure 1.9, the identified artefacts interact with the gap probe so as to cause the gap
probe to tilt about the rotary direction of motion (about the pin) and to retract back
(despite the opposing spring force) towards the inspection head. Surface contact at
two points on the probe body is still maintained but the surface is curved at these
locations resulting in variation of probe lift-off.
One characteristic of acquired gap data is high spatial data density. There have
been many studies, which have employed this feature to extract signals of interest.
The complexity of these techniques has varied. Data can be largely unfiltered [21, 22],
or highly conditioned, employing image and signal processing techniques [23, 24,
25, 26]. A common theme in these works, for highlighting artefacts of interest, is
examination of the data with special consideration of the corresponding physical
phenomena resulting in artefact data markers. Interrogation of gap data should
therefore, involve a search for effects of lift-off, which can mark the presence of pressure
tube artefacts.
There are currently two standard ways eddy current data is examined. The first
is through the Lissajous curve, which identifies resistance and inductive reactance
of acquired eddy current data on an oscilloscope display [5]. The second is through
C-Scan examination. C-Scan data is typically represented by displaying the voltage
component orthogonal to lift-off of 2D eddy current voltage data, at locations for
which eddy current data was collected [27]. An example C-Scan is given in Figure
2.2. IDENTIFYING REDUNDANCIES IN MULTIFREQUENCY
EDDY CURRENT DATA 20
5.4.
Since a desired objective of this work is to correlate acquired eddy current signals
with lift-off indications, it is necessary to identify standard methods for correlating
signals. A common method for examining the correlation between two signals is cross-
correlation [28], while convolution of two signals is another technique [29]. However,
a changing two-dimensional signal (e.g. eddy current represented on a Lissajous plot)
can be evaluated in the direction of lift-off vector using a variant of the directional
derivative (defined in Section 3.1); the dot product is defined for complex valued
signals and can be used to compute the directional derivative for complex valued
signals [30].
In extracting features of interest (e.g. artefact locations), employment of image
processing tools can be utilized. One such tool is the Canny edge detector [31], which
relies on computation of the partial derivative of source data, which has also been
employed by Gonzalez et al. [32]. The partial derivative technique will be examined
as a step in identifying characteristics of probe lift-off, relative to the pressure tube
inner diameter. This is performed in the first manuscript provided in this thesis
(Chapter 5)
particularly facial recognition and satellite image processing algorithms [35, 36]. Prin-
cipal components analysis has been employed on conventional eddy current testing,
to improve reliability of interpretation of eddy current signals for steam generator
tubes [37].
Particular to pulsed eddy current testing (a class of eddy current testing which
examines the transient responses to short voltage spikes applied to eddy current
coils) principal components analysis has been recently employed toward a number
of applications. PCA applied to pulsed eddy current was first shown to provide
enhanced classification of defects [38]. Further applications of PCA to pulsed eddy
current include detection of defects in multilayer aluminum lap joints [39, 40, 41],
steel [42], and aircraft structures [43, 44, 45]. Additionally, PCA was employed for
detection of cracks in multiple layers using pulsed eddy current and giant magneto
resistive sensors [46, 47]. A modified PCA technique was used to detect cracks in an
F/A-18 inner wing spar without wing skin removal [13, 48, 49]. Applied to pulsed
eddy current, PCA has also been combined with wavelet analysis [50] and smooth
nonnegative matrix factorization [51].
Typical employment of principal components analysis for pulsed eddy current ap-
plications considers principal component eigenvectors plotted as a function of time,
in classification of defects [49, 40, 50, 38]. It should be noted however, that conven-
tional eddy current data provides measurement results, while the magnetic circuit is
in steady state, and therefore, conventional eddy current measurement can be con-
sidered time invariant. Principal components analysis can therefore, be employed on
conventional eddy current data, removing the time dependence from interpretation of
results. Also, for conventional eddy current data, principal components analysis can
2.3. EMPLOYMENT OF AN ANALYTICAL MODEL OF EDDY
CURRENT RESPONSES TO FACTORS AFFECTING PT TO CT
GAP MEASUREMENT 22
simplify the data analysis, by relating the reduced score representation of the data to
parameters that produce variations in the signal.
Principal components analysis has not yet been applied to multifrequency eddy
current applications (e.g. [52, 24]), employing conventional eddy current testing. As
will be shown, principal components analysis will be employed to identify indepen-
dence and redundancies in multifrequency eddy current gap data.
Analytic solutions to Maxwell’s equations, which model eddy current probe coils
have been developed by Dodd et al. [16, 17, 18]. Employment of these solutions
in determining independence (and redundancies) in gap probe eddy current data is
especially useful in that these models do not exhibit variable probe frequency-gain
and noise, compared to physical systems, and as such, performing PCA with model
data will provide solutions that are repeatable and free of data corruption.
In order to employ solutions to probe coils using Dodd and Deeds analytical
models, these models have to be validated. Examples of validation of analytical
models of eddy current testing methods are available [53, 54]. An important element
of the validation process is defining physical limitations for which the validation is
correct. This should also be considered in validating Dodd and Deeds solutions [16,
17, 18] for modeling pressure tube to calandria tube gap data.
Using an analytical model for predicting effects of physical parameters such as
pressure tube resistivity, pressure tube wall thickness, and pressure tube to calandria
2.3. EMPLOYMENT OF AN ANALYTICAL MODEL OF EDDY
CURRENT RESPONSES TO FACTORS AFFECTING PT TO CT
GAP MEASUREMENT 23
tube gap has positive side effects with respect to the process of inspection qualifi-
cation. Inspection qualification, which is a nuclear operator regulatory requirement,
[19], serves to determine an inspection system’s capabilities, as compared to its in-
spection specification requirements. Physical parameters whose change in value could
affect an inspection system, such that its specified objectives would no longer be met,
are termed essential parameters [55] and are central in the process of inspection qual-
ification. The effectiveness of modeling has been highlighted for characterizing effects
of essential parameter variation, in the context of inspection qualification [20].
24
Chapter 3
Theory
The concepts presented in this section are further developed in the analysis of eddy
current signals in one of the papers (Manuscript I, Chapter 5) provided in this thesis
and are therefore, introduced only briefly. As laid out in Section 5.4, the directional
derivative is used to compute the projections of change in eddy current signals, in the
direction of the lift-off vector. The axial and rotary components of the directional
derivative are considered separately, in imaging pressure tube artefacts (Section 5.5).
The mathematical gradient is a well defined mathematical operator employed in
many analytical fields of study. The gradient of a function of three variables, f (x, y, z)
where î is the unit vector in the x direction, ĵ is the unit vector in the y direction,
and k̂ is the unit vector in the z direction is defined as [56]
∂f ∂f ∂f
∇f = î + ĵ + k̂. (3.1)
∂x ∂y ∂z
The gradient of a function is a vector, and represents the slope of the tangent of
a function. The direction of the gradient is in the greatest rate of increase of the
function, with respect to variables x, y and z.
The directional derivative of f along a given unit vector û = (ux , uy , uz ) at a point
(x0 , y0 , z0 ) represents the rate of change of f (x0 , y0 , z0 ) with respect to variables x, y, z,
along the direction of û. The directional derivative in the direction û is computed as
Z1 Z2
Zp = (3.4)
Z1 + Z2
3.2. EQUIVALENT CIRCUIT AND IMPEDANCE PLANE
REPRESENTATION 27
Figure 3.1: Model of a transmit and receive coil with a test object.
where Z1 = Np2 Rs and Z2 = jX0 . Np is the number of turns in the transmit coil,
Rs is the variable resistance of the conductive test object, and j is the imaginary
number. Where L0 is the probe inductance, X0 = ωL0 is the inductive reactance of
the coil in proximity to the test object. As Rp is a constant which is not affected by
test object impedance, Rp can be excluded from Zp in deriving an impedance plane
representation of an EC coil in proximity to a test object. Zp can be evaluated as
jNp2 Rs X0
Zp = 2 . (3.5)
Np Rs + jX0
Figure 3.2: Equivalent parallel circuit of coil and test sample represented without and
with equivalent impedance. Zp is the equivalent impedance of the parallel
circuit with Np2 Rs and ωL0 components.
where RL can be used to represent the real part of the equation, and Xp can be used
to represent the imaginary part of the equation, such that [5, 57]
Zp = RL + jXp . (3.7)
Figure 3.4: Impedance (reactance vs. resistance) diagram with effects of lift-off, test-
object resistivity (ρ) change and test-object equivalent resistance (Rs ).
3.3. MAXWELL’S EQUATIONS AND THE SKIN DEPTH
EQUATION 30
In this section a derivation of the skin depth equation is provided, beginning with
Maxwell’s equations. The skin depth equation is discussed in relation to results
presented in Manuscripts I and II (Chapters 5 and 6).
Where E is the electric field, B is the magnetic flux, ρf is the free electric charge
density, µ is the permeability, is the permittivity, and σ is the conductivity, in a
linear medium. Maxwell’s equations have the form [6]
ρf
∇·E = (3.8)
∇·B =0 (3.9)
∂B
∇×E =− (3.10)
∂t
∂E
∇ × B = µσE + µ (3.11)
∂t
Taking the curl of Equation 3.10, substituting into Equation 3.11, and using the
identities ∇ × ∇ × E = ∇(∇ · E) − ∇2 E and ∇(∇ · E) = 0, we obtain, with some
manipulation,
∂E ∂ 2E
∇2 E = µσ + µ 2 . (3.12)
∂t ∂t
For good conductors, where σ is very large (σ/ω 1 [10]), and for a time
harmonic excitation, this can be written as the Diffusion Equation (for frequencies
3.3. MAXWELL’S EQUATIONS AND THE SKIN DEPTH
EQUATION 31
∂E
∇2 E = µσ . (3.13)
∂t
∂J
∇2 J = µσ . (3.14)
∂t
If the current varies sinusoidally in time, in phasor notation, J (t) = Re{J0 e−jωt }
where J0 is the vector having magnitude and direction of the current density, and
substituting into the diffusion equation we have [10]
∇2 J0 + α2 J0 = 0, (3.15)
where α2 = jµσω. Where j is the imaginary number, and since the following mathe-
√
matical identity holds, j = √12 (1 + j),
r
µσω 1+j
α = (1 + j) = , (3.16)
2 δ
r
2
δ= , (3.17)
σµω
or [5]
r
ρ
δ = 50 , (3.18)
µr f
where δ is the skin depth measured in mm, ρ is the resistivity measured in µΩ · cm,
3.4. DODD AND DEEDS SOLUTIONS 32
µr is the relative permeability, and f is the frequency measured in Hz. Equation 3.18
is the Skin Depth Equation, where δ is the depth at which the eddy current density
has decreased to 1/e, or 36.8% of the surface density. In general, penetration depth
decreases with increased frequency, and increases with increased resistivity as per
Equation 3.18. Eddy currents produced at and below the test surface oppose changes
in the primary magnetic field produced by the alternating current in the probe coil,
according to Lenz’s law [6, 10].
The following is a high-level summary of how the eddy current probe response can be
derived analytically, by approximating the test surface(s) as one or more flat plates.
The most significant points, from Ref. [18], are summarized below.
A coordinate system is employed, where the driving current and media are axially
symmetric, reflecting the coil geometry above a conducting plane (Figure 3.5). Due to
the axial symmetry, the current density J and vector potential A have only azimuthal
components, such that
and
where êφ is an azimuthal unit vector. The magnitude of the magnetic vector potential
can be expressed as [18]
3.4. DODD AND DEEDS SOLUTIONS 33
ZZ
A(r, z) = G(r, z; r0 , z 0 )J(r0 , z 0 )dr0 dz 0 , (3.21)
where G(r, z; r0 , z 0 ) is the Green’s function for a δ-function current (having infinite
current density, but total current I) at (r0 , z 0 ), as shown in Figure 3.5. G(r, z; r0 , z 0 )
is also the magnetic vector potential at (r, z) of the δ-function current [16]. A(r, z)
is the summation of the magnetic vector potential, due to a number of δ-coils, which
together approximate real physical coils. The Green’s function satisfies the following
equation, assuming a linear, isotropic, homogeneous medium, having time-harmonic
current angular frequency ω [18],
∂2 ∂2
1 ∂ 1
2
+ − 2 + 2 − jωµσ + ω µ G(r, z; r0 , z 0 ) = −µδ(r−r0 )δ(z−z 0 ) (3.22)
2
∂r r ∂r r ∂z
where µ, , and σ are the permeability, permittivity, and conductivity of the medium,
respectively.
To find the vector potential, given in Equation 3.21, solutions of Equation 3.22
that satisfy proper boundary conditions are used. From the vector potential, all other
electromagnetic quantities of interest can be calculated [18].
The solution of Equation 3.22 for each region is given by [18]
Z ∞
0 0
(n)
Bn (α)e−αn z + Cn (α)eαn z J1 (αr)dα
G (r, z; r , z ) = (3.23)
0
Figure 3.5: δ-function coil between conductor plates k and k 0 . Conductor plates are
enumerated n = 1, 2, . . . , k − 1, k, 10 , 20 , . . . , k 0 − 1, k 0 .
and
1
Bn+1 = eαn+1 zn [(1 + βn+1,n )e−αn zn Bn + (1 − βn+1,n )eαn zn Cn ] (3.26)
2
and
1
Cn+1 = e−αn+1 zn [(1 − βn+1,n )e−αn zn Bn + (1 + βn+1,n )eαn zn Cn ], (3.27)
2
Depicted in Figure 3.6, for the case where a coil is above the conductors (e.g. gap
eddy current probe above PT and CT conductors), and in the coil region (i.e. neither
above nor below the coil), the magnetic vector potential is given as [18]
Z ∞
1 1
A(r, z) = nc I J(r2 , r1 )J1 (αr)e−α0 l1 [(γe−α0 l1 + eα0 z )(1 − e−α0 (z−l1 ) )
2 0 β0 α 0 α
+ (γe−α0 z + eα0 z )(1 − e−α0 (l2 −2) )]dα,
(3.28)
where γ is a constant, which captures the positions and conductivities of the conduc-
tors [18], nc is the turn density, I is the current, and nc I is the current density of a
densely and uniformly wound driving coil. l1 and l2 are positions of the upper and
lower surfaces of the driving coil, respectively, and
Z αr2
J(r2 , r1 ) = xJ1 (x)dx, (3.29)
αr1
where r1 and r2 are the inner and outer radii of the coil, as shown in Figure 3.6.
For each of the conductor configurations, the induced voltage and coil impedance
in a detector (receiver) coil located at (r, z) are computed, where the induced voltage
is given [18]
I
V1 (r, z) = − E · dl = jω2πrA(r, z), (3.30)
For uniformly wound, dense conductors, the total induced voltage is equal to
Z l20 Z r20
V = jω2πnd rA(r, z)drdz, (3.31)
l10 r10
where l10 , l20 , r10 , r20 are the upper and lower surfaces and inner and outer radii, respec-
tively, of the detector coil as shown in Figure 3.6, and nd is its turn density.
Figure 3.6: Coaxial transmitter and detector coils above a series of conductors.
For a detector coil located in the coil region, using Equations 3.28 and 3.31, the
total induced voltage is evaulated [18]
3.5. PRINCIPAL COMPONENTS ANALYSIS 38
Z ∞
1
V =jωIπnc nd 2 3
J(r2 , r1 )J(r20 , r10 )
0 β0 α0 α
0 0 0 0
[2α0 (l20 − l10 ) + (1 − e−α0 (l2 −l1 ) )(−e−α0 l2 (γe−α0 l1 + eα0 l2 )+ (3.32)
0
e−α0 l1 (γe−α0 l1 − eα0 l1 )]dα.
Figure 3.7: PCA transformation, reducing the dimensionality of a data set from 3 to
2 (modified from [2]).
eddy current data sets, and relate effects of physical parameter changes to principal
component scores, for PT to CT gap data.
3.5.1 Process
wT X T Xw
w(1) = arg max , (3.34)
wT w
where arg max is defined as the argument of the maximum, which is the set of points
of the given argument for which the given function attains its maximum value. In
this case w is chosen such that wT X T Xw/wT w attains its maximum value. The kth
component can be found by first subtracting the k − 1 principal components from x:
k−1
X
T
X̂k = X − Xw(s) w(s) . (3.35)
s=1
Now,
3.5. PRINCIPAL COMPONENTS ANALYSIS 40
( )
wT X̂kT X̂k w
ŵ(k) = arg max . (3.36)
wT w
The kth principal component of vector x(i) is given as a score tk(i) = x(i) · w(k) , where
the full decomposition of X is given as
T = XW. (3.37)
Chapter 4
Introduction to Manuscripts
The following journal articles, Manuscripts I, II, and III, are presented as Chapters
5, 6, and 7, respectively.
Chapter 5
Abstract: The pressure tubes (PT) in CANDU (CANada Deuterium Uranium) re-
actors undergo creep induced deformation due to operating pressure, temperature
and radiation conditions. While global deformation of the tube in the form of elonga-
tion and diametral creep are well characterized and monitored by station inspection
systems, local PT deformation and the presence of inner surface artifacts due to wear
are not as directly monitored, but can still provide additional information of fuel
channel condition. A surface profiling technique for monitoring local deformation
and identification of surface wear, using an eddy current probe mounted in a small
(50 mm x 25 mm) planar probe body is presented. The sensitivity of the eddy current
probe to small lift-off variations combined with high density C-Scan information is
used to extract information on smoothly varying local deformation as well as monitor
more significant wear on the inner surface of pressure tubes. Vector separation of
components permits independent identification of axial and circumferential surface
5.1. INTRODUCTION 44
features. Analysis of this data can be used to characterize local PT deformation due
to constrictions at fuel bundle ends and loaded garter spring spacers, as well as iden-
tify areas where shallow mechanical wear has occurred. Examples of the features that
may be identified are presented.
Keywords: surface profiling, eddy current, lift-off, signal processing, data density
5.1 Introduction
C-Scan imaging. Imaged data can either be minimally conditioned [21, 22], or al-
ternatively, a significant degree of signal and image processing can be employed
[23, 24, 25, 26].
For flaw detection applications, examination of lift-off in eddy current data is
normally avoided and is considered a source of noise [5, 62]. The application of paint
and metal deposition thickness measurement is well known. But its potential for
surface profiling is not as well recognized [63]. For the particular circumstance where
a large planar probe body is present, lift-off may be utilized to provide additional
information that would be available if the probe maintains surface following or surface
riding characteristics.
This study shows how high eddy current data-density, basic knowledge of mechani-
cal interaction between the probe and test-piece, and lift-off response can be employed
towards profiling a test-piece surface to a high degree of precision. The technique to
be described in detail is simple to implement, relying on a computation of partial
derivatives with respect to variables whose direction represent orthogonal image axes
(an image processing building block [32, 31]). Ease of technique implementation is
in contrast to more elaborate techniques for analysis of eddy current data, including
neural network implementation and wavelet transformation [25, 64, 65, 66].
The technique will be compared against another currently employed ultrasound
based method for detection of spacer and pressure tube artefact locations, important
activities in ensuring integrity of CANDU nuclear reactor fuel channels.
5.2. INSPECTION REQUIREMENT 46
Fuel channels are a critical component of CANDU power plants. They carry the fuel
bundles as they are transported across the core of the reactor, whose configuration
is shown in the top portion of Figure 5.1. Each fuel channel is made up of a six
metre long pressure tube (which directly houses twelve 0.5m long fuel bundles) and a
surrounding calandria tube. The pressure tube and calandria tube are separated by
four spacers, which are manufactured from a square cross-section Zr-alloy wire formed
into a tight helix. Further details of these Fuel Channel components are shown as a
magnification in Figure 5.1.
Fuel channels are periodically inspected in order to characterize the presence and
status of different degradation mechanisms, thereby meeting their safety and licensing
requirements [4]. This is performed by insertion of an inspection head along the
length of the fuel channel during reactor shutdown periods. A number of different
degradation mechanisms and pressure tube artefacts can be identified by various NDE
5.3. MEASUREMENT TECHNIQUE 47
• Localized deformation due to garter spring spacer loading: Garter spring spac-
ers serve the purpose of ensuring separation between the pressure tube and
relatively cool calandria tube. Loading is therefore induced on the pressure
tube by the spacer and supporting calandria tube. This leaves the pressure
tube slightly ovalized in the vicinity of the spacer.
The eddy current probe is mounted on the inspection head, with transmit and re-
ceive coils situated along the axial direction of the pressure tube, when the inspection
head is installed in-channel. The probe is spring loaded such that it rides the surface
of the pressure tube. In the event that the probe makes contact with an obstruction
embedded in the pressure tube interior, as shown in Figure 5.2, it is also permitted
limited motion with two degrees of freedom: about the inspection head rotary direc-
tion of motion (forward and backward tilt), and away from the pressure tube surface
in the direction of spring force.
Figure 5.2: Eddy current (EC) probe riding over exaggerated pressure tube protru-
sion.
The eddy current signals are generated and received via an Olympus NDT Multi-
scan MS5800 eddy current instrument. Custom software is employed to view acquired
signals in real time as well as post-acquisition. Eddy current signals are acquired on
multiple frequencies, which include 8kHz and 16kHz. Unadjusted raw measurements
have units mV.
5.4. LIFT-OFF LOCATION EXTRACTION 49
We represent the acquired eddy current data set as a complex valued function f (x, y)
mapping rotational and axial values (x and y) to the complex plane, C. We use C
to represent the domain where eddy current signals lie. Orthogonal components of
C represent the resistance and inductive reactance components in the eddy current
impedance plane. It is worthy to note that axially adjacent values of f (x, y) (along
the y-axis) are collected 360 degrees apart, due to the helical eddy current probe
trajectory; data is collected helically at 1 rotary degree (approximately 0.9mm) per
sample.
A retraction signal is sent to the eddy current probe during its acquisition sequence
and a lift-off signal (which can be represented as a horizontally oriented vector on the
Lissajous curve) is obtained. The direction of the lift-off vector is a key component
used in the calculations and is employed to allow for surface profiling using eddy
current data. ˆl is used to denote the complex valued unit vector lying in C with the
same direction as the direction of the lift-off response.
Since values of f (x, y) lie on the complex plane, C, f (x, y) can be written
and
5.4. LIFT-OFF LOCATION EXTRACTION 50
Where the dot product is defined for complex valued vectors [30], the scalars
gx (x, y) and gy (x, y) are defined as the components of ∂f (x, y)/∂x and ∂f (x, y)/∂y
in the direction of lift off, such that
∂f (x, y) ˆ
gx (x, y) = · l, (5.4)
∂x
and
∂f (x, y) ˆ
gy (x, y) = · l. (5.5)
∂y
gx (x, y) and gy (x, y) represent the change in eddy current data, in the rotary and
axial directions, respectively, in the direction of lift-off.
Examination of gx and gy reveal not only areas in the inspection space where
the eddy current probe has lifted off from the pressure tube, but also reveal key
mechanical interactions between the eddy current probe and the inspection surface.
gx represents rotary lift-off experienced by the probe. That is to say, changes in lift-off
experienced by the probe as it traverses in the rotary direction over pressure tube
obstructions extending along the axial direction of the pressure tube.
Consider Figure 5.2. The eddy current probe is not free to rotate about the axial
direction of travel. So, the physical motion of the probe as it traverses over axially
extending pressure tube obstructions is away from the surface of the pressure tube in
the direction of spring force, and is limited to one degree of freedom.
gy on the other hand, represents changes in axial lift off experienced by the probe
5.5. APPLICATIONS 51
as it traverses in the axial direction over pressure tube obstructions away from the
surface of the pressure tube. The obstructions extend in the rotary direction of the
pressure tube. Unlike the case for lift-off of the probe as it traverses in the rotary
component of travel, (again consider Figure 5.2) the probe is free to move both away
from the pressure tube wall as well as rotate (tilt) about the rotary direction of
travel. Probe motion is therefore capable of two degrees of freedom when lifting off
over rotationally extended pressure tube obstructions.
5.5 Applications
In this section a number of pressure tube artefacts are identified using the computa-
tional method introduced in the previous section.
Calculation of gx for a set of eddy current data obtained in-reactor determines probe
lift-off variation in its rotary component of travel. The direction of probe lift-off is
normal to the pressure tube wall.
Vertical stripes are evident in Figure 5.3, revealing a systematic phenomenon,
which possesses a lift-off component in the rotary direction of travel at the bottom of
the pressure tube (near 180 degrees) for a significant axial length along the pressure
tube (800mm). The vertical stripes represent pressure tube mechanical wear marks
that have been caused by fuel bundles sliding along the bottom of the pressure tube.
The depth of these mechanical wear marks is typically less than 30µm [67]; as such,
computation of gx for the 16kHz frequency has been used to identify features in the
pressure tube surface not evident in standard eddy current Lissajous representations.
5.5. APPLICATIONS 52
Figure 5.3: 16kHz gx revealing fuel bundle mechanical wear marks with nominal depth
50µm.
Figure 5.4 shows a conventional C-Scan for the same eddy current data, repre-
sented in Figure 5.3. Vertical stripes representing positions of mechanical wear marks
are evident in the top and middle of the figure (representing lower axial positions
and rotary positions at the bottom of the pressure tube). However, it is clear that
mechanical wear marks, given in the Figure 5.4 C-Scan, are much less obvious than
those given by gx in Figure 5.3. Furthermore, it is difficult using the C-Scan alone to
differentiate between vertical stripes in the middle of the figure, representing location
of mechanical wear marks, versus vertical stripes at the sides of the figure (corre-
sponding with the top of the pressure tube), where no mechanical wear marks are
present.
5.5. APPLICATIONS 53
Figure 5.4: 16kHz C-Scan, plotting eddy current data orthogonal to direction of lift-
off vector, ˆl.
Figure 5.5 provides 16kHz C-Scan eddy current data, in the direction of lift-
off. Mechanical wear marks are more evident in this figure compared to Figure 5.4.
gx given in Figure 5.3 however still provides increased discrimination of wear mark
features. An increased number of vertical stripes revealing locations of wear marks
are apparent in Figure 5.3 over Figure 5.5. Furthermore, whereas mechanical wear
marks are not evident in axial positions between 5950mm and 6000mm in Figure 5.5,
these wear marks are visible for the same axial positions in Figure 5.3.
The SNR (signal-to-noise ratio) of mechanical wear mark indications is used to
quantitatively compare imaging capability of gx , against C-Scan representations. The
SNR for this application is defined as the rotational peak-to-peak mechanical wear
mark signal amplitude, averaged over all axial positions, divided by the background
5.5. APPLICATIONS 54
Figure 5.5: 16kHz C-Scan, plotting eddy current data in direction of lift-off vector, ˆl.
noise, which is computed as the standard deviation of noise pixel intensities. The
SNR of mechanical wear mark indications imaged via computation of gx , shown in
Figure 5.3, is computed as 5.8, while the SNRs of mechanical wear mark indications
shown in C-Scan Figures 5.4 and 5.5 are computed as 1.5 and 1.6, respectively. This
demonstrates increased wear-mark imaging capability of gx , over C-Scan representa-
tions.
Figure 5.6 represents the computation of gy for the same eddy current data set em-
ployed to calculate gx , plotted in Figure 5.3.
Large horizontal bands, which extend 360 degrees circumferentially around the
pressure tube (evident at axial positions 5500mm and 6050mm) are shown in Figure
5.5. APPLICATIONS 55
5.6. These large horizontal bands represent the locations of fuel channel constrictions.
The rotary extent (360 degrees around the pressure tube) of the bands is consistent
with the circular formation of constrictions. Furthermore, the bands are approxi-
mately 0.5m apart, consistent with the length of fuel bundles where constrictions
form (Figure 5.1).
It is worthy to note that Figures 5.4 and 5.5 give C-Scan displays for the same
eddy current data revealing constrictions in Figure 5.6. Indications of constrictions
in Figures 5.4 and 5.5 coincide, and are inseparable from indications of mechanical
wear marks, contrary to indications of constrictions in Figure 5.6.
Figure 5.6: 16kHz gy revealing locations of local pressure tube constrictions (horizon-
tal stripes extending from 0◦ to 360◦ ).
5.5. APPLICATIONS 56
5.5.3 Spacers
The inspection head used to collect eddy current data also collects ultrasonic non-
destructive examination (NDE) data. Ultrasonic NDE data can be used to confirm
position of spacers and constrictions identified through computation of gy .
Two ultrasonic probes diametrically opposed, and positioned on the inspection
5.5. APPLICATIONS 58
head, rotate at the same pitch as the eddy current probe. This allows for measure-
ments to be taken at 0 degree and 180 degree rotary positions, for all axial positions
along the pressure tube. The distance, d, between a probe face and the pressure tube
wall can be determined by the simple relationship
ct
d= , (5.6)
2
where c is the speed of sound in water, and t is the travel time between ultrasound
pulse emission from, and return to, the ultrasonic probe. The pressure tube inner
diameter along the 0 degree and 180 degree vertical line can then be computed
where d0 is the distance between the probe positioned at the 0 degree rotary position
and the pressure tube inner diameter wall, d180 is the distance between the probe
positioned at the 180 degree rotary position and the pressure tube inner diameter
wall, and dl is the distance between diametrically opposed probe faces.
gy for 8kHz eddy current data is shown in Figure 5.8, revealing three constriction
locations and a single spacer location. To confirm the position of these artefacts,
a comparison between locations identified through examination of gy , and locations
identified through examination of ultrasonic measurements (dP T −ID ) is made. A
correction factor is introduced for axial positions of dP T −ID . This accounts for the
difference between axial position of ultrasonic probe centres and axial position centred
between transmitting and receiving eddy current coils.
Figure 5.9 a) gives gy along axial positions between 3600mm and 5000mm, but
5.5. APPLICATIONS 59
averaged for points centred at the bottom of the channel; the average is calculated for
values of gy between rotary positions 160 degrees and 200 degrees. The points where
gy crosses the x-axis in the vicinity of visible signal perturbations (each characterized
by a large positive, then negative fluctuation) marks the position of spacers and
constrictions. These points lie on x-axis positions: 3695mm, 4195mm, 4478mm, and
4698mm.
Figure 5.9 b) gives ultrasonic measurements (dP T −ID ), along pressure tube axial
positions between 3600mm and 5000mm. Background variation in pressure tube inner
diameter has been removed. Positions of constrictions and spacers can be identified
by localized reductions in dP T −ID , plotted as a function of axial position. For localized
reductions, the pressure tube inner diameter is relatively smaller as compared to the
immediately surrounding regions, reflecting the characteristic protrusions at spacers
and constrictions.
The positions of lowest points in localized reductions can be used as markers for
the axial locations of spacers and constrictions, where axial locations are identified
as 3694mm, 4190mm, 4474mm, and 4702mm. Localized diameter changes can also
be estimated from Figure 5.9 b), where the first, second, and third constrictions have
associated diameter reductions 60µm, 70µm, and 90µm, respectively, while the spacer
has an associated diameter reduction of 170µm.
Comparing spacer and constriction positions determined through examination of
gy , and through ultrasonic measurement reveals consistency in artefact localization
between the two methods; measurements agree within +/- 5mm.
5.6. DISCUSSION 60
Figure 5.8: 8kHz gy identifying the positions of three constrictions and one spacer.
5.6 Discussion
Eddy current data is collected rotationally every 1 degree (∼ 0.9mm) and axially
every 1mm. High data-density (typically output in C-Scan display for eddy current
applications), has been exploited in the surface profiling process.
It is worthy to note that images formed by the computation of gx (and gy ) differ
from C-Scan displays due to isolation of change in eddy current data, in the direction
of lift-off. C-Scan displays reveal the component of densely collected eddy current
data in one vector direction, usually orthogonal to lift-off. Images formed through
computation of gx and gy not only differ from C-Scan data by calculation of the
component of eddy current data parallel (not orthogonal) to the lift-off direction, but
5.6. DISCUSSION 61
make use of a maximum rate of change calculation, of eddy current signals in the
direction of lift-off (Equations 5.4 and 5.5). This key feature allows the visualization
of change in acquired eddy current data in the direction of lift-off. Without this,
features such as those visible in Figure 5.6, identifying position of constrictions would
be difficult to identify. Indeed, the position of constrictions cannot be identified from
C-Scan images (Figures 5.4 and 5.5) where the same eddy current data is employed
in generation of Figure 5.6.
Although eddy current data is gathered in a helical trajectory by the eddy current
probe mounted on the fuel channel inspection head, changes in probe lift-off due to
mechanical interaction between the eddy current probe and pressure tube protrusions
are decomposed into rotary and axial directions. Indications visible in figures gen-
erated via computation of gx and gy represent probe lift-off in the rotary and axial
directions, respectively. gx can therefore be used to detect axially extending obstruc-
tions, and is robust against noise due to lift-off in the axial direction of probe motion.
Conversely, gy can be be used to detect rotationally extending obstructions and is ro-
bust against noise due to lift-off in the rotary direction of probe motion. As is shown
in Figures 5.3 and 5.6, rotationally extending obstructions (e.g. constrictions) do not
appear in the figure generated through computation of gx , while axially extending
obstructions (e.g. mechanical wear marks) do not appear in the figure generated by
gy . This demonstrates the high degree of separability between localization of probe
lift-off in rotary and axial directions, through computation of gx and gy .
5.7. CONCLUSIONS 62
5.7 Conclusions
Surface profiling of CANDU fuel channel pressure tube artefacts has been realized
through the methods laid out in this study. Mechanical wear marks, constrictions,
and local pressure tube deformation due to spacer loading can be identified through a
simple computational manipulation of high density eddy current data. These artefacts
range in size; mechanical wear marks have depressions ranging from 50µm to 100µm,
constrictions have protrusions in the same range, while garter spring spacers have
5.7. CONCLUSIONS 63
Acknowledgements
The authors would like to thank, from Ontario Power Generation: Tulchand Harduwar
for supporting this work; as well as Andrew Hong and Jerry Piskorski for facilitating
access to source material.
5.7. CONCLUSIONS 64
Figure 5.9: a) gy computed from 8kHz eddy current data, averaged at the bottom
of the pressure tube. b) dP T −ID computed from ultrasonic measurement
data.
65
Chapter 6
6.1 Introduction
CANDU reactors contain over 400 fuel channels, each consisting of a 6 m long Zr
2.5% Nb pressure tube (PT) having 104 mm inner diameter (ID) concentric within a
129 mm ID Zircaloy-2 calandria tube (CT). Fuel bundles and heavy water required
for heat transport are contained within the PT. An annulus gas lies between the PT
and CT and insulates the hot ( 300 ◦ C) pressure tube from the ( 50 ◦ C) moderator
cooled CT. Diametral creep of the PT ID is induced by pressure, heat, and irradiation
induced creep, to a maximum ID of 111 mm, and a progressive decrease from nominal
4.2 mm wall thickness to an allowable minimum of 3.7 mm. PT to CT gap therefore
starts with an average of 8.3 mm and decreases with aging of the reactor. Potential for
contact between PT and CT is introduced by sag of the hotter PT within the cooler
CT. Hydride blister formation on the outer PT diameter (with risk of PT cracking) is
introduced with prolonged contact and PT to CT contact is therefore, to be avoided.
An eddy current based technique (complemented by ultrasonic PT wall thickness
measurements), which measures PT to CT gap, is used to monitor separation of PT
to CT. Figure 6.1 shows parameters of interest as well as the simplified measurement
configuration.
Qualification of an inspection system is an important step in determination of
the system’s capabilities against its inspection specification requirements, and has
become a nuclear operator regulator requirement [19]. Integral to the qualification
process is the study and determination of the effects of variation of an inspection
system’s essential parameters, those parameters whose change in value could affect
an inspection system such that its specified objectives would no longer be met [55].
Modelling is an important tool that has been promoted in the study of effects of
6.2. MEASUREMENT TECHNIQUE 67
essential parameter variation [20]. Modelling can be used to evaluate the effect of
specific parameter variations on gap probe response and therefore, is a tool for the
evaluation of gap measurement accuracy. However, models need to be validated in
order to provide assurance that they are providing the correct results. Examples doc-
umenting the validation of mathematical models for eddy current testing techniques
can be found in the literature [53, 54].
Mathematical models for realistic eddy current probe coils based on analytic so-
lutions to Maxwell’s equations were developed by Dodd, Deeds and other colleagues
[18, 16, 17]. A software tool which emulates Dodd and Deeds equations (expressing
analytical solutions to eddy current problems containing stratified conductors) has
been employed to determine the eddy current response to changes in essential param-
eters affecting accuracy of the PT to CT gap system. In this model the pressure tube
and calandria tube approximate parallel plate conductors for the eddy current probe
near the surface.
This paper documents validation performed to quantify the agreement between
modeled and physical responses to variation of essential parameters. Validation of
modelling effects of essential parameter variation on eddy current response was im-
plemented by comparing model output against physical measurements where PT to
CT gap, PT resistivity, and PT wall thickness were varied. Validation of the model
was critical to legitimizing its use in the inspection qualification process.
A transmit-receive eddy current probe acquires eddy current signals in a helical tra-
jectory. This probe is mounted on an inspection head, which is driven axially and
6.2. MEASUREMENT TECHNIQUE 68
Figure 6.2: Geometries in close proximity to the gap probe. Pressure tube and Ca-
landria tube modelled as flat plates.
6.2. MEASUREMENT TECHNIQUE 69
rotationally along the pressure tube. The probe’s helical travel is approximated by a
series of circular trajectories, due to its axial pitch being small (1 mm), compared to
its rotational travel.
Transmit and receive coils belonging to the eddy current probe are situated along
the axial direction of the pressure tube when the inspection head is installed in-
channel. The probe is spring loaded such that it rides the surface of the pressure tube
[68]. The surface riding characteristic of the probe minimizes lift-off variations, which
would otherwise introduce additional variability in the probe response [5]. In addition,
deviations of lift-off from the ideal surface-riding condition, which could compromise
calibrated response of the probe to changes in gap, can be monitored using a gradient
response analysis as outlined in Ref. [68], helping to identify conditions where lift-off
variations may have compromised calibrated gap probe response. The center-to-center
spacing of the coil pair of 11 mm is much smaller than the nominal 330 mm inner
circumference of the PT. A schematic representation of coil positioning in relation to
other components is provided in Figure 6.2.
An Olympus NDT Multi-scan MS5800 eddy current instrument is used to induce
eddy currents and measure pickup coil response, while acquired signals are viewed in
real time. Multiple frequencies, including 4kHz, 8kHz, and 16kHz are used to acquire
eddy current signals. Unadjusted raw measurements have units mV .
Measurement of pressure tube to calandria tube gap is primarily a calandria tube
lift-off measurement, obfuscated by the presence of the pressure tube [69, 70]. Vari-
ations in eddy current probe response are monotonic with variation in the distance
between the probe and the calandria tube.
Calibration data is generated from measurement of multiple concentric pressure
6.2. MEASUREMENT TECHNIQUE 70
tube to calandria tube segments, where for each pressure tube to calandria tube com-
bination, pressure tube to calandria tube gap is varied from the expected minimum
to maximum. The concentric segments differ in pressure tube wall thickness, where
beginning of life (larger) to end of life (smaller) wall thicknesses are employed. Cal-
ibration data, along with ultrasonic measurement of in-channel pressure tube wall
thickness, are used to map eddy current data collected in-channel for each frequency
to in-channel pressure tube to calandria tube gap values [70].
Responses are attenuated for larger pressure tube wall thicknesses. The basic skin
depth equation describes shielding of a uniform electromagnetic field source near a
thick conductor, where the conducting domain is a homogeneous half-space and the
uniform magnetic flux density is directed tangentially to the surface of the material
[71, 5]:
J
= exp − x .
n o
J0 (6.1)
δ
Here, J is the eddy current density at distance x into the conductor, J0 is the surface
eddy current density and δ is the electromagnetic skin depth of penetration into the
conductor given by
r
ρ
δ = 50 . (6.2)
f µr
At δ, the eddy current density has been reduced to 1/e (36.8%) of the surface eddy
current density, J0 [5]. Equation 6.2 reveals that the effect of pressure tube resistivity
variation on eddy current voltage response for each frequency employed in pressure
6.2. MEASUREMENT TECHNIQUE 71
Carefully produced samples were used to measure responses to change in gap, wall
thickness, and resistivity. One key sample consisted of a singular pressure tube, fixed
within a calandria tube, where PT to CT gap between the pressure tube and calandria
tube varied between 0.8mm and 16mm. This sample also allowed for measurement of
responses to pressure tube wall thickness changes, as two of the pressure tube sections
were machined from their inner diameters, reducing their respective wall thicknesses.
The resulting pressure tube is made up of three sections with wall thickness 3.48 mm,
3.84 mm, and 4.36 mm (with a constant resistivity of ∼53.8 µΩ · cm at 20◦ C).
A set of samples was also produced to measure the response to changes in re-
sistivity and gap. These samples consisted of two composite pressure tubes. Each
composite pressure tube was made up of two pressure tube sections that had been cut,
machined, and welded together so that both could be fixed within the calandria tube.
Variations in composite tube wall thickness were small (a few microns) such that the
voltage response due to resistivity could be isolated, from potential wall thickness ef-
fects. The first composite pressure tube was made up of medium and high resistivity
sections (55.0 µΩ · cm and 57.4 µΩ · cm, respectively) while the second composite
pressure tube was made up of low and high resistivity sections (52.7 µΩ · cm and 57.2
µΩ · cm, respectively). The first composite pressure tube is shown in Figure 6.3.
6.3. ANALYTIC DETERMINATION OF EDDY CURRENT
RESPONSES TO CHANGES IN PHYSICAL PARAMETERS 72
Figure 6.3: Composite pressure tube and associated test rig assembly.
that from the receive coil of the real transmit-receive gap probe, but reduced by the
appropriate ratio of areas, namely that of the gap probe receive coil to that of the
annulus between the differential detectors.
A 2D model was assumed to be sufficient, since changes in gap are in the radial
direction with respect to the PT, and axial and circumferential directions can be
treated as planar to a first approximation, due to the relative localization of induced
eddy currents. As indicated in the introduction, the nominal ID for as-installed PT
is 104 mm and at end-of-life ID is 111 mm. The inner circumference of the PT is
therefore, 327 to 349 mm, as compared to the 12 mm x 11 mm axially oriented probe
sensing area of the transmit-receive coil pair, thereby justifying first-order treatment
by a planar geometry. It also neglects the potential effects of PT and CT curvature,
not considered a factor due to field spread on a 52 mm PT ID radius, compared
to the probe, which only has an 11 mm spacing. Furthermore, the mathematical
model assumes that the electromagnetic properties (electrical resistivity and magnetic
permeability) of the materials (PT, CT and Shield) are linear, homogeneous and
isotropic. Effects of any anisotropy in conductive material properties are ignored.
The analytical model based on Dodd and Deeds equations was employed to model
eddy current response to a number of varying PT and CT conditions. The modelling
program was applied to determine the voltage developed in the receive coil of the
probe at frequencies 4, 8, and 16 kHz. Modelling conditions that were employed in the
simulation and subsequent validation are given in Table 6.1, where all permutations
of given parameters were modeled. The resistivities (varying between 45 µΩ · cm and
6.4. VALIDATION OF ANALYTICALLY DETERMINED
RESPONSES 74
60 µΩ · cm) bound resistivities for non-irradiated PTs [5, 74]. Similarly, the range of
wall thickness values (3.48 mm and 4.38 mm) employed in modelling was chosen to
emulate the range of wall thicknesses of in-service pressure tubes.
Parameter Values
0, 0.5, 1.0, 1.5, 2.0
PT to CT Gap (mm) 2.5, 3.0, 4.0, 5.0, 6.0
7.0, 8.0, 9.0, 11.0, 16.0
PT Wall Thickness (mm) 3.48, 3.76, 4.36
PT Resistivity (µΩ · cm) 45, 49, 52.4, 56.9, 60
Figure 6.4: Impedance plane display of measured response to change in gap of 0.6
to 16.3 mm is shown by circles (
), while scaled modeled response to
change in gap of 0 to 16 mm is shown by squares ().
mm wall thickness. A comparison of the modeled and measured amplitudes for the
different frequency signals, and an increase in wall thickness is shown in Figure 6.5.
Evident in examination of Figure 6.5 are the monotonic relationships between eddy
current voltage, wall thickness and gap; eddy current voltage magnitude increases
with gap, while it decreases with wall thickness.
To correct for a small mismatch in PT wall thickness between modelled responses
at 3.76 mm and measured responses at 3.84 mm, modelled responses at 3.84 mm
were generated via interpolation. For each gap value (listed in Table 6.1) of 1.0 mm
and greater, a cubic spline interpolation [75] was used to output voltage response at
pressure tube wall thickness of 3.84 mm using modeled responses at 3.48 mm, 3.76
6.4. VALIDATION OF ANALYTICALLY DETERMINED
RESPONSES 76
|Vmeasured |
Agreement = × 100%. (6.3)
|Vmodelled |
The maximum effect of wall thickness increase from 3.48mm to 4.36mm on eddy
current amplitude (at 16mm gap), the maximum measured response (%), for the
different frequencies is also identified in Table 6.2.
Table 6.2: Statistical Computations for 4 kHz, 8 kHz, and 16 kHz Frequencies.
Figure 6.5: Effect of wall thickness variation on modeled versus measured eddy cur-
rent amplitudes for a) 4kHz, b) 8kHz, and c) 16kHz frequencies.
6.4. VALIDATION OF ANALYTICALLY DETERMINED
RESPONSES 78
only evident in comparing responses to bulk wall thickness changes (e.g. responses
between 3.48 mm, 3.84 mm, and 4.36 mm sections), but also responses to wall thick-
ness changes within individual sections where wall thickness was machined to a near
constant value. Deviations from a smooth monotonic curve representing the mea-
sured relationship between absolute voltage and gap are a result of small changes in
wall thickness about the circumference of the pressure tube. These deviations are
identified at the same gap values across Figures 6.5 a) to c), but are more prominent
for higher frequency responses, particularly at 16kHz.
Table 6.3: Resistivity and Wall Thickness Parameters for Measured Composite Re-
sistivity Pressure Tubes.
PT Change in
Number of Average Wall
Resistivity at Wall
Pressure Tube Measure- Thickness
20 ◦ C Thickness at
Section ments (mm)
(µΩ · cm) 0 ◦ and 180 ◦
4.03, 4.0, 0.0, 0.0, 0.0,
Medium ρ in 4 55.0
4.01, 4.015 -0.01
Med-High Tube
4.08, 4.08,
High ρ in Med- 4 57.4 0.0, 0.0, 0.02
4.08
High Tube
Low ρ in Low- 2 52.7 4.055, 4.05 0.01, 0.01
High Tube
High ρ in Low- 2 57.2 4.056, 4.055 0.0, -0.01
High Tube
p
Amplitude = (x0 − x1 )2 + (y0 − y1 )2 (6.4)
where 0.6mm gap occurs at (x0 , y0 ) and 16.3mm gap at (x1 , y1 ). The measured am-
plitude is used to scale the modeled gap probe response using one standard resistivity
pressure tube. Recorded horizontal and vertical voltages are in units of mV .
Figure 6.6 shows the measured amplitudes for 0.4mm to 16.3 mm change in gap,
obtained from the probe at 4 kHz for the various resistivity sections of the Low-
High and Med-High composite tubes, detailed in Table 6.3. Modeled results for an
equivalent change in gap at the same frequency, coil spacing and wall thickness have
been scaled to the experimental results at 52.7 µΩ · cm, while model results at other
resistivities have been scaled by the same amount. Table 6.4 summarizes standard
deviation, σ, between scaled model and experimental results and σ normalized by
gap amplitude, presented as a percentage. Excellent agreement between the two sets
6.4. VALIDATION OF ANALYTICALLY DETERMINED
RESPONSES 80
v
u N
uX (f (ρi ) − Ai )2
σ=t (6.5)
i=1
N −1
Ratio of
Composite Amp. vs ρ Amp.
Frequency σ σ/A Amplitude rate at 54 variation
(kHz) (mV) (%) Scaled to 4 µΩ · cm with ρ to wall
KHz (mV /µΩ · cm) thickness
(µΩ·cm/mm)
4 29.5 0.9 1 51.7 25
8 28.5 1.1 1.2 85.0 17
16 26.4 2.6 3.3 121 5.7
6.5. DISCUSSION 81
Figure 6.7 shows the gap probe response rate of change with resistivity as a func-
tion of frequency, obtained from the slope at 55 µΩ · cm in the scaled model curve
in Figure 6.6 as well as slopes for 8kHz and 16kHz frequencies. Slope values are
shown in the second last column of Table 6.4. These results indicate that the effects
of resistivity on gap response are greater at higher frequencies, again consistent with
Equation 6.2.
The last column in Table 6.4 shows the ratio of rate of change of gap amplitude
with resistivity relative to that of wall thickness as obtained from model results. The
values show that the effect of resistivity, relative to wall thickness, on gap measure-
ment decreases with increasing frequency. The largest ratio arises at 4 kHz frequency
(25 µΩ · cm/mm) indicating that this frequency is the most impacted by changes in
resistivity when compared with response due to PT wall thickness variations. The
next largest, with similar order of magnitude (17 µΩ · cm/mm), is the 8 kHz ampli-
tude response. These results are consistent with the relative % error, presented in
the second column of Table 6.4. The relative % error shows a progressive increase
in % standard deviation with increasing frequency that results in increased scatter,
attributed to the small wall thickness variations in this tube.
6.5 Discussion
Although experimental results have been employed to show agreement with a discrete
number of modeled results, an advantage to modelling effect of parameter variation
is the ability to model all values within a range of an essential parameter (or ranges
of essential parameters). This is obvious, considering voltage responses are through
6.5. DISCUSSION 82
Figure 6.6: Amplitude as a function of resistivity for (a) 4kHz, (b) 8kHz, (c) 16kHz
frequencies for scaled model and experimental data from composite tubes.
6.5. DISCUSSION 83
Figure 6.7: Variation of scaled model based on gap probe response amplitude (0.5
to 16 mm gap) with resistivity at 55 µΩ · cm plotted as a function of
frequency.
Dodds and Deeds equations [18, 16, 17], which are analytic and closed form. However,
this presents clear advantages to examining effects of essential parameters in the
context of inspection qualification. Effects of essential parameter variation can be
observed to an arbitrary level of specificity, which can be performed rapidly and
inexpensively, compared to performing a large number of experiments.
In the model results presented here, the relative sensitivity of the eddy current
based gap response to bounding pressure tube wall thickness and resistivity variations
has been studied. The model results have assisted in the identification of frequencies
most sensitive to either the effects of wall thickness or resistivity, relative to the desired
PT to CT gap response. This information can be applied to select the most robust
frequency, that which is most independent of the essential parameter variations, for
in-reactor gap measurement.
6.6. CONCLUSIONS 84
Voltage responses to wall thickness and gap have been modeled by Dodds and Deeds
equations [18, 16, 17], where the pressure tube and calandria tube geometries (relative
to the eddy current probe) have been approximated as parallel conductive plates. For
the probe configuration employed in this study, this has shown not to be a significant
source of error. However, if select probe parameters are modified, the parallel plate
approximation may cease to approximate pressure tube to calandria tube geometries
to sufficient accuracy. If, for example, frequency is lowered well below 4kHz, there is
potential for increased eddy current penetration for larger PT and CT surface areas,
which, relative to the probe surface, no longer approximate parallel plates. Similarly,
if the probe was situated away from the surface of the pressure tube, curvature of PT
and CT components relative to the probe surface would discount their approximation
as paraellel plates.
6.6 Conclusions
Validation of a model based on Dodds and Deeds equations [18, 16, 17] used to es-
timate eddy current voltage response to PT wall thickness and resistivity changes
(with varying PT to CT gap) has been performed via comparison with physical mea-
surements.
For wall thickness values between 3.48mm and 4.36mm, and gap values between
1mm and 16mm, average agreement exceeded 95 % for all frequencies used, despite
small machining deviations (microns) in the wall thickness of test pieces, which re-
sulted in an exaggeration of disagreement between measured and modeled results.
Standard deviation of the agreement ranged between 7.9 % (4kHz signal) and 18.5
6.6. CONCLUSIONS 85
Acknowledgements
The authors wish to thank: Ontario Power Generation management for supporting
this work; as well as Stuart Craig from Atomic Energy of Canada Limited and John
Sedo for valuable discussions.
87
Chapter 7
7.1 Introduction
reveal unapparent and/or simplified trends in the data [11, 12]. Particular to conven-
tional eddy current testing, principal components analysis (PCA) has been employed
to increase reliable interpretation of steam generator tube signals [37]. PCA has also
been applied to pulsed eddy current, and was first shown to provide enhanced classi-
fication of defects [38]. Further applications of PCA to pulsed eddy current include
detection of defects in multilayer aluminum lap joints [39, 40, 49], steel [42], and
aircraft structures [43, 44, 45]. Independent component analysis (ICA), a technique
used to separate multivariate signals into independent non-gaussian signals, has also
been used on eddy current data towards flaw characterization [78]. Integration of ICA
and PCA has been used for defect detection and separation for eddy current pulsed
thermography video [79].
Inspection speed is a critical factor in non destructive examination of nuclear com-
ponents, such as steam generator tubes [80], since shortened inspections will reduce
facility downtime. Techniques which have been employed to increase eddy current
inspection speed of steam generator tubes, have involved multiplexing, simultaneous
injection, or employment of eddy current arrays [81, 82, 83]. This is the first work
to apply PCA to a multi-frequency eddy current application (using analytically gen-
erated model data), where the ability to monitor variation in physical parameters of
interest, using a reduced number of variables is shown. Acquisition of the reduced
number of variables is an alternative to previously employed strategies for increasing
speed of eddy current inspection, as relevant inspection data acquired in real time
is significantly reduced. PCA using model data is validated against PCA data from
physically acquired data.
7.2. PRESSURE TUBE TO CALANDRIA TUBE GAP
MEASUREMENT 90
Acquisition of multi-frequency (4 kHz, 8 kHz, and 16 kHz) eddy current data is per-
formed by a transmit-receive eddy current probe. The eddy current probe is mounted
on an inspection head, which is driven axially, while rotating circumferentially within
a pressure tube [70].
The eddy current probe is spring loaded and therefore makes contact with the
pressure tube surface during acquisition of gap measurement data. Also, eddy current
probe transmit and receive coils are aligned in the axial direction of the pressure tube.
Due to a relatively small center-to-center spacing between the transmit and receive
coil (11 mm), compared to the 330 mm pressure tube inner circumference, a planar
approximation of the geometries in proximity to the gap probe has been employed
[77]. Figure 7.1 identifies coil positions, relative to other component geometries.
Eddy current probe amplitude response increases with increased gap variation,
as this interaction is fundamentally a calandria tube lift-off response [70, 69]. Data
from multiple non-concentric pressure tube to calandria tube segments is acquired to
generate calibration data. In calibration sets the pressure tube wall thickness is varied
between segments, representing beginning of life, mid-life, and end-of-life conditions,
since pressure tube wall thickness decreases with in-reactor life [3]. For each eddy
current frequency, calibration data combined with eddy current data acquired in-
channel along with ultrasonic measurements, are used to generate in-channel gap
measurements [70].
The skin depth equation for a plane magnetic field incident on a planar surface
[5] is given by
7.2. PRESSURE TUBE TO CALANDRIA TUBE GAP
MEASUREMENT 91
Figure 7.1: Material components in the vicinity of the gap probe, where pressure tube
and calandria tube are approximated as flat plates.
J
= exp − x ,
n o
J0 (7.1)
δ
where J0 is the surface eddy current density, J is the eddy current density at x
mm into the conductor, and δ is the electromagnetic skin depth (mm) of penetration
into the conductor. δ is given by [5]
r
ρ
δ = 50 , (7.2)
f µr
U T ΣU = Λ. (7.3)
When raw data vectors in X are in different units, the correlation matrix can be
used instead of the covariance matrix to scale data appropriately [11]. To employ the
correlation matrix for PCA, vectors xi , i ∈ {1, . . . , p} are scaled by 1/σi , where σi2 is
the variance of xi .
Principal component scores Z = [z1 , z2 , . . . , zp ] are computed as Z = XU .
Eigenvalues λ1 , . . . , λp are the variances of the principal components, where the
percent variance explained by values of principal component i (scores) is given by
P
λi / j λj × 100%.
The original data set X can be recovered from principal component scores via the
computation X = ZU T .
Pm P
Principal component scores z1 , z2 , . . . , zm , i=1 λi / j λj ≈ 1, contain the major-
ity of variance accounted for in principal component scores z1 , z2 , . . . , zp . Therefore,
principal components u1 , u2 , . . . , um can be used in computing a subset of the full set
of principal component scores, and also approximating the original data set X from
this same subset. Considering Z = XU , computation of z1 , z2 , . . . , zm , and recon-
structing X (represented by the approximation X̃) from these scores is performed via
the following transformations,
and
The analytical model based on Dodd and Deeds equations [16] was employed to model
eddy current response to a number of varying PT and CT conditions. A modelling
program emulating Dodd and Deeds equations [77, 84] was applied to determine the
voltage developed in the receive coil of the probe at frequencies 4 kHz, 8 kHz, and
16 kHz. Modelling conditions that were employed in the simulation and subsequent
validation are given in Table 7.1, where all permutations of given parameters were
modeled. The resistivities (varying between 45 µΩ · cm and 60 µΩ · cm) bound
resistivities for non-irradiated PTs [5, 74]. Similarly, the range of wall thickness
values (3.48 mm and 4.36 mm) employed in modelling was chosen to emulate the
range of wall thicknesses of in-service pressure tubes.
Parameter Values
0, 0.5, 1.0, 1.5, 2.0
PT to CT Gap (mm) 2.5, 3.0, 4.0, 5.0, 6.0
7.0, 8.0, 9.0, 11.0, 16.0
PT Wall Thickness (mm) 3.48, 3.76, 4.36
PT Resistivity (µΩ · cm) 45, 49, 52.4, 56.9, 60
The Matlab function pca [85] was applied to the analytical model data, where resistive
and reactive components of individual frequency responses were represented as row
vectors in X (Sect. 7.3). A set of principal component scores was output from
the PCA analysis. The percentage of the total variance explained by each principal
component (ordered in terms of significance) is as follows: 96.92%, 2.76%, 0.29%,
7.4. PCA APPLIED TO ANALYTICAL MODEL DATA 95
Figure 7.4 provides a flow chart illustrating the process of compressed data acquisition
for multi-frequency eddy current data, where PCA has been employed, as presented
in Section 7.3. In the described process, data is acquired along the first m principal
component vectors, where m is less than p, the total dimension of multi-frequency
eddy current data. p = 2 × q, where q is the number of frequencies employed, and 2
7.5. COMPRESSED DATA ACQUISITION 98
Figure 7.3: Score 1 peak-to-peak variation (due 0 to 16 mm change in gap) for model
data.
represents horizontal and vertical voltages plotted on the impedance plane. Process
hardware (performing arithmetic operations) [86], can be used to linearly transform
p dimensional multi-frequency eddy current data into m dimensional PCA scores (via
Equation 7.4), thereby realizing data compression during real time acquisition.
Original uncompressed multi-frequency eddy current data sensed at the probe, can
be reconstructed from compressed data, by a simple linear transformation (Equation
7.5), again employing m principal components. Process hardware or software can
be employed during the reconstruction step, as this is performed post-acquisition.
The closer m is to the dimension of the original uncompressed multi-frequency eddy
current data (p), the more the reconstructed signals will represent the originally sensed
7.5. COMPRESSED DATA ACQUISITION
p
Amplitude error = (x0 − x00 )2 + (y0 − y00 )2 , (7.6)
where (x0 , y0 ) is the original uncompressed voltage response (resistive and reac-
tive components) and (x00 , y00 ) is the reconstructed voltage response. % error for an
approximated voltage response is given as
amplitude error
% error = , (7.7)
max gap response
where max gap response is defined as the maximum amplitude response due to a
7.6. EXPERIMENTAL VALIDATION 102
Two sets of samples with changing pressure tube wall thickness, pressure tube resis-
tivity, and pressure tube to calandria tube gap were used for experiments discussed in
this paper. The first sample varied PT to CT gap continuously between 0.8 mm and
16 mm. Pressure tube to calandria tube wall thickness also varied, as two PT sections
had their inner diameters machined to different depths. To summarize, this sample
was made up of three sections with wall thicknesses 3.48 mm, 3.84 mm, and 4.36 mm
(resistivity of the PT was ∼ 53.8 µΩ · cm at 20 ◦ C). The experimental configuration
is described in detail elsewhere [77].
The other sample set that was used included changes to PT resistivity and gap.
7.6. EXPERIMENTAL VALIDATION 103
Two different pressure tube samples, each with differing resistivity (51.38 µΩ · cm
and 55.95 µΩ · cm) were cut and welded, and their respective wall thicknesses were
machined such that they were equal within a few microns. A composite pressure tube
was housed within a calandria tube such that the gap varied continuously, from just
above 0 mm to just below 16 mm [77].
Weighted PCA (where the inverse variable variances are used as weights) was per-
formed on physical measurements of eddy current responses to changes in PT wall
thickness and PT to CT gap (detailed in Section 7.6). The percentage of the total
variance explained by each principal component (ordered in terms of significance) is
as follows: 87.94%, 11.66%, 0.36%, 0.03%, 0.00%, and 0.00%.
A polynomial surface of the form
is fit to the first principal component score values shown in Figure 7.6 (blue circles),
such that f (x, y) = 48.82 + 0.91x − 22.87y − 0.01694x2 − 0.1077xy + 2.439y 2 .
Towards validating the model data against experimental data, PCA was performed
on a subset of the model data, where parameters were restricted to those with values
near the physical parameter values present in the physical experiment, where PT wall
thickness and PT to CT gap were varied. More specifically, PT resistivity was fixed at
52.4 µΩ · cm, while CT resistivity was fixed at 74 µΩ · cm. The percentage of the total
variance explained by each principal component (ordered in terms of significance) is
as follows: 87.98%, 11.90%, 0.09%, 0.03%, 0.00%, and 0.00%.
A polynomial surface of the same form used in surface regression for PCA of
experimental data (Equation 7.8), is fit to the first principal component score values
for the model data, shown in Figure 7.6 (yellow data), such that f (x, y) = 52.05 +
1.862x − 24.91y − 0.02486x2 − 0.3097xy + 2.784y 2 .
Examination of Figure 7.6 reveals that polynomial fits with respect to the exper-
imental and model data (restricted to approximate physical parameter values) are
similar, serving to validate the PCA approach for PT wall thickness and PT to CT
gap variations.
Eddy current scans of a composite pressure tube, having two sections with different
resistivities (51.38 µΩ · cm and 55.95 µΩ · cm) were taken. A calandria tube enclosed
the pressure tubes during the scan, to simulate in-channel conditions. For each axial
position along the pressure tube, gap varied uniformly between minimum and max-
imum (i.e. cross sectional gap profiles are identical along the pressure tube in the
7.6. EXPERIMENTAL VALIDATION 105
Figure 7.6: Component 1 score with regression surfaces, for variations in PT wall
thickness and PT to CT gap. Experimental data is shown in blue while
analytical model data is shown in yellow.
axial direction).
Weighted PCA (where the inverse variable variances are used as weights) was per-
formed on physical measurements of eddy current responses to changes in PT resis-
tivity and PT to CT gap (detailed in Section 7.6.2). The percentage of the total
variance explained by each principal component (ordered in terms of significance) is
as follows: 81.46%, 18.05%, 0.43%, 0.05%, 0.01%, and 0.00%.
A polynomial of the same form as Equation 7.8, where x represents PT to CT
7.6. EXPERIMENTAL VALIDATION 106
gap (µΩ · cm), and y represents PT wall thickness (mm), is fit to the first princi-
pal component score values shown in Figure 7.7 (blue circles), such that f (x, y) =
−20.57 + 0.337x + 0.309y − 0.033x2 + 0.010xy + 0y 2 .
Figure 7.7: Component 1 score with regression surfaces, for variations in PT resis-
tivity and PT to CT gap. Experimental data is shown in blue while
analytical model data is shown in yellow.
Towards validating the model data, PCA was performed on a subset of the model
data. Parameters were restricted to those with values near physical parameter values
present in the physical experiment, where PT resistivity and PT to CT gap were
varied. More specifically, PT wall thickness was fixed at 4.36 mm, while CT resistivity
was fixed at 74 µΩ · cm. The percentage of the total variance explained by each
principal component (ordered in terms of significance) is as follows: 88.32%, 11.11%,
0.54%, 0.03%, 0.00%, and 0.00%.
7.7. DISCUSSION 107
A polynomial surface of the form of Equation 7.8 is fit to the first principal
component score values shown in Figure 7.7 (yellow circles), such that f (x, y) =
−46.98 − 0.3423x + 0.8312y − 0.018x2 + 0.01466xy + 0y 2 .
Examination of Figure 7.7 reveals that polynomial fits with respect to the experi-
mental and model data exhibit similar trends, serving to validate the PCA approach
for PT resistivity and PT to CT gap variations.
The weighted PCA of both the experimental and analytical model data, generated
via Dodd and Deeds equations [16] generally agree and reveal similar trends (Figures
7.6 and 7.7). Disagreement in PCA output for the two differently generated data
sets can be attributed to difference in experimental and analytical phase response to
variations in physical quantities: PT wall thickness, PT resistivity, and PT to CT
gap. Limitations of Dodd and Deeds equations may be the result of the multi-coaxial
coil approximation used to model the experimental transmit-receive configuration
[77]. This may be the cause for minor difference between experimental and analytical
response, exhibited by score 1. Agreement between analytical and experimental am-
plitude responses of the entire signal to variations in these physical quantities have,
however, been validated [77].
7.7 Discussion
to allow for the signals of interest to align in the vertical direction of the impedance
plane. Employing PCA allows for this rotation to be emulated by computation of
principal components; direction of subsequent principal component vectors, identify
subsequent directions of maximum data variance.
Normalization of scale is achieved via the weighted PCA process, as resultant
variance of columns of Σ are 1. This has the advantage of normalizing data sets be-
longing to different multi-frequency ET signals, having different gain settings, and/or
acquired by different probes and instruments.
Prior to field use of PCA, for interpretation of multi-frequency eddy current gap
measurement data, consideration should be given to practical implications of field
implementation. Probe lift-off due to oxide and sharp pressure tube inner diameter
surface changes will induce impedance plane responses, which would also be required
as inputs into a fully inclusive PCA model for analysis of gap measurement data.
Other physical effects to consider include variable cable impedance for different cable
lengths and changing probe position, thermal signal drift, electrical noise, and prox-
imity to other reactor components outside the calandria tube (e.g. Liquid Injection
Shutdown System nozzles).
In implementing compressed data acquisition, information content contained in the
last few principal component scores will be lost upon recovery of the full dimensional
multi-frequency data set. Although it has been shown that responses to gap, wall
thickness, and resistivity are captured within the first three principal component
scores (Figure 7.2), generalization of compressed data acquisition to other applications
7.8. CONCLUSION 109
should verify that responses of particular physical phenomena of interest, which could
be present in the last few principal component scores, are not lost. To mitigate loss
of inspection data of interest, data used to generate principal components, should
include responses to relevant physical phenomena. Also, bench testing should be
employed to ensure a sufficient number of principal components are used in data
acquisition, for identification of physical phenomena.
7.8 Conclusion
data, and on eddy current data analytically derived from Dodd and Deed equations
[16] via computer program [84], was performed. Agreement was observed between
the PCA performed on analytical data and experimental data, with small differences
attributed to incomplete representation of the transmit-receive experimental data by
Dodd and Deeds equations.
Acknowledgment
The authors would like to thank Dr. Ross Underhill and Stuart Craig for valuable
discussions, as well as Ryan Howard from Ontario Power Generation for supporting
this work.
111
Chapter 8
Discussion
Traditional methods employed to examine eddy current measurement data have in-
volved examination of impedance plane diagrams, and more recently, C-Scan displays.
These methods have provided an intuitive and basic introduction for the interrogation
of eddy current data with minimal analysis; voltage responses to physical phenomena
such as lift off, resistivity changes, or the presence of an edge or defect have clear and
pre-defined responses for an operator to detect and interpret. Acquisition trajectory is
implicit in how voltage responses are displayed, as curves on impedance planes corre-
spond to voltage response to probe movement relative to a test piece, and C-Scan axes
correspond to probe location. The techniques developed in this thesis demonstrate
the effectiveness of particular analysis techniques applied to data that are independent
of acquisition location. These methods rely on representation of inspection data, in
the context of other data acquired either locally or globally. For example, the surface
profiling technique employs directional derivatives in multiple directions, relying on
neighbouring data to highlight areas of probe lift off. Furthermore, weighted prin-
cipal components analysis of multi-frequency eddy current data normalizes principal
component scores by their standard deviations [87]. This allows for normalization
112
of phase and scale for data sets acquired by different instruments and gain settings,
providing a common framework for representation of multi-frequency eddy current
data, and analyst review, as shown in the third manuscript provided in this thesis
(Chapter 7).
The surface profiling technique developed in this thesis employs a transmit-receive
probe, initially developed for pressure tube to calandria tube gap measurement. This
surface profiling technique has been employed at Ontario Power Generation for the
detection and characterization of near-surface pressure tube flaws, providing inde-
pendence (with respect to ultrasonic measurements) for detection and disposition of
pressure tube artefacts. Worthy to note is the dual functionality of the gap probe,
now employed for both relative eddy current measurement, towards gap estimation,
and pressure tube artefacts (including flaw) detection. Specialized engineered tools
having drastically different, yet critical functionalities, are rare, and development pro-
cesses for analysis tools employing the gap probe should be noted in evaluation of
other engineered tools in extending their functionality beyond their initially intended
purposes.
Recent advances in signal processing have introduced compressed sensing as a
methodology allowing for reducing traditional acquisition data density requirements
(per the Shannon-Nyquist theorem [88]), needed for full reconstruction. In simi-
lar fashion, compressed data acquisition, facilitated through calculation of principal
components of multi-frequency eddy current data, allows reduction of acquisition data
density requirements, while providing 99.9 % reconstruction capability, using only half
the normally acquired data, and still 97 % reconstruction capability using only one
principal component. Although different in their approach and requirements, these
113
two methods for reducing requirements for data acquisition can be employed where
the cost (whether due to cost of materials, or induced delays) of acquiring data is
prohibitive.
114
Chapter 9
Conclusions
using ultrasonic gauging measurements collected in reactor within fuel channels. Fi-
nally, this tool is generalizable to other conductive surfaces and eddy current probes.
The conditions for extending functionality beyond pressure tube applications, include
high eddy current data density and relatively large surface area of the probe, com-
pared to artefacts required to be imaged. As a result of work done towards this
thesis, the surface profiling tool is now a commissioned fuel channel inspection tool,
employed at Ontario Power Generation. It is used in tandem with ultrasonic mea-
surements to assess fitness for service of fuel channels. Analyst technicians have been
trained in its use, and the condition of numerous fuel channels has been evaluated
using the tool.
Contributing to a comprehensive characterization of gap eddy current data, inde-
pendent and redundant information across multi-frequency data sets has been iden-
tified through principal components analysis. Eddy current responses to PT to CT
gap, PT wall thickness, and PT resistivity have been modeled via Dodd and Deed
solutions to Maxwell’s equations. 97% of the data variance is represented by the first
principal component scores, while 99.7% (nearly 100%) is represented by the first
two principal component scores, reflecting a high degree of redundancy across multi-
frequency data sets. Minimum-to-maximum response of the first principal component
to gap variation, with varying PT wall thickness and resistivity is consistent with ex-
pected responses, considering the skin depth equation. The model data has been
validated against physical experiments, varying PT to CT gap, PT wall thickness,
and PT resistivity. Demonstration of compressed data acquisition for multi-frequency
gap data has identified a methodology, which is generalizable to potential applications
116
that acquire redundant information, and are limited due to acquisition hardware con-
straints. In the case of multi-frequency eddy current gap data, using half the number
of principal components (compared to the dimension of the original data set) allows
for acquisition and reconstruction fidelity to within 1%, for both mean and standard
deviation errors.
Finally, Dodd and Deeds solutions have been experimentally validated using sam-
ples with varying PT to CT gap, PT resistivity, and PT wall thickness variations.
This has facilitated the application of principal components analysis of eddy cur-
rent gap data using model data generated via Dodd and Deeds solutions. PCA of
model data was used to extract the significant components that would also arise
in experimental data due to variations in PT resistivity, wall thickness and PT to
CT gap. Employment of Dodd and Deeds solutions is generalizable to conductive
surfaces, whose intersection with eddy current probe magnetic field spread, can be
approximated as flat plates.
117
Chapter 10
Future Work
In this section the potential extension of the work of this thesis is introduced and
discussed. For purposes of discussion, future work can be grouped into two different
areas: implementation of generalized applications for multi-frequency eddy current
data, and development of mathematical tools for further characterization of the fuel
channel condition.
Generalized applications can be realized for the surface profiling methods, mod-
elling employing Dodd and Deeds equations, and principal components analysis of
multi-frequency eddy current data. Surface profiling of conductive surfaces can be
realized for numerous geometries, given eddy current data is collected with sufficient
data density, the direction of the lift-off vector is known, and the surface area of the
probe is larger than area of surface protrusions requiring imaging. Increased degrees
of freedom of probe movement (requiring probe redesign) will allow for ability to im-
age surface variations of increased geometrical variation. Furthermore, modelling of
factors affecting measured eddy current impedance employing Dodd and Deeds solu-
tions can be extended beyond the gap probe and fuel channel geometries, provided the
area in which the interaction between the magnetic field induced by the eddy current
118
probe, and conductive material can be approximated by a flat plate. Finally, in addi-
tion to employing principal components analysis (PCA) to highlight the relationship
between physical variables and high dimensional multi-frequency eddy current data,
PCA can be employed in the realization of compressed data acquisition. It has been
demonstrated how compressed data acquisition can increase the acquisition speed
of eddy current inspection given fixed limitations of data acquisition hardware. To
realize compressed data acquisition for eddy current systems, process hardware can
be employed to reduce the dimensionality of the data sensed by the probe for data
acquisition, while not overly restricting ability to reconstruct the originally sensed
data.
A number of mathematical tools could be exploited to further characterize the
fuel channel condition. PCA could be combined with techniques such as cluster anal-
ysis, support vector machines, or artificial neural networks to simultaneously extract
multi-parameter data, collected from in-reactor fuel channels. Reduction of data set
dimensionality via PCA can provide for tools exploiting geometric and topological
arguments, which are either not available or are only apparent at higher dimensions.
The potential for extraction of resistivity from pressure tube data, which could give
information on pressure tube condition in terms of irradiation induced effects of creep
on hardness and ductility and potentially hydrogen/deuterium pick-up, is a clear ex-
ample of an important candidate parameter for extraction. Specific to the problem
of PT resistivity extraction, as made evident from examination of the skin depth
equation, different excitation frequencies have variable sensitivity to resistivity, and
variation in this sensitivity may provide sufficient independence across measurements
to solve for resistivity. Towards extraction of new information identifying physical
119
Bibliography
[2] M. Scholz, Approaches to analyse and interpret biological profile data. PhD thesis,
University of Potsdam, 2006.
[3] S. Shokralla and T. Krause, “Methods for evaluation of accuracy with multiple
essential parameters for eddy current measurement of pressure tube to calandria
tube gap in CANDU reactors,” CINDE Journal, vol. 35, no. 1, pp. 5–8, 2014.
[7] M. Trelinski, “Selected CANDU pressure tube degradation mechanisms and in-
spection related issues,” in 17th World Conference on Nondestructive Testing,
October 2008.
[17] C. V. Dodd and W. E. Deeds, “Some eddy-current problems and their integral
solutions,” Tech. Rep. Contract No. W-7505-eng-26, Oak Ridge National Labo-
ratory, April 1969.
[18] C. C. Cheng, C. V. Dodd, and W. E. Deeds, “General analysis of probe coils near
stratified conductors,” International Journal of Nondestructive Testing, vol. 3,
pp. 109–130, 1971.
[20] European Network for Inspection and Qualification, Luxembourg, ENIQ Recom-
mended Practice 6: The Use of Modelling in Inspection Qualification, 2nd ed.,
2011.
[27] A. A. Diaz and R. A. Mathews, Assessment of Eddy Current Testing for the
Detection of Cracks in Cast Stainless Steel Reactor Piping Components. Pacific
Northwest National Laboratory, 2006.
[32] R. Gonzalez and R. Woods, Digital Image Processing. Upper Saddle River, New
Jersey, USA: Pearson Education, Inc., third ed., 2007.
[33] K. Pearson, “On lines and planes of closest fit to systems of points in space,”
Philisophical Magazine, vol. 2, no. 11, pp. 559–572, 1901.
[36] K. I. Kim, K. Jung, and H. J. Kim, “Face recognition using kernel principal
component analysis,” IEEE Signal Processing Letters, vol. 9, February 2002.
[40] C. Stott, P. Underhill, V. Babbar, and T. Krause, “Pulsed eddy current detection
of cracks in multilayer aluminum lap joints,” Sensors Journal, IEEE, vol. 15,
pp. 956–962, Feb 2015.
[41] C. A. Stott, “Pulsed eddy current inspection of second layer wing structure,”
Master’s thesis, Royal Military College of Canada, 2014.
[42] Y. He, G. Tian, H. Zhang, M. Alamin, A. Simm, and P. Jackson, “Steel corrosion
characterization using pulsed eddy current systems,” Sensors Journal, IEEE,
vol. 12, pp. 2113–2120, June 2012.
[43] M. Pan, Y. He, G. Tian, D. Chen, and F. Luo, “PEC frequency band selection
for locating defects in two-layer aircraft structures with air gap variations,” In-
strumentation and Measurement, IEEE Transactions on, vol. 62, pp. 2849–2856,
Oct 2013.
[44] Y. He, F. Luo, M. Pan, F. Weng, X. Hu, J. Gao, and B. Liu, “Pulsed eddy
current technique for defect detection in aircraft riveted structures,” NDT & E
International, vol. 43, no. 2, pp. 176 – 181, 2010.
[45] Y. He, M. Pan, D. Chen, and F. Luo, “PEC defect automated classification
in aircraft multi-ply structures with interlayer gaps and lift-offs,” NDT & E
International, vol. 53, no. 0, pp. 39 – 46, 2013.
BIBLIOGRAPHY 126
[46] G. Yang, J. Kim, L. Udpa, and S. Udpa, “Analysis of pulsed eddy current - gmr
data using principal component analysis,” Studies in Applied Electromagnetics
and Mechanics, vol. 33, pp. 207–214, 2010.
[47] J. Kim, G. Yang, L. Udpa, and S. Udpa, “Classification of pulsed eddy current
gmr data on aircraft structures,” NDT&E International, vol. 43, no. 2, pp. 141–
144, 2010.
[49] P. Horan, P. Underhill, and T. Krause, “Real time pulsed eddy current detection
of cracks in F/A-18 inner wing spar using discriminant separation of modified
principal components analysis scores,” Sensors Journal, IEEE, vol. 14, pp. 171–
177, Jan 2014.
[50] G. Tian, A. Sophian, D. Taylor, and J. Rudlin, “Wavelet-based pca defect clas-
sification and quantification for pulsed eddy current ndt,” IEEE Proceeding -
Science, Measurement and Technology, vol. 152, no. 4, pp. 141–148, 2005.
[51] B. Gao, H. Zhang, W. L. Woo, G. Y. Tian, L. Bao, and A. Yin, “Smooth nonneg-
ative matrix factorization for defect detection using microwave nondestructuve
testing and evaluation,” IEEE Transactions on Instrumentation and Measure-
ment, vol. 63, no. 4, pp. 923–934, 2014.
BIBLIOGRAPHY 127
[55] European Network for Inspection and Qualification, Luxembourg, ENIQ Recom-
mended Practice 1: Influential/Essential Parameters, 2nd ed., 2005.
[57] G. Tian, Z. Zhao, and R. Baines, “The research of inhomogeneity in eddy current
sensors,” Sensors and Actuators A: Physical, vol. 69, no. 2, pp. 148 – 151, 1998.
[58] B. G. Koronev, Bessel Functions and Their Applications. CRC Press, first ed.,
2002.
[60] S. Tian, Z. Chen, M. Ueda, and T. Yamashita, “Signal processing schemes for
eddy current testing of steam generator tubes of nuclear power plants,” Nuclear
Engineering and Design, vol. 245, no. 0, pp. 78 – 88, 2012.
[63] ASM International, Metals Park, OH, USA, Materials Handbook. Volume 17:
Nondestructive Evaluation and Quality Control. p. 168, ninth ed., 1989.
[64] M. Wrzuszczak and J. Wrzuszczak, “Eddy current flaw detection with neural
network applications,” Measurement, vol. 38, no. 2, pp. 132 – 136, 2005.
[65] G. Chen, A. Yamaguchi, and K. Miya, “A novel signal processing technique for
eddy-current testing of steam generator tubes,” Magnetics, IEEE Transactions
on, vol. 34, no. 3, pp. 642–648, 1998.
[68] S. Shokralla, “Reducing the negative effects of grating through a-scan post pro-
cessing when compensating for poor element directivity,” Canadian Institute for
NDE, vol. 31, no. 4, pp. 8–11, 2010.
[70] S. Shokralla, T. W. Krause, and J. Morelli, “Surface profiling with high density
eddy current non-destructive examination data,” NDT&E International, vol. 62,
pp. 153 – 159, March 2014.
[71] J. D. Jackson, Classical Electrodynamics, ch. 5, p. 220. New York, USA: Wiley
& Sons, third ed., 1999.
[72] V. S. Cecco, J. R. Carter, and S. P. Sullivan, “An eddy current technique for
detecting and sizing surface cracks in carbon steel,” Materials Evaluation, vol. 51,
pp. 572–577, May 1993.
[76] D. Horn and R. Roiha, “Multifrequency analysis of eddy current data,” in 16th
World Conference on NDT, August 2004.
[83] M. O’Connor, “Design of an eddy-current array probe for crack sizing in steam
generator tubes,” NDT& E International, vol. 36, no. 7, pp. 515 – 522, 2003.
[86] F. Jovic, Process Control Systems Principles of design, operation and interfacing.
Springer Netherlands, first ed., 1992.
[87] Z. Fan, E. Liu, and B. Xu, Weighted Principal Component Analysis, pp. 569–574.
Berlin, Heidelberg: Springer Berlin Heidelberg, 2011.
[88] A. J. Jerri, “The shannon sampling theorem;its various extensions and applica-
tions: A tutorial review,” Proceedings of the IEEE, vol. 65, pp. 1565–1596, Nov
1977.