Journal of Nondestructive Evaluation, Vol. 18, No.

1, 1999

Review of Advances in Quantitative Eddy Current

Nondestructive Evaluation

B. A. Auld1 and J. C. Moulder2

Received May 11, 1998; revised March 1, 1999

A comprehensive review of advancements in eddy current (EC) modeling is presented. This paper
contains three main sections: a general treatise of EC theory, the thin skin EC forward modeling,
and the EC inverse problem. (1) The general treatise of eddy current theory begins with an exposition
of the reciprocity formulas for evaluating probe impedance changes, which are derivable from first
principles. Two versions of the reciprocity formulas, one with a surface integral and the other with
a volume integral, are given. Any particular type of defect, as well as both one-port and two-port
probes, can be treated. Second, a brief account of analytical and numerical methods for calculating
the field distributions is presented. Third, theory of probe/material interactions with various defect
types is described. (2) The paper then proceeds to the forward modeling section, which contains a
detailed treatment of the eddy current forward problem for surface breaking cracks and EDM
notches in the thin skin approximation. (3) The inverse problem section begins with a general
review of commonly used inversion methods, exemplified by selected references from the literature,
followed by more detailed examinations of EC inversions for surface breaking cracks and slots.
The last part of this section is devoted to the inverse problem for layered structures. Although being
a review in nature, the paper contains a number of new accounts for time-domain eddy current
interactions. In particular, a modification is proposed to the reciprocity formula in order to take a
better account of pulsed eddy current signals.

KEY WORDS: Eddy current modeling; quantitative NDE; reciprocity formulas; thin skin theory; eddy current
inversion; pulsed eddy current.

1. INTRODUCTION resistive losses in the testpiece. The induction of eddy

currents in the testpiece was characterized by the lumped
The original development of eddy current probes for mutual inductance of the transformer, which was taken
defect detection and material characterization was based to vary inversely with the distance between the probe
on circuit concepts. In this approach, the metallic testpiece and the metal surface (liftoff distance) according to some
was pictured as the short-circuited secondary of a trans- empirical law. A defect in the metal was considered to
former, with the probe coil constituting the primary. This alter the series resistance in the secondary of the trans-
model was formalized by an equivalent circuit diagram, former, by an amount depending on the size of the defect
in which a series resistance in the primary represented coil and its position relative to the probe coil. Similarly, global
losses and a series resistance in the secondary represented changes in the resistivity of the metal and resistive inho-
mogeneities, etc., were also considered as changes in the
1 secondary resistance.
E. L. Ginzton Laboratory, Stanford University, Stanford, California
94305.
The circuit model qualitatively predicted the general
2
Center for Nondestructive Evaluation, Iowa State University, Ames, behavior of changes in the probe input impedance due
Iowa 50011. to the presence of a defect and variations of the liftoff
3
0195-9298/99/0400-0003$16.00/0 © 1999 Plenum Publishing Corporation
4 Auld and Moulder

distance. It showed, for example, that the probe imped- • variations in the properties of homogeneous met-
ance change AZ has a different phase angle for a defect als due, for example, to processing errors occuring
and for a liftoff variation. But it could not provide quanti- during manufacture; and
tative information about the defect and liftoff signals, or • variation in the thickness and material properties
about the effect on these signals of changes in the defect of surface coatings.
and probe geometries. The first step beyond this concept For all of the listed defects, having a model that accurately
was taken by Burrows,(1) who first introduced the use of predicts the AZ of the probe is a prerequisite to the satis-
reciprocity in eddy current probe analysis. His method factory performance of eddy current diagnostics. First,
(for defects in the form of small ellipsoidal inclusions such a model should provide a solution to the forward
with altered resistive and magnetic properties) first calcu- problem—that is, a quantitative analysis of probe
lated the magnetic moment of the inclusion induced response to defects of all types. Second, the model should
through its interaction with the eddy current and magnetic be formulated so as to facilitate the development of solu-
field distributions excited in the metal testpiece by a tions to the inverse problem—that is, the quantitative
fixed input current / injected at the probe terminals. The characterization of a particular defect from measurements
reaction of the defect back on the probe was then evalu- of changes in the eddy current probe impedance as the
ated by finding the voltage change AV at the probe termi- probe is scanned over the defect. Finally, a satisfactory
nals, created by the altered flux linkages originating from model should also accurately predict probe impedance
the induced magnetic moment of the inclusion. Reciproc- changes due to the liftoff and tilt variation as the probe
ity arguments were used to simplify the analysis. Combin- is scanned over a metal surface. The background "clutter"
ing the fixed input current / and the voltage change AV arising from liftoff and tilt variations constitutes the pri-
due to the presence of the defect provided an expression mary limitation on the effective signal-to-noise of a
for the probe impedance change AZ due to the defect. defect signal.
The defect's geometry was also included in the analysis, The body of the paper begins, in Section 2.1, with
through the dependence of its induced moment on the the development from first principles of relations for the
change in impedance AZ of an eddy current probe in
shape of the inclusion.
the presence of a defect. General formulas are obtained
Burrows' analysis, which was applied to both single-
without restriction to any particular type of defect, for
coil (one-port) and drive-coil/pickup-coil (two-port)
both one-port and two-port probes. The results are stated
probe geometries, was an important advance in eddy
in two alternative forms: (i) as a field integral over a
current probe modeling. It introduced explicitly the
surface enclosing the defect and (ii) as a field integral
dependence of the defect signal on the field distribution
over the volume of the defect. These integrals involve
of the probe and therefore on the probe geometry, showing
the probe field distribution in the absence and in the
that quantitative characterization of a defect could be
presence of the defect, as well as the dimensions and
achieved only if the probe field itself was accurately
shape of the defect. To evaluate accurately a defect signal
known. But the model had shortcomings. It applied rigor- AZ, it is necessary to begin with a model for the probe
ously only to ellipsoidal defects, and then only if the field in the absence of a defect. Section 2.2 presents a
defect was so small that it lay in an essentially uniform brief overview of analytical and numerical methods used
region of the probe field. These difficulties can be over- for calculating the field distributions of both air-core and
come by treating the probe/defect interaction as an elec- ferrite-core probes over a metal surface. Then, finally,
tromagnetic field problem rather than a circuit problem. Section 2.3 considers probe/material interactions for the
This approach, to be presented here in detail, strongly various types of defects listed above. After this general
resembles the analysis used in microwave theory to obtain treatment, Sections 3 continues with a detailed treatment
equivalent circuit representations for a distributed field in the thin skin approximation of the eddy current forward
structure.(2,13) But, in the present case, the method is problem for surface-breaking cracks and EDM notches.
applied to quasistatic fields of kilohertz frequencies, Section 4 begins with a general review of commonly used
rather than to propagating fields at gigaherz frequencies. inversion methods, exemplified by selected references
Current applications of eddy current probes include from the literature. Section 4.2 then examines in more
the detection and characterization of detail the inversion of eddy current signals from surface-
• surface-breaking and internal cracks; breaking cracks and slots, while Section 4.3 treats the
• pits, internal voids, and inclusions; inversion problem for layered structures. In both cases
• nonlocalized material inhomogeneities, as in cor- examples are given of new developments in the use of
rosion problems; pulsed eddy current instruments.
 Advances in Eddy Current Nondestructive Evaluation 5

2. GENERAL ELECTROMAGNETIC FIELD

MODELING

2.1. AZ Relations

The traditional approach to probe modeling used an

equivalent circuit model to provide a qualitative descrip-
tion of probe performance.(4) Burrows later developed a
quantitative equivalent circuit model, for defects con-
sisting of small ellipsoidal inclusions and voids.(1) His
approach was based on use of the electromagnetic field
reciprocity relation (Lorentz) to establish a circuit reci-
procity relation for two-port probes. Change in the cou-
pling between the drive coil and the pickup coil, due to
the presence of a defect, was formulated using polariza- Fig. 1. General one-port (or absolute) probe geometry with a coaxial
line feed.
tions of the defect induced by exciting the drive and
pickup coils in turn. Applying the circuit reciprocity rela-
tion then yielded the quantitative change AZ21 in mutual
impedance between the two coils.
Later applications of the Lorentz reciprocity relation
to probe modeling were based on the concepts of micro-
wave circuit theory.(2,3) This approach was followed by
Bahr(5) to analyze 100-GHz eddy current detection in a and the input impedance is
two-port waveguide driven system, by Auld(6) for a one-
port microwave ferrite resonator probe, and by Zaman et
al.(7) for a one-port coil probe. In all of these papers,
exact expressions give AZ for arbitrary defects in terms The basic probe field/impedance relationship can be
of the probe fields in the absence and in the presence of established using Poynting's theorem. Complex power
the defect. Reference 5 expresses AZ in the form of a delivered at the input terminal plane in Fig. 1 is defined as
surface integral, while Refs. 6 and 7 give both volume and
surface integral forms. These results generalize Burrows'
work and define the electromagnetic boundary value
problems that must be solved to evaluate AZ. The three From Poynting's theorem, this is also given as
references cited all begin with an integration of the
Lorentz reciprocity relation over the domain of the prob-
lem and arrive at impedance relations by manipulating
this integral. In the following development, a more physi-
cal approach is presented. Probe impedances for the where E and H are the probe fields corresponding to V
unflawed and flawed states are first expressed in terms and I in 3(a) and Sc is the area of the coaxial line section
of Poynting-type field integrals, and the difference AZ at the terminal plane.
of these impedances is subsequently related to either a Because the impedance of a circuit is independent
surface integral or a volume integral over the region of of the excitation phase, the input current at the terminal
the defect. plane can, without loss of generality, be taken to have
zero phase, so that / is pure real and the conjugate sign
can be dropped in (3a). Since the magnetic field H of
2.1.1. One-Port Probes the coaxial line mode has the same phase as I at any
particular cross section, the complex conjugate can also
To illustrate the field-analytic approach, the one- be dropped in (3b). Then from (3a) and (3b),
port probe geometry in Fig. 1 is chosen as a first example.
In complex notation, the time-harmonic voltage and cur-
rent at the probe terminals are
6 Auld and Moulder

The impedance in (4a) is defined to be the probe imped- and the defect (Fig. 1), use of the divergence theorem
ance in the absence of a defect, where Ea and Ha are the converts the result to the surface integral form
probe fields excited in the unflawed testpiece by the probe
terminal current Ia. In the presence of a defect the probe
impedance is
where n is defined in Fig. 1 as the inward normal to the
entire boundary surface 5. The contribution from the
enclosing sphere vanishes as R -> I (Appendix A),(8)
and the integral around the source reduces to an integral
where Eb and Hb are the probe fields excited in the over the coaxial cross section Sc, if the shielding around
presence of a defect by the probe terminal current Ib. the source is assumed perfectly conducting. There
Since the input impedance at the terminal is independent results, then,
of the drive current, the condition

can always be imposed in (4a) and (4b). The change in

probe impedance due to the defect is then, subtracting
(4a) from (4b), given as
AZ = Zb - Za where SF is any surface enclosing the defect (Fig. 1). (The
most convenient definition of SF depends on the geometry
of the defect, as shown below.) Substitution of this result
into (7) gives an expression for AZ in terms of an integral
around the defect,
Because the H field at Sc is proportional to the input
current, it follows from

where the minus sign in (9) has been suppressed by

that reversing the order of the cross products. Recall that
subscript a defines fields in the absence of the defect and
subscript b defines fields in the presence of the defect.
on Sc. Consequently, in (5) the subscripts to H can be Appendix B shows how this result can be converted
interchanged between the first term under the integral to a volume integral format,
and the second, yielding

The expression for AZ in (7) can now be converted where VF is the volume of the flaw and de and du are
to a field integral over the defect by invoking the Lorenz the differences of permittivity and permeability between
reciprocity relation [3] the flawed "b" and the unflawed "a" states of the test-
piece. Both the permittivity and the permeability are com-
plex. In the normal range of frequencies for eddy current
between arbitrary solutions to Maxwell's equation in a testing, displacement currents in metal testpieces are neg-
source-free region with fixed material properties, where ligible compared with conduction currents and the
the a and b solutions are the fields defined in (4a) and approximation
(4b). In particular, (8a) applies to the fields defined in
(4) at all points in the region exterior to the source and
the defect in Fig. 1—that is, at all points outside the
surfaces Sc and SF in Fig. 1. If this region is enclosed by can be made. Equation (10b) generalizes Burrows' analy-
a spherical surface of radius R and (8a) is integrated over sis to completely arbitrary geometries. It should be
the volume bounded by the total surface 5 comprising emphasized that no approximations have been made in
the enclosing sphere plus the surfaces around the source the above derivation.
Advances in Eddy Current Nondestructive Evaluation 7

There are no restrictions on the anisotropy and inho- coil 2. In many reflection probes there is no transfer
mogeneity of e and u, nor are there any restrictions on the signal in the absence of a defect.
shape of the defect. Choice between the surface integral Regardless of the specific probe geometry, any two-
format (10a) and the volume integral format (10b) port reflection probe can be represented schematically as
depends on the type of defect involved and on the analytic shown in Fig. 3. Figure 3a is a general T-network equiva-
approximations or numerical methods used in evaluating lent for calculating the transfer signal. Changes in the
the b-subscripted fields under the integrals. Note that the impedances shown are to be evaluated in terms of the
surface SF in Fig. 1 can be chosen arbitrarily, as long as actual fields around the flaw using AZ formulas analogous
it encloses the defect. This flexibility of choice may often to (10a) and (10b). The derivations follow the same gen-
be used to simplify calculation of AZ. eral procedure as before. For example, to derive AZ12,
the two solutions for the Lorenz reciprocity relation in
(8a) and (8b) are defined as in Figs. 3b and c, so that the
2.7.2. Two-Port Probes counterpart of (9) is
The analysis presented above applies to the one-port
(or absolute) probe illustrated in Fig. 2a and to any of
the various absolute probe geometries used in practice:
coil axis horizontal,(9,10) fertile core,(11,12) rectangular-
shaped coil with rounded corners and horizontal axis,(13)
nonrectangular winding cross section,(14) uniform field
ferrite core,(15,16) encircling coil,(17) pull-through coil,(18) As in Fig. 1, the magnetic field at any terminal plane in
and waveguide probe.(19) Fig. 3 is proportional to the current at the same terminal
The same analysis is applicable to the differential plane. Because of the boundary conditions on the currents
probe illustrated in Fig. 2b. In this case, the one-port in Fig. 3, the left-hand side of (12) reduces to
analysis is used to evaluate AZ separately for the left-
hand and right-hand coils, and taking the difference.
Alternatively, using bridge circuil eleclronics, the probe
may be regarded as a two-port probe with the input port
at the bridge excitation terminal and the output port at
the bridge balance terminal. In (13), the electric fields E and the magnetic fields H,
To illustrate the formulation of AZ for a two-port as well as the voltages V and currents /, are evaluated at
probe, the reflection probe in Fig. 2c is taken as an exam- the terminal planes S1 and S2 in Fig. 3, where only the
ple. In this case, the defect is interrogated by exciting fundamental coaxial transmission line mode is present.
the drive coil 1 and the presence of the defect is detected For this mode, the electric field has only a radial compo-
by observing a change in the transfer signal to the pickup nent and the magnetic field only an azimuthal component,

Fig. 2. Basic types of eddy current probes.

8 Auld and Moulder

Fig. 3. Reflection probe analysis. (a) Equivalent circuit. (b) Solution a. (c) Solution b.

both components being independent of the azimuthal This result is converted to a transfer impedance
coordinate P and inversely proportional to the radial coor- change by writing out the impedance matrix equations
dinate r.(2,3) Consequently, the cross products under the of the T-network for solutions "a" and "b" in Fig. 3. For
integrals on the left-hand side of (13) reduce to algebraic the a-subscripted solution (I 1a = I, I2a = 0), noting that
products of the form ErHp. The p part of the surface Z12 = Z21 by reciprocity,
integrations then introduce a multiplier 2P, which con-
verts the magnetic field factor into the coaxial line current,
while the radial part of the integration converts the electric
field factor into the coaxial line voltage. This produces the
voltage-current products on the right-hand side of (13). and for the b-subscripted solution (I1b, = 0, I2b = I)
Advances in Eddy Current Nondestructive Evaluation 9

general review of b-field modeling is given for the various

types of defects listed in Section 1, with particular empha-
sis on cracks. Section 3 continues with a detailed treat-
Substitution of (15a) and (14b) into (13) and (12) yields ment of surface-breaking cracks and EDM notches.

2.2. Probe Field Modeling

2.2.1. Air Core Coils

A classical study of circular air core coils over a
As in the case of (10a) and (10b), the surface integral in
metal surface is given in an analytical formulation by
(16a) can be converted to volume form, yielding
Dodd and Deeds, first performed for a single circular coil
over various planar structures and for a single encircling
coil over uncladded and cladded cylinders.(39) An essen-
tial feature of the Dodd and Deed analysis at typical eddy
current frequencies is that the actual current distribution
in the wire cross section is approximated by a current
filament. This method was later extended to multilayer
The above formulation of AZ12 is applicable to any
cylinders(40) and to circular drive/pickup assemblies. (41)
of the two-port probe types currently in use: differential
The approach is based on use of the vector potential,
pickup coil,(20) remote field,(21-23) encircling coils,(24-26)
A, which, for the axisymmetric structures considered, has
horizontal drive and pickup coils,(13,27,28) vertical drive
only an azimuthal component.
and horizontal pickup coils,(29,30) geometrically aniso-
The a-field for an absolute probe (Fig. 2a) is calcu-
tropic,(31) and ferrite core.(32,33) It can also be applied to
lated in terms of the current / in the coil by constructing
the differential probe in Fig. 2b, as noted at the beginning
the vector potentials in the air medium and in the metal,
of this subsection.
and imposing continuity at the interface. Solution of any
These and other AZ relations can also be formulated
electromagnetic problem is simplified if the dimensions
for eddy current probe arrays.(34-38) The AZ formulas
of the physical structure are small compared with the
(10a) and (10b) and(16a) and (16b) are exact provided
"characteristic length" of the electromagnetic field. In
that exact distributions are used for the a-field (defect
this case the time derivative term in the wave equation
absent) and the b-field (defect present). These two fields
can be neglected, leaving only the V2 term. This is called
may be described in operational terms as the interrogating
the "quasistatic" approximation because the field satisfies
field a and the response field b. A number of mathematical
the Laplace (static field) equation, even though all compo-
methods have been used to model the a-field. As noted
nents are time varying. The quasistatic approximation is
in Section 1, the accuracy of an eddy current measurement
applied, for example, in calculating the values of alternat-
depends critically on having a precise evaluation of this
ing current circuit elements such as coils and capacitors.
field. However, having a correct field analysis for an
Eddy current testing is typically performed at frequencies
ideal probe is not sufficient to ensure accurate signal
lower than 10 MHz, with probe coil and defect dimen-
prediction. In an actual probe, small artifacts in probe
sions less than 1 cm. In the air medium outside the metal
construction are known to induce very significant changes
testpiece, the characteristic length is the wavelength X.
in the a-field, and techniques have been devised to mea-
At usual eddy current test frequencies, X is in the range
sure the a-field distribution directly. These questions are
of kilometers, so the quasistatic approximation is always
addressed in Section 2.2.3. Determination of the b-field
applicable in the air medium. However, in a highly con-
is considerably more complicated. But it is seen from the
ducting testpiece, the time derivative terms in the wave
AZ formulas that this field is needed only in the near-
equation reduce to
vicinity of the defect, a feature that lends itself to the use
of approximations and simplifies numerical evaluation
of AZ.
Section 2.2 reviews in general terms the various
methods, analytic and numerical, currently used in evalu- in the frequency range for eddy current testing(11) and
ating the a-field (unflawed testpiece). Modeling of the convert the wave equation into the diffusion equation,
b-field (flawed testpiece) is considered in Section 2.3. A where the characteristic length is the skin depth(2,3)
10 Auld and Moulder

a-field can be calculated separately by the Dodd and

Deeds formulation, then superimposed.
Evaluations of the a-field by analytical methods have
At 1 MHz in aluminum 8 = 0.5 mm, and it is much been reported for some complicated air core structures:
smaller in magnetic metals, so that the quasistatic coil axis horizontal,(9,10) rectangular-shaped coil with
(Laplace) approximation is not applicable in the metal rounded corners and horizontal axis,(13) and nonrectangu-
at the higher range of test frequencies. In this range, lar winding cross section.(14) Some of these methods, as
penetration of the a-field into the testpiece is governed in Ref. 28, can be applied to D-shaped coils oriented
by the skin depth. At frequencies in the very low kilohertz either horizontally or vertically.
range the quasistatic approximation does apply for typical For more complicated air core coil structures (coils
probe coil dimensions, and the a-field penetration is then located close to an edge of the testpiece, etc.) the a-
governed by the coil dimensions rather than by the field can generally be obtained only by using numerical
methods (finite difference, finite element, volume ele-
skin depth.(42)
Certain types of two-port probes, such as the reflec- ment, boundary-element hybrid, etc.).(43-45) Finite differ-
ence was the only method available during the early
tion probe in Fig. 2c, are also axisymmetric and the
development of electronic computers but is now rarely
vector potential fields of the two coils again have only
used for problems in eddy current NDE. But Section 3.4
an azimuthal component. Since the a-field in Fig. 3b has
does show how this method can be usefully applied in
coil 1 excited and coil 2 open circuited, and since the
connection with certain analytic approximations. Finite
open-circuited coil does not perturb the drive coil at eddy
element and boundary element, sometimes combined in
current frequencies, the Dodd and Deeds analysis can be
a hybrid formulation, are now the most widely used meth-
applied directly to coil 1. A particularly simple form of
ods for solving the a-field problem. Extensive reference
axisymmetric two-port probe is frequently used in some
listings are given in Refs. 43—45, with a detailed compari-
eddy current sensors for material processing control.(24-26)
son of finite element and boundary element given in Ref.
Here the drive coil is a long solenoid encircling a cylinder
45. Generally speaking, the boundary-element procedure
of the material under test and the pickup coil is a short
reduces the amount of discretization required for prob-
single-turn solenoid at the midpoint of the drive solenoid,
lems involving only isotropic and homogeneous materi-
where the a-field is uniform and is described by a simple als, compared with finite element. This advantage, which
closed form expression. derives from the use of free-space Green's functions in
A more complicated example is the differential the various geometrical regions of a test geometry, leads
reflection probe of Fig. 4a. Here there is a pair of circular to a frequent preference for the boundary-element
counterwound pickup coils, placed side by side and con- method, especially for testpieces of complex shape(46)
nected in series. In Fig. 2c the larger (drive) coil is labeled and probe location close to an edge.(47,48) In such cases,
1 and the smaller (pickup) coil is labeled 2. From Fig. 3 boundary element is often combined with another numeri-
this means that the a-field is calculated by exciting the cal procedure: hybrid volume-boundary element(49) and
large coil in the absence of the defect, and the b-field by hybrid finite-boundary element.(50)
exciting the small coil in the presence of the defect. Since
coupling between two coils is always reciprocal, the trans-
fer impedance can also be calculated with the large coil 2.2.2. Ferrite Core Coils
labeled 2 and the small coil 1. The a-field is then obtained Ferrite core probes(11,12,15,30-32) are now widely used
by exciting the small coil in the absence of the defect, to concentrate or otherwise modify the spatial distribution
and the b-field by exciting the large coil in the presence of the a-field. Figure 4b illustrates a probe of this type
of the defect. Because both coils are axisymmetric this that is specially designed to produce a spatially uniform
change of coil labeling does not reduce the difficulty of a-field in the vicinity of a defect.(15,16) (It is shown in
solving the b-field problem. In the case of the differential Section 3.3 that this probe can greatly simplify the ana-
reflection probe (Fig. 4a), there is, however, an advantage lytic evaluation of AZ.) For ferrite core probes, use of
in labeling the differential pickup coil 1 and the drive one of the numerical methods listed above is imperative.
coil 2. In doing this, the simpler a-field problem is excited Boundary element and finite element are still the preferred
by the more complicated coil geometry 1, while the more methods, but volume element has also been used.(11,12)
difficult b-field problem is excited by a simple axisymme- The remote field eddy current technique for locating
tric coil. Since the two axisymmetric coils in 1 are driven defects in ferromagnetic tubes has been modeled by the
in series by the same current I, their contributions to the finite-element method(21-23) and by a boundary-element
Advances in Eddy Current Nondestructive Evaluation 11

Fig. 4. More complex probe geometries. (a) Differential reflection probe. (b) Uniform-field ferrite probe.

mesh, using an iterated Born approximation.(51) In model- 2.2.3. Photoinductive Probe Characterization
ing this application, some difficulties arise from uncer-
tainty as to the value of the permeability and the extent In (10a) and (10b) and (16a) and (16b) the a-field
of its spatial variations.(22) When the permeability is inho- (the interrogating field) characterizes the properties of
mogeneous, only the finite-element method is applicable. the probe itself. An accurate evaluation of this field is a
Although AZ relations have not yet been applied to this prerequisite to an accurate evaluation of the b-field, which
type of probe, (16) is usable in principle once the fields characterizes the probe-defect interaction. To model the
excited by the drive and pickup coils have been evalu- defect signal AZ correctly, both the a-field and the b-
ated numerically. field must be known precisely. It has been observed exper-
12 Auld and Moulder

imentally, however, that having a precise mathematical For the particularly simple case in Ref. 26, a two-
field model is not sufficient to guarantee accurate model- port probe consisting of a long solenoidal drive coil and
ing of AZ. Nominally identical probes have been found a single-turn pickup coil, both closely fitted a long cylin-
to give signals that differ by as much as 35%, even though drical testpiece, is used to sense property changes during
the probe inductances were identical to better than 2%. processing. The change in transfer impedance can be
For truly quantitative NDE it is therefore essential to found by simply calculating the change in flux linkage
measure the a-field distribution experimentally. This can through the pickup coil, so that the change AZ21 is
be done, for example, by the photoinductive method, expressed completely, without using (16), in the form(25)
where the probe under test is placed over a "witness plate"
consisting of a carbon layer deposited on a nonmetallic
substrate. A laser beam focused on the carbon layer from
below induces a localized temperature and conductivity
where I2 is the current in the winding of the drive coil,
change.(52,53) Substituting
j is the current density in the pickup coil, and

from (11) and the Burrows' approximation for a small is the change in electric field (the "scattered" field) due
inclusion [from (24) in Section 2.3.1] into (10b) shows to the change in material properties. The integral is per-
that localized heating by the laser beam generates a AZ formed over the entire winding of the pickup coil. To
that is proportional to El at the position of the laser spot. evaluate (20) for some more complex geometries, the
Scanning the laser beam over the witness plate generates exact analytic solutions of Dodd and Deeds(39-41) may
a mapping of the electric part of the a-field in the car- also be used in (20).
bon layer. Although few analytical b-field solutions exist, an
Reference 53 shows explicitly how the three-dimen- increasing number of numerical solutions is becoming
sional electric a-field in the absence of the carbon layer available in the literature. Some examples are finite ele-
ment(18,55-57) volume element,(58,61) and boundary ele-
can be determined from photoinductive mapping of the
ment(62,63) (Collocation).(64-69) Formulation (20) may also
electric field in the layer. Since this electric field E is
equal to the vector potential A inside the test piece, the be used with these numerical solutions. But for numerical
corresponding three-dimensional magnetic a-field (used rather than analytical solutions, there is an advantage in
in the modeling techniques described here) can be calcu- using AZ. In effect, using the AZ relations reduces the
lated from A .(39-41) precision required in computing both the a- and the b-
field. This can be demonstrated by assuming a small
For simple probe geometries the measured and cal-
culated field distributions have been shown to be in good mean percentage error p in the computed fields Ea, Ha
agreement. The technique has proved to be useful for and Eb, Hb. For one-port probes it can be seen from Ref.
directly characterizing probes in field operations and for 4 that Za and Zb then have a percentage error 2p; and,
quality control in probe manufacture. from (10), AZ has the same percentage error. In contrast,
if AZ is evaluated from the one-port version of (20), the
percentage error in AZ is
2.3. Probe-Defect Interaction Modeling

2.3.1. General Considerations

In Section 2.2 a review was given of analytic and This is much larger than p when the defect is small
numerical methods for obtaining the a-field distribution.
The b-field distribution is much more difficult because
of the more complicated shape of the flawed testpiece. Stated in words, evaluation of the probe impedance
Exact and numerical solutions are available in some change from (20) requires taking a small difference
cases, such as dimensional defects (such as errors in the between two large quantities. Evaluation from the AZ
dimensions of a plate, a cylinder, etc.) and homogeneous relation does not have the limitation. This argument may
material property defects in cylinders.(26,54) Dodd and also be applied to two-port probes.
Deeds(39-41) provide a general catalog of exact analytic Since the integrals in the AZ relations are localized
a- and b-field solutions for such problems in layered to the region of the defect, a second advantage of this
plates and cylinders. format is that the computational load is reduced by
Advances in Eddy Current Nondestructive Evaluation 13

restricting discretization to the locality of the defect.(61,69) where dE, dH are of the same order as du, ds. Substitu-
This feature can be exploited in all of the volume-ele- tion of (25a), (25b) into (10a), (10b) shows that, to first
ment,(58-61) boundary-element,(62-69) and finite-boundary order in the 8 quantities, AZ can be obtained using the
element(50) computational procedures. For geometrically approximation
simple testpiece shapes, where an analytical form of the
Green's function for the bounded body is available, only
the region of the defect need be discretized when using This result is evident from (24) for the case of a
the boundary element method.(10,61,69) But in the general small, weak spherical inclusion. The advantage of the
case, it is necessary to use the unbounded Green's func- Born approximation over the quasistatic approximation
tion and discretize both the region of the flaw and the is that it is applicable to weak inclusions of arbitrary
boundary of the testpiece. shape and size. But the Born method cannot be used
Other advantages arising from the "localization" with surface integral formulations of the AZ relations.
property of the AZ relations are that it facilitates intuitive Substitution of (26) into (10a), (16a) makes this clear. In
physical reasoning and, by extension, development of the surface integral format, the AZ calculation is most
analytical approximations. Four types of approximation easily approximated when the testpiece skin depth is
are most commonly used. much smaller than the dimensions of the testpiece and
2.3.1.1. Quasistatic Approximation. If the volume of the defect.
an inclusion or a void is sufficiently small that the a- 2.3.1.3. Variational Approximation. This third mem-
field is essentially constant in the neighborhood of the ber in the trio of most widely used approximations in
defect, the quasistatic approximation is valid and the b- physics and engineering has also been applied to the
field has a static field distribution in and around the evaluation of AZ,(73-75) although not yet fully developed.
defect. For an inclusion or a void of ellipsoidal shape, Here, both the volume and the surface forms, (10a) and
the solution is particularly simple.(70,71) The b-field is then (10b), (16a) and (16b), of AZ have been reformulated as
uniform within the defect volume VF in (10b), (16b) and variational expressions. Substitution into these expres-
the integration reduces to multiplication by VF, giving sions of approximate solutions to the a- and Wields, with
Burrows' result.(1) For a spherical inclusion, with proper- a small mean percentage error p, yields a AZ with an
ties ub, £b, in a testpiece with properties ua, £a the b- error of order p2. This property facilitates the use of
field within the defect is related to the a-field at the physically intuitive field approximations and also has
position of the defect by important implications for numerical evaluations of AZ.
2.3.1.4. Thin Skin Approximation. This approxima-
tion, inspired by the methods of microwave circuit the-
ory,(2,3) applies when the skin depth (18) of the testpiece
is much smaller than the dimensions of the defect. It is
where approximation (11) has been used. Substitution therefore restricted to surface-breaking, or very nearly
into (10b), (16b) gives Burrows' formula. A complete surface-breaking, defects of various kinds—cracks, EDM
tabulation of the multiplying factors to be used in (24) notches, pits, voids, etc. Figure 5 illustrates, for the case
for ellipsoidal defects is given in Ref. 71. Other relevant of a surface-breaking crack or EDM notch, the three steps
information is available in Ref. 1. A more general involved in analyzing this type of problem. Figures 5a
approach (low-frequency asymptotics) extends the qua- and b define, respectively, the a-field and b-field problems
sistatic approximation to defects of arbitrary shape.(72) for the real physical configuration. The b-field problem
2.3.1.2 Born Approximation. If an inclusion has is approximated through the intermediate step defined
properties ub, ab that differ only slightly from the test- by Fig. 5c, for the idealized physical configuration of
piece properties, the Born approximation may be used in a perfectly conducting flawed testpiece. The tangential
evaluating (10b), (16b). For magnetic field excited by the probe at the defect/testpiece
surface is first calculated for this idealized problem. The
actual tangential electric field at this surface is then
obtained, in the thin skin approximation, from the skin
impedance(2,3) for the lossy testpiece material,
the relation between the b- and the a-fields is

applied to all points on the interior walls of the defect

14 Auld and Moulder

In the idealized defect problem (Fig. 5c) the mag-

netic field is quasistatic in the air and zero in the metal.
To understand more clearly the physical nature of the
probe field in this idealized case, it is useful to consider
the lossless Maxwell curl relations at a test frequency w,

and to substitute the power series expansions [77]

The quasistatic fields, as defined previously are clearly

the zero-order terms £(0) and H(0) in (29), so that the
quasielectrostatic and quasimagnetostatic fields are
uncoupled in (28). In this order, the probe behaves as a
lumped circuit element—a lossless inductor near a loss-
less metal screen. The quasimagnetostatic part of the field
is generated by current flow in the coil; the lowest-order
contribution to the electric field of the probe arises from
the voltage drop across the inductive reactance of the
coil. Although the resulting lowest-order E field has, in
effect, a quasistatic spatial distribution, it is proportional
to to [first order in (29)] because it is excited from the
current by way of the inductive reactance wL of the coil.
For the actual problem [Fig. 5(b)] both the coil wind-
ing and the testpiece have finite conductivity, with

Fig. 5. Probe-defect configurations for the AZ analysis, (a) Unflawed

testpiece. (b) Actual flaw. (c) Idealized flaw. in (28), so that the quasistatic fields E(0) and H(0) are
coupled. In eddy current modeling, the contribution of
the winding conductivity is normally ignored. Following
and the upper surface of the testpiece in Fig. 5b. [Correc- Dodd and Deeds(41-43) and the various numerical field
tions for the tip and the upper edges (or lips) at the calculations, the coil current is taken as an ideal current
mouth of a surface breaking crack or EDM notch(63,76) filament. The testpiece conductivity, on the other hand,
are discussed in Section 3, which gives full details of the is included. In the thin skin approximation this inclusion
above procedure.] of testpiece losses is simplified by using the skin imped-
It is useful at this point to examine briefly the nature ance (27), as outlined above.
of air/metal boundary conditions in eddy current prob- An alternative approach to the thin skin approxima-
lems. The thin skin approximation begins with a calcula- tion was developed in connection with the alternating
tion of the probe field in the presence of a lossless flawed current potential difference (ACPD) method for the detec-
testpiece. In microwave circuit theory, the solution to tion of surface-breaking defects. In its first realization
such an idealized problem is a Maxwell, or nonquasistatic this method interrogated the surface defect with a uniform
field; in eddy current theory the solution is quasistatic. current injected into the testpiece by a pair of remote
Section 2.2.1 defined the quasistatic approximation with electrode contacts. The defect was detected by means of
reference to the wave equation in air and the diffusion a pair of point voltage contacts, aligned parallel to the
equation in metal. It was seen that this approximation is direction of the unperturbed interrogating current. Pertur-
always valid in the air medium but not, for the higher bation of the measured voltage difference indicated the
test frequencies, in the metal medium. presence of a defect. Probe interaction with a surface
Advances in Eddy Current Nondestructive Evaluation 15

breaking crack was modeled in the thin skin approxima- approach is shown to be very convenient for treating
tion by means of the "unfolding" technique.(78) This ferrite core probes and very effective in modeling edge
approach was based on the analytical proof that, in this effects.(83) It permits the use of the boundary-element
measurement, the distribution of the eddy current field method for these types of problems, with a reduced com-
over the surface of the testpiece is described by the two- putational load but without sacrificing flexibility.
dimensional vector Laplace equation. The same descrip- The following subsections give brief overviews of
tion also applies to the current distribution on the inner analytic approximations and exact numerical evaluations
surfaces of the surface crack. This three-dimensional thin for the general classes of defects listed in Section 1, as
skin geometry was then unfolded into two dimensions well as for liftoff and tilt. For each topic a sampling of
by bisecting the flawed testpiece along the plane of the references is listed in chronological order. The AZ rela-
crack and then flattening each part of the bisected crack tions of the present paper are applicable to all of these
into a two-dimensional problem plane. Using the required examples cited but have not always been used explicitly
continuity conditions at the tips and edges of the half- in the cited references.
crack sections, it was shown that the "unfolded" vector
Laplace solution in two dimensions corresponds to a
known hydrodynamic flow problem. 2.3.2. Surface Cracks and Slots
In essence, use of the unfolding technique requires Because of their potential importance, natural sur-
interrogation of the defect by an essentially uniform eddy face cracks, and notches or slots (fabricated by electrical
current field. This condition was met in the earliest experi- discharge machining, EDM, as experimental simulations
ments by placing the current injection electrodes far from of cracks), have always been the subject of intensive
the defect. Because of difficulties in certain environments modeling studies. Both rectangular and semielliptical (or
with contacting the voltage difference pickup probe, it semicircular) notches have been examined. Approximate
was replaced by a horizontal pickup coil to sense the analytical methods have been developed for both the thin
horizontal component of the fringing magnetic field skin (skin depth d « crack depth a) and the thick skin
above the crack.(27,28) Eventually, for the same practical (skin depth d » crack depth a) regimes. In addition,
reason, the drive current contacts were also replaced by numerical methods have now been developed for arbitrary
a drive coil designed to provide a uniform eddy current ratios of skin depth to crack depth.
field distribution.(13) This approach, called the alternating When the diameter of an eddy current probe is small
current field measurement (ACFM) approach, then cor- compared with the skin depth of the testpiece material,
responds to use of a two-port eddy current probe, as it is the probe diameter that determines the penetration
described in Section 2.1.2. It might use a driver of the of the a-field into the material.(38,42) In such instances,
form shown in Fig. 4b. Another probe of similar form the ratio of the probe diameter to the crack depth can
evolved from a Hall probe detection scheme for observing strongly influence the choice of analytic approximation.
the fringing magnetic field of a surface crack interrogated Two important cases arise in the thin skin regime: probe
by the eddy current field of a large drive coil.(29,30,79) This diameter » crack depth (uniform field interrogation)
type of probe has a vertical drive coil and a horizontal and probe diameter « crack depth (localized field inter-
pickup coil. In Refs. 13, 27, and 28, both the drive and rogation). It is shown in Section 3 that uniform field
the pickup coils are horizontal. The "unfolding" approach interrogation accurately senses the depth of a crack but
to modeling the probe-defect interaction requires that does not clearly display the length of a crack. In contrast,
the eddy current field near the defect be in a region far localized field interrogation accurately senses the length
removed from the source fields, as in Ref. 78. When the (or surface image) of a crack but does not clearly display
current injection electrodes are replaced by a drive coil its depth. This behavior, which can be explained by noting
this condition is no longer strictly observed. A detailed that a crack (or EDM slot) is in effect a very small short
analysis of the air-metal interface and the fringing field circuited cutoff waveguide. In localized field interroga-
of a crack, in Ref. 27, addresses this question. tion the probe preferentially excites very strongly cutoff
Use of the quasistatic approximation (with the mag- modes that do not reach the bottom of the crack. This
netic field in air represented as the gradient of a scalar behavior, illustrated explicitly in Reference 80 for a rect-
potential) for the surface-breaking crack problem is angularly shaped EDM slot, will be seen to play an
detailed in Section 3.1.(20,80) To treat the complex testpiece important role in the inversion procedure.
shapes now encountered during in-service eddy current 2.3.2.1. Approximate Analytical Methods. 2.3.2.1.1.
inspection and computer simulation,(81) an extended sca- Thin Skin Regime. Reference 84 distinguished between
lar potential method had been developed.(82) This thin skin (large a/8) and thick skin (small a/8) modeling
16 Auld and Moulder

approximations. Here a is the crack depth and 8 the three-dimensional cracks were modeled by using the
skin depth. Both two-dimensional cracks (infinite surface hydrodynamic flow analogy. This analogy and Burrows'
length) and three-dimensional cracks (finite surface ellipsoidal void model were briefly reviewed in Ref. 85.
length) were modeled. The a-field (interrogating field) A variational approach(89) extended the treatment of ellip-
was assumed in all cases to be uniform over the surface soidal voids in Ref. 1 to voids of arbitrary shape. In
of the testpiece. For a two-dimensional crack the AZ principle, this technique could be applied to surface break-
integral over SF reduced to a line integral around the ing cracks of arbitrary shape. The unfolding technique,(78)
contour enclosing the crack cross section. Using the thin as applied to the ACFM method,(13,27,28) was applied to
skin surface impedance on the crack walls, performance model the b-field of a surface breaking crack in the thin
of this integral separated the contributions of the crack- skin regime. For the thick skin regime, Ref. 90 used the
wall losses and the crack opening. Contributions of the hydrodynamic flow analogy to model the ACFM response
crack lips and the crack tip(63,76) were added as correction of a semielliptical surface breaking crack. Beissner et
terms. For a three-dimensional crack the b-field was al.(29) applied the same method to obtain the b-field for
obtained from the unfolding model(78) for part-circular a semicircular surface breaking crack.
surface cracks. This model describes the surface field 2.3.2.2. Numerical Methods. In recent years there
by a potential that is continuous at the fold line. This has occurred a remarkable evolution and expansion of
assumption has been verified for ferromagnetic metals the applications of numerical methods to eddy current
but not for nonferrous metals. NDE.(81-83-91-97) These techniques are especially useful
In Ref. 85, previous results were reviewed and the when the testpiece has a complicated shape. But numeri-
modeling was extended to a rectangularly shaped surface- cal methods are also advantageous even when the test-
breaking defect (EDM slot) interrogated by an a-field piece shape is simple, because these methods are
that was spatially nonuniform on the testpiece surface. applicable even in the intermediate skin regime. However,
This analysis applied the Born approximation to the tan- despite this flexibility of numerical methods, the compu-
gential magnetic field in the mouth of the crack. That is, tational efficiency can always be improved by using the
HT in the crack mouth was assumed to be the same as thin skin approximation when appropriate. Use of a scalar
the field that would have existed in the area coinciding potential model also improves the computational effi-
with the crack mouth, but on the surface of an ciency. The finite-difference method was first applied
unflawed testpiece. in evaluating the scalar potential model of Ref. 85, for
In the second approximation, the magnetic field in semicircular and semielliptical surface breaking cracks.
the interior of the flaw was calculated for the case of a Subsequently, an improved scalar potential model(64,87)
perfectly conducting testpiece. Losses in the flaw walls was developed for boundary-element modeling of three-
were then introduced by means of the skin impedance of dimensional cracks (asymmetric probe scans) in a flat
the actual testpiece material (27). This model was used testpiece. And, more recently, an extended scalar potential
to obtain the response (flaw profile) of an absolute probe theory,(82) by placing the model on a completely rigorous
scanned along the length of a rectangular EDM slot.(86) basis, permitted smooth performance of boundary-ele-
The method was extended in Ref. 87 by applying the ment calculations of the b-field, even in the presence of
formal thin skin perturbation theory of Ref. 9 and includ- such geometric singularities as edges and comers.
ing the possibility of probe scans that are not centered 2.3.2.2.1. Finite Difference. Although this numerical
on the plane of the EDM slot. A detailed study of the method is now rarely used in eddy current modeling, the
effects of EDM slot closure(60) was based on the use of scalar potential approach of Ref. 85 is sufficiently simple
the thin skin theory of Ref. 27 to obtain the b-field without to merit application of a very simple numerical method
restriction to the Born approximation used in Ref. 85. In for solving the Laplace scalar potential problem in the
Ref. 88, Hartfield and Bowler present a more accurate interior of a crack or an EDM slot. Reference 80 presented
version of the two-dimensional crack model in Ref. 84, flaw profile curves for a rectangular EDM slot, calculated
calculating the b-field by the Wiener-Hopf technique. from the analytical model of Ref. 85. This was compared
Further applications of this technique are discussed in in Ref. 91 with profile curves obtained for the same
Section 2.3.3. problem using the finite difference method. Curves for
2.3.2.1.2. Thick Skin Regime. The seminal work of semielliptical EDM slots, obtained by finite differences,
Burrows(1) falls into this category. In this case, surface- were also presented in the same paper. These results were
breaking cracks and EDM notches were modeled approxi- all for an absolute probe. Similar results were given in
mately by bisecting a thin ellipsoidal void and placing it Ref. 20 for a two-port (reflection) probe, with the numeri-
at the testpiece surface. In Ref. 84 two-dimensional and cal procedure extended to the three-dimensional case
Advances in Eddy Current Nondestructive Evaluation 17

(asymmetric probe scans), as in Refs. 64 and 87. This been appended to the dipole layer distribution function
work is detailed in Section 3 as an illustration of the as an indication that this is associated with the b-field.
forward problem. Reference 97 reviews the application the extended scalar
2.3.2.2.2. Finite Element. Finite-element models(92-94) potential representation(82) to a complex surface breaking
have recently been developed for surface-breaking cracks crack problem. The status of software implementation
and EDM notches. A reciprocity relation was used to for in-service eddy current inspection and simulation,
show that (20) is equivalent to the electrical terms in based on the extended scalar potential method, is reported
(10b), where the integral is performed only over the flaw in Ref. 81.
region, with the advantages noted in Section 2.3.1. These
new models can also extend the treatment to two-port
probes and evaluate AZ from the electrical term in (16b). 2.3.3. Subsurface Cracks
2.3.2.2.3. Volume Element. Conversion of (20) to 2.3.3.1. Approximate Analytical Methods. In the
an integral over only the flaw region has also been applied class of subsurface cracks the thin skin regime is not
to volume element modeling.(59,61,95) Because VEM is relevant, except as a description of an almost surface-
based on the use of a Green's function, this reduces the breaking crack. The following examples therefore con-
amount of discretization required. sider the skin depth to be comparable to or greater than
2.3.2.2.4. Boundary Element. This approach is much the crack dimensions. Burrows' ellipsoidal void model(1)
more widely used at present than volume element for applies to deep subsurface cracks in the thick skin regime.
modeling eddy current problems. References 64 and 87 There exist only a few references that deal analytically
applied BEM to the model in Ref. 85, with an improved with more realistic subsurface crack problems. Three of
scalar potential formulation and application to more com- these applied the Wiener-Hopf technique to solving b-
plex geometries. A general review of the boundary-ele- field problems for subsurface cracks. Undercladding
ment method was given in Ref. 96. References 65, 66, cracks were treated in Ref. 98 and pages 68-81 in Ref.
and 67 illustrated application of the method to a complex 99, and subsurface cracks in noncladded testpieces were
geometry of great practical importance—the corner studied in Ref. 100.
crack. An ideal, or zero-thickness, surface-breaking crack 2.3.3.2. Numerical Methods. 2.3.3.2.1. Finite Ele-
was studied in Ref. 69. By considering the electromag- ment. References 55 and 99, Appendix F, apply this
netic continuity conditions across the plane of the crack, method to subsurface cracks, the first citation for subsur-
it was shown that the jump in the transverse electric field face in uncladded testpieces and the second for interface
at the crack is given by the surface gradient of a surface cracks in cladded testpieces. General three-dimensional
scalar function p, now commonly known as the Bowler FEM software suitable for subsurface cracks and other
potential, general problems is described in Ref. 101.
2.3.3.2.2. Volume Element. The surface crack model
in Ref. 61 could be extended to subsurface cracks, and
Ref. 102 applied the volume-element method to a general
where s is the testpiece conductivity. The crack is repre- subsurface defect. This could be specialized to subsurface
sented by an equivalent secondary source consisting of and interface cracks.
a layer of current dipoles directed normal to the crack sur- 2.3.3.2.3. Boundary Element. The surface crack
face, model in Ref. 69 could be extended to subsurface cracks,
and the general boundary-element procedure described
in Ref. 96 could be specialized to subsurface and inter-
The function p is obtained by solving the scattered field face cracks.
problem by BEM, with the current boundary condition

2.3.4. Other Types of Defects—Pits, Voids, Inclusions,

at the crack, and constructing the b-field as the sum of and Nonlocal Inhomogeneities
the a-field and the scattered field. The electrical term in 2.3.4.1. Approximate Analytical Methods. In the
(10b) then becomes thick skin regime, the ellipsoidal model of Ref. 1 is appli-
cable to pits, voids and inclusions, and Ref. 89 could be
applied to arbitrarily shaped defects of the classes listed
under this heading, again in the thick skin regime. Some
where S0 is the surface of the crack and a subscript b has of the other analytical references cited here are for the
18 Auld and Moulder

Fig. 6. Single surface layer of thickness l (after Ref. 99).

thin skin regime: corrosion and other inhomogeneities an angular spectrum of plane waves. In this case, the E
were studied, in Ref. 103 using spatial frequency analysis, and H fields are expanded directly rather than using vector
pits in the form of a spherical cap cavity interrogated by and scalar potentials
a uniform current field in Ref. 104, spatial frequency
(angular spectrum) analysis at low temporal frequencies
of spheroidal defects,(30) shallow flat-bottomed holes
interrogated by a uniform field probe,(16) low-frequency
asymptotic solutions for irregularly shaped inhomogenei-
ties,(72) corrosion pits interrogated by the remote field
eddy current method and modeled in the Born approxima-
tion,(51) and hemispherical pits interrogated by an
ACFM probe.(105) where z is the layering direction (Fig. 6) and the Fourier
2.3.4.2. Numerical Methods. The finite-element components individually satisfy Maxwell's equations. In
method was applied to a through hole in a tube wall(18) this way, the quasistatic approximation is automatically
and to oxide buildup in nuclear steam generator tubing.(106) introduced as the angles of incidence of the angular plane
Sophisticated numerical software is now available(62,82,94-96) waves increase. In the general case, Fourier components
for a much wider range of modeling applications to all polarized both perpendicular and parallel to the plane of
classes of defects in this category. incidence are required. Axisymmetric probe coils require
only parallel polarized components. For a one-port probe
2.3.5. Layers and Topography the a- and b-fields in (10a) are both expanded as in (34),
and SF lies along the dashed line in the figure, with
2.3.5.1. Analytical Methods. 2.3.5.1.1. Layers. The closure at infinity where the quasistatic field makes no
surface (and subsurface) layer b-field problem is of very contribution (Appendix A). Only the surface-tangential
significant importance for quality control of deposited field components enter into the integrand of (10a) and
layers, and for detection and characterization of defects the E-field components can be related to the H-field com-
in the form of a layer or layers—i.e., corrosion. This is ponents through the a-subscripted and b-subscripted sur-
one of the few b-field problems for which there exists a face impedances at the dashed line in the figure, where
rather complete set of exact analytical solutions. The the cross-hatched region is a deposited layer and the
earliest examples(39-41) deal with both planar and cylindri- region labeled "Metal" is the uncoated testpiece. The
cal geometries. Equation (20) gives AZ in terms of an final result is
integral over the probe coil. But this is equivalent to
taking the difference of Zb and Za. For small AZ it is
advantageous precisionwise [see (22) and (23)] to evalu-
ate the a- and b-fields in the layer and then apply the AZ
relation (10a) and (10b).
In the references cited, the field solutions are with
obtained in the form of Fourier-Bessel expansions, which
limit the method to axisymmetric problems. A more gen-
eral procedure is the spatial frequency (or angular spec- where the b-configuration is as shown in Fig. 6 and l = 0
trum) method,(30,99,103) where the fields are expressed as for the a-configuration. The subscript t denotes magnetic
Advances in Eddy Current Nondestructive Evaluation 19

fields in the xy plane. This formalism is applicable to In a topographical problem the angular Fourier com-
nonaxisymmetric probes, both one- and two-port probes, ponents are coupled by the topographical relief and (35)
and multiple layers. Layers with continuously variable is not applicable. But, in the spirit of the thin skin approxi-
properties are treated in Refs. 107 and 54. mation,(84-88) field nonuniformities near steps and edges
Modeling of corrosion layers in lap junctions are ignored to a first approximation, and in the piecewise
deserves special mention at this point. Defects of this uniform regions of the topographical structure the surface
type are commonly modeled as subsurface air gap layers impedance is defined as for a normally incident plane
because of the low conductivity of corrosion products.(108) wave—a reasonable assumption when the fields have
2.3.5.1.2. Topography. Eddy current probes have slow spatial variations. Substituting this spatially varying
been designed to determine the position and orientation surface impedance into (16a) then yields
of surface topographical features such as steps and
edges(109) Figure 7 illustrates a crossed differential two-
port probe used for this purpose. The two differential
pickup pairs are read individually. When the probe is
oriented such that one pair is parallel and the other perpen- This approach yields satisfactory qualitative agreement
dicular to an edge, the first pair does not respond to the with experiment.
edge and the second pair has a maximum output. But 2.3.5.2. Numerical Methods. A wide range of sophis-
small changes in the orientation of the edge are easily ticated software is now available for modeling layer and
detected by the first pair.(110,111) Figure 8 illustrates the topographical problems. The extended scalar approach
response of this probe to a curved step. to the boundary element method(82) is especially worthy

Fig. 7. Topographic sensor probe (after Ref. 109, with permission).

20 Auld and Moulder

Fig. 8. Response of a curved step (after Ref. 110, with permission).

of note with regard to edge and step modeling for topo- defect responses AZ have been displayed as trajectory
graphical problems. curves on the impedance plane illustrated in Fig. 10.
Examples are given in Figs. 4 and 5 of Ref. 94. This
display has the advantage of showing the angular differ-
2.3.6. Liftoff, Tilt, and Material Characterization
ence in the impedance plane between the liftoff signal and
Quasi-random variations of the liftoff distance /, the defect signal. By arranging the detection electronics to
where the cross-hatched region in Fig. 6 is an air gap, read only impedance plane variations perpendicular to
and of the tilt angle in Fig. 9(112) create a large "clutter" the liftoff trajectory (Q axis in Fig. 10), the liftoff "clutter"
signal that can mask the defect signal during scanned signal can be minimized. Alternatively, the flaw profile
probe measurements unless special precautions are display of the amplitude and phase of AZ as a function
taken.(113) Fixed probe measurements(112) and vertically of a probe scan position along the plane of a crack—the
scanned probe measurements, for material characteriza-
flaw profile in Ref. 86 and Fig. 7 in Ref. 94 will be
tion, are also subject to errors due to inaccuracies in the
seen to be an especially convenient display for inversion.
liftoff and tilt positioning.
Details are given in Section 4.
2.3.6.1. Analytical Methods. 2.3.6.1.1. Liftoff. The
2.3.6.1.2. Tilt. No analytical evaluations of the tilt
Fourier-Bessel(39-41) and angular Fourier(30,99,103) expan-
sion methods described in Section 2.3.5 for layer prob- effect have been made. Figure 11 illustrates a possible
lems are directly applicable to liftoff when the cross- approach to this problem, by the angular spectrum
hatched material layer in Fig. 6 is replaced by an air gap. method. The dashed line is a part of the surface of integra-
The a-field evaluations for air-core coils of complicated tion SF in (10a). This surface is closed at a distance from
shape in Refs. 9, 10, 13, 14, and 28 from Section 2.2.1 the probe where the integrand becomes negligibly small.
can be applied to two liftoff distances, using either (20), 2.3.6.1.3. Material Characterization. This topic has
or the AZ relation (10a) and (10b) to evaluate the AZ due already been covered in Section 2.3.5.3.1, where it is
to the change in liftoff. Figure 10 shows the variations shown in Fig. 10 that the angle of the liftoff trajectory
of the real and imaginary parts of AZ as a function of depends on the material conductivity through the skin
the liftoff distance Z0, with l = 0 in Fig. 6, and as a depth 8.
function of the testpiece skin depth. Traditionally scanned 2.3.6.2. Numerical Methods. All of the previously
Advances in Eddy Current Nondestructive Evaluation 21

Fig. 9. Errors due to probe tilt (after Ref. 112, with permission).

described numerical procedures are applicable to liftoff A first approximation assumes that the tangential
calculations, using either (20) or (10a) and (10b). component of the magnetic field in the open mouth of
the flaw (Fig. 5) is the same as at the same position on
the surface of the unflawed metal (Fig. 5a). That is,
3. THE FORWARD PROBLEM

where capital subscripts are used to avoid confusion with

3.1. Probe-Defect Interaction a and b in (10a) and (16a). One argument in favor of
Analysis of probe-defect interactions in the thin this approximation is that an almost-closed surface flaw
skin approximation is demonstrated for a one-port probe produces only a small perturbation of the magnetic field
and a surface-breaking crack or EDM notch. In (10a) the above the metal surface. This is in the same spirit as the
surface SF, which can be any surface enclosing the crack, Bom approximation used for weak internal inclusions
is first chosen to lie on the walls and mouth of the crack, (25a) and (25b). Otherwise, the approximation may be
as illustrated schematically in Fig. 5. Figure 5a shows considered as the first step in an iteration process, where
the position of SF in an unflawed testpiece. The a-field the interior and exterior magnetic fields in and around
is calculated by one of the methods in Section 2.2, while the flaw are calculated in alternating steps.
the b-field is analyzed in the thin skin approximation. On the parts of SF that coincide with metal surfaces,
For an "almost-closed" surface-breaking crack this the surface area M in Fig. 5a and the inner walls of the
approach has been found to be in good agreement with flaw in Figs. 5b and c, the boundary condition relating
experiments in nonmagnetic testpieces. Figure 5b illus- the magnetic field, which is tangential at the surface of
trates the problem to be solved, a surface crack (or EDM a good conductor, to the tangential electric field is defined
notch) in a metal of finite conductivity. In an intermediate by the skin impedance. That is, for the A-, B-, and C-
step the analysis treats the same crack in a perfectly subscripted fields
conducting metal, shown in Fig. 5c.
22 Auld and Moulder

Fig. 11. Schematic for evaluation of AZ from (10a) due to tilt, by the
angular spectrum method (after Ref. 99).

same as if the walls were perfectly conducting (Fig. 5c).

That is, the approximation

can be applied.
In summary, the fields under the integral in (10a)
are obtained for the defect in Fig. 5b by first finding the
magnetic field Hc inside the idealized defect of Fig. 5c,
subject to the boundary condition (37) in the mouth of the
flaw, and then approximating the B-subscripted magnetic
field by (39). The tangential electric field EB on the flaw
walls is then obtained from (38b). In the mouth area
M the B-subscripted electric field must be determined
indirectly, as described below.
At this point it is convenient to enlarge the enclosing
Fig. 10. "Comma" curves showing the variation of liftoff AZ with z0 surface SF so that it coincides with the surface of the
and 8 (after Ref. 113). flawed testpiece. On this surface the a- and b-subscripted
E fields (in 10a) differ only on and near the crack mouth
opening M. As a final approximation the integration is
on M in Fig. 5a, and restricted to the crack mouth itself, reducing (10a) to

on the inner walls in Figs. 5b and c. [Recall that the

normal vector n is directed toward the outside of SF (Fig. where the first term represents the removal of a part of the
1). It is therefore directed away from the metal surface unflawed surface and the second represents the addition of
in (38a), but toward the metal surface in (38c).] The skin a defect. [A corresponding result for a two-port probe
impedance in (38b) and (38c) is defined as is obtained from (16a).] The first term in (40) can be
reduced to

where a is the metal conductivity, and 8 is the skin depth

as defined in Eq. (18). This is an approximation, using the
skin impedance of a uniform plane surface and completely
neglecting the disturbance of eddy current flow at the tip
and edges of the defect. (These effects are considered in by a vector triple product manipulation (Appendix B)
Section 3.3, below.) and application of (37). The term on the right is obtained
The second approximation is based on the assump- from the boundary condition (38a).
tion of a highly conducting metal in Fig. 5b. In this case The next step is to convert the second term in (40)
it can be shown, from microwave resonator theory,(2,3) to a form in which the magnetic field solution of Fig. 5c
that the magnetic field inside the flaw is essentially the can be used (Appendix C),
Advances in Eddy Current Nondestructive Evaluation 23

of the magnetic field in (43) therefore reduces to a solu-

tion of a Laplace problem with well-defined boundary
conditions. This can be carried out analytically for rectan-
gular-shaped flaws, where a series solution for the interior
magnetostatic problem can easily be obtained. This is not
the case for the semielliptical shapes encountered in real
This substitution simplifies subsequent numerical evalua- fatigue cracks. For these geometries, the interior Laplace
tion of AZ. Since the Lorentz reciprocity relation (8a) problem must be solved numerically, by either finite-
applies to any two field solutions, it can also be written difference or finite-element methods. In any case, it is
in terms of the B-subscripted and C-subscripted field convenient at this point to express (43) in quasistatic
solutions in Fig. 5 for all points inside SF. Conversion to a format using a scalar potential to represent the magnetic
surface integral by application of the divergence theorem field in the interior of the flaw,
then yields

where P is a solution to Laplace's equation satisfying (37)

By vector triple product manipulation (Appendix C), the and the perfectly conducting wall boundary conditions in
wall contribution in (42b) is converted to Fig. 5c. The following analysis originates with Refs. 84
and 85. Scalar potential modeling is also used by Beis-
sner,(64,87) Burke,(86) and Hartfield and Bowler.(88) The
extended scalar potential formulation by Nakagawa and
Chao(82) puts the scalar potential method on a rigorous
The first term is zero from (38c). After substitution of basis and renders it applicable to complicated geometries
(38b) and (39) the second term becomes such as edges.(83)
The quantity integrated over the flaw mouth on the
right side of (43) is zero on the flaw walls because, from
Appendix C,
Combining these results in (42a) and (42b) and applying
(37) yield the final result (Appendix B),

and this is zero on the walls (38c). Consequently, the

integral can be extended over all of SF, and the use of
the divergence theorem yields

Here the left-hand side is the total power delivered to the

flaw, the first term on the right is stored energy inside
the flaw, and the remaining term is dissipated power.
Assembling (40), (41), and (43) then gives a general
probe-defect interaction relation for almost-closed sur-
In the quasistatic approximation (44) the volume integ-
face-breaking cracks and EDM notches.
rand above can be written

3.2. Quasistatic Formulation

Since eddy current probes operate at very low fre-
quencies, where the dimensions of the probe and of the
defect are much smaller than a wavelength, the magnetic
field Hc (Fig. 5c) in the interior of a surface-breaking since P is a magnetic Laplacian potential. In the qua-
defect can be represented by the gradient of a scalar sistatic regime, where (44) applies, the first approxima-
potential. But note that for calculating the magnetic field tion to EC is given by the Faraday induction relation,
outside the surface of the testpiece, where currents are
present, the vector potential must be used. Calculation
24 Auld and Moulder

and it follows that and the crack lips correction is

in (43). Similarly, the second integral on the right side where X extends along the crack tip and / along the crack
of (43) becomes, from (44), lips. These correction terms are to be added to the right
side of (40). Since the changes in eddy current distribution
occur within a distance of several skin depths from the
tip and the lips of the crack, it is clear that these correction
terms become much smaller than dF for cracks whose
Finally, in (40), dimensions are much larger than a skin depth.

3.4. Analytic Evaluation of AZ

The interior magnetic field is defined in the qua-
sistatic approximation by (44), with
is expressed entirely in terms of the interior Laplace
solution to the problem illustrated in Fig. 5c. In evaluating
d0 in (41) the field Da is obtained from the Dodd and [Note that, according to the approximation (39), the inte-
Deeds vector potential calculation(39-41) where the qua- rior magnetoquasistatic potential field for Fig. 5c is identi-
sistatic approximation has already been applied. Combi- cal to the interior magnetostatic potential field for Fig.
nation of these results in (40) gives the final quasistatic 5b.]
formulation of AZ in terms of the unperturbed probe field In the magnetoquasistatic approximation, the bound-
in Fig. 5c and the interior Laplace potential in Fig. 5c. ary condition (38c) is rewritten as
For an almost-closed surface crack d0 is usually much
smaller than dF and can often be neglected in a first
approximation. following the waveguide analysis in Ref. 3; and, in terms
of the magnetostatic potential,
3.3. Crack Tip and Lip Correction Terms
Near the tip and lips of a surface-breaking crack,
the eddy currents in the metal no longer flow parallel to on the inner walls of the surface-breaking crack. This
the surface, as they do in a flat metal specimen. An can be restated as
analysis of this phenomenon carried out by Kahn(63,76)
for a two-dimensional surface-breaking crack shows that
the eddy current is concentrated where it flows around
the tip of the crack and avoids the lips at the mouth of and in the mouth of the flaw, (37) requires that
the crack. In other words, the current flow behaves like
water flow around an obstacle. These changes in current
distribution, which are localized over distances on the where the subscript T denotes a vector in the plane of
order of a skin depth, modify the wall integral contribu- the crack mouth. The field HT can be obtained from the
tion considered in (47).(84,85,88) In applying this two- a-solution in Fig. 12.
dimensional analysis to the general problem above, the To illustrate, consider the rectangular EDM model
corrections given in the reference cited are weighted at in Fig. 12a. This geometry is, in fact, a short-circuited
each point on the crack periphery by the magnitude rectangular waveguide that is below cutoff for all wavegu-
squared of the local magnetic field. In the quasistatic ide modes. When w is far below cutoff the TE modes
approximation the tip correction term is then given by become magnetoquasistatic in nature(3) and constitute an
orthogonal basis set for solving the problem at hand. The
quasistatic modal solutions can be shown to be, from the
reference cited and the boundary condition (49b),
Advances in Eddy Current Nondestructive Evaluation 25

Fig. 12. Surface-breaking crack models. (a) Rectangular slot (EDM notch), (b) Semielliptical (or semicircular) slot.

where the subscript T denotes the transverse (or xy part)

of the field.
The mode expansion for the transverse part of the
interior field of the flaw can be written

with and, using the boundary condition (37), the expansion

coefficients amn are obtained by applying the orthogonal-
ity condition (51). That is, from Ref. 3,

These solutions satisfy the following orthogonality rela-

tion [3] in the mouth of the flaw,

The integral is evaluated using the a-solution in Fig. 12

to obtain HT. The probe-crack interaction formula
26 Auld and Moulder

This relation reduces (47) to

is then evaluated by substituting (52) into (47) and (48).

This general three-dimensional formalism allows the
crack to be placed either symmetrically (Fig. 13a) or
asymmetrically (Fig. 13b) with respect to the probe. A The analytical approach in Section 3.4 is applied to the
scan of AZ versus probe position (crack profile) can be rectangular two-dimensional problem by taking the mode
made either in the plane of the crack, as in the figure, or index n to be zero in (52a), which restricts the field
normal to the crack plane. The formalism can also be expansion to modes that are uniform across the dimension
applied to very shallow open defects, or recesses, where b = Au of the crack (Fig. 13a).
the Born approximation (37) can again be applied, justi- For numerical evaluation of the two-dimensional
fied by the fact that the defect is very shallow. problem, it is convenient to convert the wall integral of
(55) into an integral over the flaw mouth. From Green's
theorem and (49a)

3.5 Numerical Evaluation of AZ

For semicircular and semielliptical slot geometries An integration over the interior volume of the flaw and
(Fig. 12b), the AZ formula cannot be easily evaluated by application of the divergence theorem yields
an analytical approach. In such cases Az was first evalu-
ated by the finite-difference method.(20,91) Only the case
of an almost-closed crack, with the probe centered over
the plane of the crack and scanned in the plane of the crack
Fig. 13a, is considered here. The fields and potentials
are, in this case, approximately uniform across the flaw
opening, and the three-dimensional (x, y, z) problem
From (49b) the right-hand integrand in (56a) is nonzero
detailed in Section 3.1 can be reduced to a two-dimen-
only over the mouth, and substituting from (54) yields
sional (x, z) problem. In applying the method it is useful
to manipulate (47) so that only one integral is involved.
First, it is noted that for the two-dimensional problem
the wall integral in (47) is twice the integral over one
wall. Furthermore, the volume integral is the integral over Then (55) becomes
one wall multiplied by the crack opening AM,

Fig. 13. Defect profile scans. (a) Symmetric. (b) Asymmetric.

Advances in Eddy Current Nondestructive Evaluation 27

When P and (dP/dn) are approximated as uniform across

the opening, the right-hand integral (Fig. 13a) in (57)
becomes

where the AM in (58) cancels the Au [in (57)].

The potential function boundary condition in the
mouth of the flaw is defined by the unperturbed interro-
gating field of the probe, from (37), and integration of
(49c). It is clear from (55) that AZ is independent of
the arbitrary constant of integration. The wall boundary
condition (49b) completes the statement of the interior
potential problem. Finite-difference solutions(20,91) are
generated numerically by starting with a coarse finite-
difference mesh and then successively refining the mesh
size until the desired convergence is obtained. In this
procedure new mesh values were obtained from the old
values and interpolation.
Special care must be taken in applying boundary
conditions on a semicircular or semielliptical bound-
ary.(91) Starting from a rectangular mesh, nodes are Fig. 14. Schematic representations of defect profile curves. (a) Small
defined as exterior, interior, or border points. Exterior probe diameter. (b) Large probe diameter. (After Ref. 80, with permis-
sion.)
points are ignored, while interior points are treated by a
standard fourth-order approximation. Exterior points are
ignored, while interior points are treated by a standard
fourth-order approximation. Border points are catego-
rized using geometric relationships relative to the bound-
ary. In the iteration procedure pointwise successive properly choosing the probe diameter. Dashed circles
relaxation is used, automatically adjusting the overrelaxa- indicate the position of the probe coil and the eddy current
tion parameter to avoid divergence. From the interated vortexes in the test piece under the coil.
solution the normal derivative required in (57) is calcu- In Fig. 14a small flat spots appear in the profile
lated numerically and the integration along the flaw when the center of the vortex is over one end of the
mouth is carried out numerically by Simpson's rule for crack. The reason is that the eddy current strength goes
unequal intervals. The singularities of the normal deriva- to zero at the vortex center, so that the probe signal does
tive at the ends of the flaw cause some slowing of the not change for small probe displacements near the end
convergence, but this is not a serious problem. of the crack. In Fig. 14b, where the crack length is much
Comparisons of experiment and theory in(20,80-91) smaller than the probe diameter, the flat spots do not
provide an empirical justification for assuming P and appear. The flaw profile scan then simply displays the
(dP/dn) to be uniform across AM for almost-closed surface squared profile of the probe field distribution.
breaking defects. As noted in Section 2.2.1 the depth of penetration
The above numerical evaluation of probe response of the probe field into the substrate is related to skin
to a symmetrically scanned surface breaking crack (Fig. depth in the thin skin limit and to the coil diameter in
13a) generate a flaw profile (Fig. 14) that directly indexes the thick skin limit.(42) These features must be kept in
the ends of the crack. Figure 14 clearly shows the physical mind when applying the flaw profile method to defect
reason for this behavior and illustrates the importance of characterization.
28 Auld and Moulder

4. INVERSION of the extreme degree of attenuation in an eddy current

medium. But it should be noted that the principles of
geometric optics have long been applied to dissipative
4.1. Mathematical Inversion, Imaging, Defect
media(115) so that it is logical to speak of eddy current
Reconstruction, and Defect Characterization
imaging, first demonstrated by Copley(116) as a means to
General eddy current inspection for defects com- classify partially defects as to type (cracks, voids, etc.).
prises three functional steps: detection, location, and The diffusion equation governing eddy current behavior
description. The third step is very commonly termed is, in fact, the limit case of the dissipative wave equation
defect sizing. In the forward problem AZ of the probe is when the losses become very large.(117) However, in eddy
analyzed for a particular type of defect in a particular current imaging, the complicated geometric shapes of the
location, as a function of the scanned position of the object (defect) and the source (probe), as well as the very
probe. In the inverse problem, the aim is to determine large material losses and interface reflections, lead to
the nature and location of a defect from the scanned AZ very severe imaging aberrations. The result is that the
data output of the probe. image does not closely resemble the object. Various image
In early eddy current practice (Chapters 2 and 5 in processing and restoration techniques have been applied
Ref. 4) the approach was qualitative and empirical, using to the problem, including what is commonly termed eddy
equivalent circuit conceptual models and experimental current holography.(118,119) Holographic (or "complete")
simulated defect models, fabricated to varying degrees of recording was developed in optics as a means for
realism. Recorded data were displayed in the impedance recording both amplitude and phase information as ampli-
plane format (Fig. 10) for phase discrimination against tude-only information, permitting restoration of a three-
the time varying liftoff signal generated by a hand-held dimensional image. In the eddy current case, the very
probe scan. The nature of a defect was deduced by com- severe attenuation of the medium requires certain modifi-
paring the impedance plane trajectory for the defect with cations of the technique, including use of an electronically
impedance plane trajectories for a selection of simulated injected phase reference signal rather than the phase refer-
defects. In other words, the identification procedure was ence beam used in optics. The phase of the recorded AZ
a qualitative and subjective form of pattern recognition. data, which encompasses a very limited range, is also
But even at the time (1965) the need for a more quantita- augmented (or multiplied) electronically. For these rea-
tive approach was recognized (Chap. 10 in Ref. 4). sons it may be reasonable to use the term "phasogra-
Programs of research targeted to quantitative nonde- phy"(120) to distinguish the eddy current technique from
structive evaluation were first initiated in the 1970s, lead- that used in optics.
ing to the organization of a series of annual conferences The following lists selected references to illustrate
devoted to the subject and beginning publication in 1982 the details of commonly used inversion procedures.
of Volume 1 in the series of annual bound conference
proceedings, Review of Progress in Quantitative NDE.
For eddy current NDE a rapid increase in the sophistica-
tion of modeling and experimental techniques resulted. 4.1.1. Mathematical Inversion
As analytical and numerical models evolved, it became
possible to reconstruct defect geometries by fitting In Ref. 121 a direct linear inverse method is pre-
recorded AZ data with computed AZ data for assumed sented for eddy current measurement of three-dimen-
types of defects. This defect reconstruction procedure is sional conductivity measurements. The form of the probe
now widely practiced. Improved mechanical scanners and is a spatially periodic current sheet. It is supposed that
electronic instrumentation provide an extensive choice the conductivity variations are small so that the Born
of AZ data display modes, including two-dimensional approximation (25a) and (25b) may be used, thereby
scanned amplitude and phase images that permit classifi- linearizing the inversion equations. Inversion of the three-
cation of defects into their particular categories as dimensional problem requires inversion of a coupled Fou-
required for implementation of inversion procedures. rier-Laplace transform. In one-dimension, the depth and
Inversion of the system of governing equations for conductivity of coatings may be reconstructed exactly
the interaction of the probe with a defect has been per- from the inverse Born profiles. Reference 122 uses the
formed rigorously (mathematical inversion) for some spe- same probe structure to obtain by inversion, in the asymp-
cial problems that satisfy the required invertibility and totic low-frequency limit, the exact dc conductivity of a
uniqueness conditions. These conditions are more severe surface layer on a substrate of Ref. 85 of different conduc-
for eddy current, rather than optical problems, because tivity.
Advances in Eddy Current Nondestructive Evaluation 29

4.1.2. Imaging solution of a two-dimensional Laplace analysis for the

crack interior. The problem is solved for a crack of general
Reference 116 first introduced the practice of using
shape using the boundary-element method. The forward
imaging to categorize quickly a defect according to its
model in Section 3 also represents the interior field of
general type. An electromagnetic model for eddy current
the crack in terms of a two-dimensional Laplace potential
imaging, based on use of the AZ relations, was later used
(magnetostatic, in this case), so that the same inversion
to compare optical and eddy current imaging systems
procedure(128) may again be adopted, using the boundary-
with regard to aberrations, point spread functions, and
element method and iteration.
the choice of image processing techniques.(123) In Ref.
A very different approach is followed in Ref. 110
124 image restoration was formulated as a maximum-
for the problem of reconstructing electrical conductivity
likelihood estimation problem, treated by constrained
profile distributions. Variable frequency data, with vary-
iterative gradient descent. Synthetic images of a number
ing penetration depths, make it possible to recover the
of defect configurations were created. Eddy current
conductivity as a function of depth. Reference 129
holography (or phasography) using phase reference multi-
describes an iterative least-squares method for inverting
plication to overcome the problem of small eddy current
eddy current probe impedance data, based on analytically
penetration and to achieve adequate resolution was dis-
formulating the spatial gradient of the scanned probe
cussed in Refs. 118 and 119. Imaging with eddy current
impedance data. An important feature is that this result
arrays was reviewed in Ref.37, where the advantages of
is independent of the type of numerical discretization
comprehensive array generated data sets were compared
used. Examples of a layer problem and an ellipsoidal
with traditional impedance plane Lissajous patterns from
crack are given. In Ref. 130, the impedance gradient
a single line scan.
approach is applied to the inversion of flaw profile data(91)
for both semielliptical and irregular cracks. Reference
131 presents a defect reconstruction procedure based on
4.1.3. Reconstruction
the forward model in Section 3, for a long crack of uni-
Image restoration in Section 4.1.2 was concerned form depth and opening, following the thin skin treatment
with maximizing the information content of the two- of an infinite (two-dimensional) crack.(85) The inversion
dimensional image recorded by the probe and with procedure performs a least-squares polynomial fitting of
improving the sharpness and resolution of the detail. the data. Use of eddy current arrays for defect reconstruc-
Defect reconstruction, in contrast, deals with estimation tion has been treated in detail in Ref. 37. Arrays have
of the defect's boundary shape (crack, pit, void, etc.) or the useful feature that data may be taken with variable
the spatial distribution of conductivity changes (inclu- geometry arrays(35) to acquire a more complete database.
sions, layers, etc.). The procedure used is to minimize In Ref. 37 three-dimensional reconstruction is discussed
the error between the measured data and the data produced and the relation of this procedure to the restoration of
from a field model of the probe-defect interaction, by the defect's two-dimensional top view is analyzed.
varying the defect parameters. Details of the minimization
procedure vary among the references cited. Some treat-
4.1.4. Characterization
ments chose a more restricted range of defect parameters
than others. Defect characterization is defined here as the proce-
In Refs. 84 and 85 the model used is the thin skin dure that estimates some of a defect's parameters by
model and the defect is a rectangular slot, while Refs. matching certain features of the experimental and model
125 and 126 present a numerical method applicable to data sets, rather than by performing an iterative least-
two- and three-dimensional conductivity distributions, squares fitting of the complete data sets, as in reconstruc-
following contemporary work in electromagnetic and tion. This approach relies heavily on physical insight and
geophysical inversion. Reference 127 begins with a brief visual selection of the features to be matched and the
general discussion of the eddy current inversion problem parameters to be characterized. The result is a rapid and
and then formulates the problem of a surface-breaking efficient procedure, realized at the cost of a reduced infor-
crack as a parameter estimation problem using a finite- mation output concerning the geometrical shape of the
element model of the forward problem. The same defect defect. Reference 85 considers procedures that share to
geometry is treated in Ref. 128 in the context of the acfm some extent the reconstruction and characterization
technique,(13) but the method is also applicable to a related approaches to the problems of rectangular surface-break-
eddy current technique (79). In Ref. 128, the forward ing cracks and slots. In Refs. 80, 131, and 91 the length,
model is based on the "unfolding" technique,(78) involving depth, and opening of rectangular and semielliptical sur-
30 Auld and Moulder

face-breaking defects are estimated. The procedure is method with an exact formulation of the electromagnetic
based on flaw profile data (Fig. 14) and the sensitivity Green's function. Reference 130 applies this forward
of depth data to the diameter of the probe (Section 2.3.2). solution to an iterated minimization of the difference
Another approach to defect characterization, in Ref. 127, between estimated AZ and measured AZ. The procedure
works from data sets consisting of Lissajous trajectories uses a descent algorithm requiring knowledge of the gra-
in the impedance plane data display mode (Fig. 10). The dient of the error with respect to a variation of the crack
technique relies on a Fourier descriptor mapping to extract geometry. The boundary-element method is well adapted
and compress the data into a reduced set of signature (or to this kind of problem because it easily adapts to prob-
feature) parameters so as to reduce computational effort lems with irregularly shaped crack boundaries.
and improve performance of the classification procedure. Figure 15 illustrates a general inversion performed
Inversion of eddy current data from homogeneous by this technique. Following the example of Ref. 127,
and inhomogeneous surface layers has attracted increased this lengthy inversion procedure could be shortened by
attention in recent years. Reference 133 treats the problem using the approximate scalar potential model in Section 3.
of determining the thickness and conductivity of a uni-
form layer, while Ref. 134 solves the forward problem
4.2.2. Characterization
for a nonuniform layer. In Ref. 135 a feature-based proce-
dure for rapid inversion of layer thickness and conductiv- The inversion process can be further accelerated by
ity is described. Reference 136 gives a preliminary report matching only a finite number of flaw profile features.
on pulsed eddy current characterization of corrosion lay- From the flaw profile in Fig. 14 the length c of the crack
ers . As a result of recent advances in electronic instrumen- can be obtained visually. The crack depth a and crack
tation the quality of pulsed eddy current measurements opening AM can be obtained from the amplitude and phase
has substantially improved. Time-domain displays permit of AZ at the center point of the crack. It is found empiri-
new approaches to inversion and provide improved lift- cally that the amplitude and phase data, when plotted as
off discrimination. in Fig. 16 for a known crack length c and skin depth 8,
exhibit contours of constant a and constant AM that are
nearly orthogonal. Consequently, these parameters can
4.2. Surface-Breaking Cracks and Slots be estimated from the measured amplitude and phase
There is now a growing trend in eddy current practice coordinates. It should be noted in Fig. 16 that, as predicted
toward assembling sets of benchmark defect simulations, in Section 2.3.2, the method breaks down when the crack
together with corresponding analytical and numerical depth a is much greater than the probe radius ? (normal-
models.(86,94,137) These simulations and models are used ized to a).
as were those described in Ref. 4 but for much more
complicated geometries and much more sophisticated the-
4.3. Layers
oretical models. They are also applied as forward solu-
tions for defect reconstruction procedures.
4.3.1. Frequency-Domain Inversion
In Ref. 129 a layered conductor with unknown layer
4.2.1. Reconstruction
conductivities is treated as a test of the reconstruction
The forward problem for an ideal crack of arbitrary procedure. But for a single unknown layer the character-
shape is solved in Ref. 69 by the boundary element ization procedure in Ref. 135 is more efficient.

Fig. 15. Inversion by iteration of an irregular crack. The profile after 20 iterations is dashed; that after 100 iterations is bold.
(After Ref. 129, with permission.)
Advances in Eddy Current Nondestructive Evaluation 31

Fig. 16. Quasi-orthogonal contours of crack depth a and opening Au, in the magnitude-phase plane—McFettridge chart (after Ref. 131).

4.3.2. Time-Domain Inversion motivate a search for commensurate improvements in

modeling methods. The time-domain character of pulsed
Reference 136 demonstrated the characterization of
eddy current testing has not yet been fully exploited with
a multilayer model of corrosion in aircraft skin splices,
regard to both the forward and the inverse problems.
proving the efficiency of this procedure. It should be
It is highly desirable to begin searching for a time-
noted again that, because of the time delay of the probe
domain modeling methodology with a flexibility and
response, the sensitivity to liftoff is very substantially
completeness comparable to that of the AZ frequency-
reduced for time domain operation. In the reference cited,
domain approach reviewed in this paper. In pulsed eddy
standard probe coils (air and ferrite core) were used and
current measurements the attenuation in the test piece is
the time-domain signal was obtained from the frequency-
large, and the time delay is small. The result is that
domain signal by a Fourier transform. Figure 17 illustrates
changes in voltage at the probe terminals contain strongly
the raw experimental data and the inverted results.
overlapping reflected and incident pulses. This masks the
Modeling of probe fields specialized to the optimiza-
crucial defect information contained in the waveform of
tion of the probe-defect interaction and interpretation of
the reflected pulses.
the forward problem response has received some atten-
All of this suggests that one can profit by choosing
tion. Reference 138 considered optimized probe design,
working directly in the time domain, while Ref. 139 AF (the change in reflection coefficient at the probe termi-
nals), rather than AZ, as the defect parameter of primary
considered the probe-defect interaction. The second ref-
interest. This choice is reinforced by the fact that low-
erence worked directly with the time domain version
frequency directional couplers to separate the reflected
of the reciprocity relation, applied to a two-dimensional
wave from the incident wave are readily available in the
surface crack interrogated by a plane wave.
form of suitably designed bridge circuits.
To model the performance of a one-port pulsed eddy
4.4. AF (or AS) Relations current system, there exists a reciprocity-based AF rela-
tion of similar form that that of the AZ relation. From
The three pulsed eddy current references(136,138,139)
Ref. 6,
cited above considered either a simplified defect structure
(layers) or a simplified interrogation field (plane wave).
But recent advances in pulsed eddy current technology,
coupled with corresponding successes in defect detection,
32 Auld and Moulder

Fig. 17. Pulsed eddy current inversion of the thickness dimensions of a two-plate structure with an air gap. (a) Time-domain data.
(b) Dimensional characterization.

where P is the power in the incident wave, and the evalua- With this information, all of the modeling methods
tion is performed in the frequency domain. The change applied to AZ in the frequency domain become applicable
in the frequency spectrum of the reflected pulse is to AF in the time domain.
obtained by multiplying (59) by the frequency spectrum
of the incident pulse. Fourier transform of the result gives
the change in the time-domain wave form of the reflected
pulse, due to the presence of a defect. 5. SUMMARY
The integral in (59) is of exactly the same form as
the AZ, when subscript a is appended to the unprimed A full derivation has been presented for a set of AZ
field quantities and subscript b to the primed field quanti- relations that express in a unified format the response of
ties. Therefore, all of the modeling techniques developed an arbitrary (one-port, two-port, array) eddy current probe
for AZ are again applicable. Since pulsed eddy current to an arbitrary defect. This result is given as an integral
probes are customarily two-port, rather than one-port, the over the region of the defect, with the integrand expressed
required two-port variant of (59) can be obtained by in terms of the defect parameters (shape, material proper-
following the procedure in Section 2.1.2. This yields ties, etc.) and the probe field distribution in the vicinity
changes in the elements of the reflection-transmission of the defect both in the absence (a-field) and in the
coefficient matrix, i.e., AG11, AF12 = AF21 and AF22. presence (b-field) of the defect. The probe response is in
When P is normalized to unity at both ports, AG11, etc., the frequency domain, but can be converted to the time
represent changes AS11, etc., in elements of the scattering domain for pulsed eddy current applications(135) by taking
matrix for the two-port probe. Fourier transform.
Advances in Eddy Current Nondestructive Evaluation 33

Fig. 17. Continued

The AZ format is applicable to any eddy current skin approximation and to introduce the importance of
modeling problem and requires only calculation of the the mode of data display (impedance plane, flaw profile,
a- and b-fields, by either analytical or numerical means. image, phasography, etc.) with regard to defect feature
This permits treatment of classes of similar defects in a classification, optimization by signal processing, and sup-
unified way. Restriction of the integral to the vicinity pression of liftoff clutter. A brief review of inversion
of the defect (the "localization" property) improves the procedures is given, with a selection of reference exam-
precision of numerical evaluations of small AZ signals ples.
by calculating AZ directly rather than taking the small An alternative approach to reciprocity-based model-
difference between the much larger probe impedances Zb ing using AF (orAS) relations has also been introduced.
and Za. Furthermore, the AZ format reduces the computa- These relations lead to expressions for changes in the
tional load in numerical evaluations by restricting the reflection-transmission (or scattering) coefficients of a
discretization to a small region around the defect. And, probe, due to the presence of a defect. These quantities
finally, by focusing on the region of the defect, the AZ are given in terms of localized integrals over the region of
formulation fosters improved physical insight and the the defect. Although they are calculated in the frequency
development of analytical approximations. domain, these scattering coefficients may be used to eval-
References cited give examples of a-field and b- uate in the time domain changes in the reflected and
field calculations, both analytical and numerical, for a transmitted pulses generated by the presence of a defect.
wide variety of probes and defects. Emphasis is placed on Such a formulation of the pulsed eddy current problem
physical interpretations wherever possible. The specific retains all of the advantages of the AZ method reviewed
example of a surface-breaking crack or notch is chosen above, while operating in the time domain. It therefore
to illustrate solution of the forward problem in the thin promises to motivate the development of sophisticated
34 Auld and Moulder

approaches to modeling and inversion in pulse-based or a void it is convenient to shrink the surface SF around
eddy current NDE. the defect, so that VF becomes the volume of the inclusion
Future trends in eddy current testing emphasize a or void.
rapid evolution of new probe technology (arrays for rapid
scanning, SQUID detectors for increased material prepa-
ration, etc.), a continuing development of sophisticated APPENDIX C
modeling for complex geometries, and some continuing
extension into the high-frequency and microwave Applying the vector triple product identity
domains.

to the first term in the integrand of (40), with C = n,

APPENDIX A yields

As R -> oo, the a-field and the b-field in (8b) can

be expressed as superpositions of outgoing spherical Application of (38a) and (37) reduces this result to
modes,(8)

which generates (41).

The integrand of the first member in (42a) can be
and converted as follows, using (B1) and (37):

Because the spherical modes are orthogonal over the

surface of the sphere,(3) substitution of (A1) and (A2)
into the left side of (8b) yields
From (C.1), again,

and
so that the contribution of the enclosing infinite sphere
vanishes.
in the integrand of (42b).
To obtain (43), note from (42a) and (42b) that
APPENDIX B

In (10a) the surface SF enclosing the defect in Fig.

1 can be chosen in a variety of ways, depending on the
shape of the defect. In Section 2.2 this is demonstrated
for the case of a surface-breaking crack. For defects such
as inclusions and voids, it is convenient to express AZ
By using the same argument that was used in developing
as a volume integral rather than a surface integral. This
(B4), the integrand of the first term on the right can be
is accomplished by first using the divergence theorem to
expressed as
write (10a) as

Substitution of (C7) and (C8) into (42a) and (42b) then

yields (25).
where VF is the volume enclosed by SF in Fig. 1. Using
the identity
ACKNOWLEDGMENT

and Maxwell's equations, (B1) can be converted to the The invaluable assistance of Norio Nakagawa in
volume integral format (10b). In the case of an inclusion preparing the manuscript is gratefully acknowledged.
Advances in Eddy Current Nondestructive Evaluation 35

REFERENCES 36. M. Uesaka, T. Nakanishi, and K. Miya. Rev. Prog. QNDE

13A:327-334 (1994).
37. M. Gramz and T. Stepinski. Res. Nondestr. Eval. 6:157-174
1. M. L. Burrows. A Theory of Eddy-Current Flaw Detection, Ph.D. (1994).
dissertation, University of Michigan, Ann Arbor, MI (1964). 38. M. Uesaka, T. Nakanishi, K. Miya, H. Komatsu, K. Aoki, and K.
2. D. M. Pozar. Microwave Engineering, Addison—Wesley, Reading, Kasai. IEEE Trans. Magnet. 31:870-876 (1995).
MA (1990). 39. C. V. Dodd and W. E. Deeds. J. Appl. Phys. 39:2829-2838 (1968).
3. R. E. Collin. Foundations for Microwave Engineering, 2nd ed., 40. C. V. Dodd, C. C. Cheng, and W. E. Deeds. 7. Appl. Phys. 45:638-
McGraw-Hill, New York (1992). 647 (1974).
4. H. L. Libby. Introduction to Electromagnetic Nondestructive Test 41. C. V. Dodd, W. E. Deeds, and J. W. Luguire. Int. J. NDE 1:29-
Methods, Wiley-Interscience, New York (1971). 90 (1969).
5. A. J. Bahr. In Eddy-Current Characterization of Materials and 42. P. Stucky and W. Lord. Rev. Prog. QNDE 11A:299-306 (1990).
Structures, ASTM STP 722, G. Birnbaum and G. Free, eds. (1981), 43. W. Lord. Electromagnetic Methods of Nondestructive Testing,
pp. 332-347. Gordon and Breach, London (1985).
6. B. A. Auld. In Eddy-Current Characterization of Materials and 44. A. Krawczyk and J. A. Tegopoulos. Numerical Modeling of Eddy
Structures, ASTM STP 722, G. Birnbaum and G. Free, eds. (1981), Currents, Clarendon, Oxford (1985).
pp. 332-347. 45. S. Nath. Finite Element and Boundary Element Analysis of Elec-
7. A. J. M. Zaman, C. G. Gardner, and S. A. Long. J. Nondestr. tromagnetic NDE Phenomena, Ph.D. thesis, EE and Computer
Eval. 3:37-43 (1982). Engineering, Iowa State University (1992).
8. R. F. Harrington. Time Harmonic Electromagnetic Fields, 46. R. E. Beissner, J. Appl. Phys. 60:352-356 (1986).
McGraw-Hill, New York (1961). 47. N. Nakagawa. Rev. Prog. QNDE 10A:249-254 (1991).
9. S. K. Burke. J. Phys. D Appl. Phys. 19:1159-1173 (1986). 48. Y. Liu, N. Nakagawa, and F. Rizzo. Rev. Prog. QNDE 12A:235-
10. J. R. Bowler. J. Appl. Phys. 61:833-839 (1987). 242 (1993).
11. J. R. Bowler, L. D. Sabbagh, and H. A. Sabbagh. IEEE Trans. 49. K. Murphy and H. A. Sabbagh. Rev. Prog. QNDE 14A:267-
Magnet. 25:2650-2664 (1989). 274 (1995).
12. H. A. Sabbagh, J. R. Bowler, and L. D. Sabbagh. Nondestr. Test. 50. I. Elshafiey, L. Udpa, and S. S. Udpa. IEEE Trans. Magnet.
Eval. 5:67-79 (1989). 30:3160-3163 (1994).
13. J. Zhou, R. Collins, and D. H. Michael. Rev. Prog. QNDE 51. R. Beissner, G. L. Burkhardt, J. L. Fisher, and T. Kikuta. Rev.
13A:295-342 (1994). Prog. QNDE 13A:273-278 (1994).
14. J. M. Prince, B. P. Hildebrand, and G. L. Hower. J. Nondestr. 52. J. C. Moulder and N. Nakagawa. Rev. Nondestr. Eval. 4:221-
Eval. 12:209-217 (1993). 236 (1992).
15. J. C. Moulder, P. J. Shull, and T. E. Capobianco. Rev. Prog. QNDE 53. N. Nakagawa and J. C. Moulder. Rev. Prog. QNDE 13A:295-
6:601-610 (1987). 301 (1994).
16. J. C. Moulder, N. Nakagawa, and P. J. Shull. Rev. Prog. QNDE 54. E. Uzal, J. C. Moulder, S. Mitra, and J. H. Rose. J. Appl. Phys.
7A:147-155 (1988). 74:2076-2089 (1993).
17. A. J. M. Zaman, S. A. Long, and C. G. Gardner. IEEE Trans. 55. R. Palanisamy, R. B. Thompson, and D. O. Thompson. Rev. Prog.
Instr. Measure. 30:41-45. QNDE 3A:569-577 (1984).
18. N. Ida. Rev. Prog. QNDE 3A:547-554 (1984). 56. P. C. French and L. J. Bond. J. Nondestr. Eval. 7:55-69 (1988).
19. C. Yeh and R. Zoughi. Res. Nondestr. Eval. 6:35-55 (1994). 57. D. J. Lynch, A. Mahmood, L. D. Philipp, and O. H. Nguyen. Res.
20. B. A. Auld, J. C. Moulder, S. Jeffries, P. J. Shull, S. Ayter, and Nondestr. Eval. 3:221-234 (1991).
J. Kenney. Res. Nondestr. Eval. 1:1-11 (1989). 58. W. S. Dunbar. J. Nondestr. Eval. 7:34-54 (1988).
21. Y. Sun, H. Lin, Y. K. Shin, Z. You, S. Nath, and W. Lord. Rev. 59. D. McA. McKirdy. J. Nondestr. Eval. 8:45-52 (1989).
Prog. QNDE 9A:319-326 (1990). 60. R. E. Beissner. J. Nondestr. Eval. 13:175-183 (1994).
22. D. L. Atherton, W. Curza, T. R. Schmidt, S. Sullivan, and C. Toal. 61. J. R. Bowler, S. A. Jenkins, L. D. Sabbagh, and H. A. Sabbagh.
J. Nondestr. Eval. 8:37-43 (1989). J. Appl. Phys. 70:1107-1114 (1991).
23. D. L. Atherton and W. Curza. Res. Nondestr. Eval. 5:123-134 62. R. E. Beissner. J. Appl. Phys. 60:352-356 (1986).
(1993). 63. A. H. Kahn. J. Nondestr. Eval. 7:3-14 (1988).
24. K. P. Dharmasena and H. N. G. Wadley. Rev. Prog. QNDE 64. R. E. Beissner. J. Nondestr. Eval. 7:15-24 (1988).
10A:1111-1118 (1991). 65. N. Nakagawa. Rev. Prog. QNDE 10A:249-254 (1991).
25. H. N. G. Wadley, K. P. Dharmasena, and H. S. Goldberg. Rev. 66. N. Nakagawa, S. Mitra, and J. C. Moulder. Rev. Prog. QNDE
Prog. QNDE 10B:1159-1166 (1991). 11A:233-240 (1992).
26. H. N. G. Wadley, A. H. Kahn, Y. Gefen, and M. Mester. Rev. 67. N. Nakagawa and J. C. Moulder. Rev. Prog. QNDE 12A:259-
Prog. QNDE 7B:1589-1598 (1988). 263 (1993).
27. H. M. Lewis, I. H. Michael, M. C. Lugg, and R. Collins. J. Appl 68. Y. Liu, N. Nakagawa, and F. Rizzo. Rev. Prog. QNDE 12A:235-
Phys. 64:3777-3783 (1988). 242 (1993).
28. D. H. Michael, A. M. Lewis, M. Mclver, and R. Collins. Proc. 69. J. R. Bowler. J. Appl. Phys. 75:8128-8137 (1994).
Roy. Soc. London A434:587-603 (1991). 70. C. Kittel. Introduction to Solid State Physics, Wiley, New York
29. R. E. Beissner, M. J. Sablik, and C. M. Teller. Rev. Prog. QNDE (1971), pp. 450-454.
2B:1237-1254 (1983). 71. J. A. Osborne. Phys. Rev. 67:351-357 (1945).
30. R. E. Beissner and M. J. Sablik. Rev. Prog. QNDE 3A:633- 72. S. M. Nair and J. H. Rose. J. Appl. Phys. 70:1924-1937 (1991).
641 (1984). 73. G. L. Hower. J. Nondestr. Eval. 6:177-180 (1987).
31. M. Mayos and J. L. Muller. J. Nondestr. Eval. 6:109-116 (1987). 74. G. L. Hower and D. E. Hadlock. J. Nondestr. Eval. 8:247-255
32. G. L. Burkhardt, E. A. Creek, and J. L. Fisher. Rev. Prog. QNDE (1989).
13A:343-356 (1994). 75. D. E. Hadlock and G. L. Hower. Res. Nondestr. Eval. 2:255-
33. B. Wincheski, J. Fulton, and S. Nath. Rev. Prog. QNDE13B:1939- 237 (1990).
1946 (1994). 76. A. H. Kahn, R. Spal, and A. Feldman. J. Appl. Phys. 48:4454-
34. W. Sheppard, D. Mih, and K. Tam. Rev. Prog. QNDE 1:395- 4459 (1977).
398 (1982). 77. D. T. Paris and F. K. Hurd. Basic Electromagnetic Theory,
35. B. A. Auld. Rev. Prog. QNDE 10A:951-955 (1991). McGraw-Hill, New York (1969), pp. 516-519.
36 Auld and Moulder

78. W. D. Dover, F. D. W. Charlsworth, K. A. Taylor, R. Collins, and 107. S. J. Norton, A. H. Kahn, and M. L. Mester. Res. Nondestr. Eval.
D. H. Michael. In Eddy Current Characterization of Metals and 1:167-179 (1989).
Structures, ASTM STP 722, G. Birnbaum and G. Free, eds. (1981), 108. J. H. Rose, E. Uzal, and J. C. Moulder. SPIE2160:164-175 (1994).
pp. 401-427. 109. B. A. Auld, J. Kenney, and T. J. Lookabaugh. Rev. Prog. QNDE
79. R. E. Beissner and M. J. Sablik. J. Appl. Phys. 56:448-454 (1984). 5A:681-690 (1986).
80. B. A. Auld, G. McFettridge, M. Riaziat, and S. Jefferies. Rev. 110. A. J. Bahr. Rev. Prog. QNDE 5A:691-698 (1986).
Prog. QNDE 4A:623-634 (1985). 111. A. Rosengreen, A. J. Bahr, and D. M. Marsland. Rev. Prog. QNDE
81. N. Nakagawa, J. Chao, and A. N. S. Prasad. In Nondestructive 7A:493-500 (1988).
Testing of Materials, R. Collins et al., eds., IOS Press, London 112. C. V. Dodd and W. E. Deeds. Rev. Prog. QNDE 1:387-394 (1982).
(1996), p. 203. 113. M. Riaziat and B. A. Auld. Rev. Prog. QNDE 2A: 189-204 (1983).
82. N. Nakagawa and J. Chao. Rev: Prog. QNDE 15A:339-345 114. A. J. Bahr and D. W. Cooley. Rev. Prog. QNDE 2A:225-244
(19%). (1983).
83. J. Chao, D. Lether, J. C. Moulder, and N. Nakagawa. Rev. Prog. 115. J. J. Brandstatter. An Introduction to Waves, Rays, and Radiation
QNDE 15A:355-360 (1996). in Plasma Media, McGraw-Hill, New York (1963).
84. B. A. Auld, F. Muennemann, and D. K. Winslow. J. Nondestr. 116. D. C. Copley. Rev. Prog. QNDE 2B:1527-1540 (1983).
Eval. 2:1-21 (1981). 117. P. M. Morse and H. Feshbach. Methods of Theoretical Physics,
85. B. A. Auld, F. G. Muennemann, and M. Riaziat. In Nondestructive Wiley, New York (1953), Part I, pp. 137-138.
Testing 7, R. S. Sharpe, ed., Academic Press, London, (1984), 118. J. H. Hippler, H. Emert, and L. von Bernus. J. Nondestr. Eval.
pp. 38-75. 12:153-162(1993).
86. S. K. Burke. J. Nondestr. Eval. 7:35-44 (1988). 119. J. M. Prince, B. P. Hildebrand, and G. L. Hower. J. Nondestr.
87. R. E. Beissner. J. Nondestr. Eval. 7:25-34 (1988). Eval. 12:209-217 (1993).
88. N. Harfield and J. R. Bowler. J. Appl. Phys. 76:4853-4856 (1994). 120. H. D. Collins, T. J. Davis, and L. J. Busse. Acoust. Imaging
89. G. L. Hower and R. W. Rupe. J. Nondestr. Eval. 4:59-63 (1984). 11:37-43 (1982).
90. D. H. Michael, R. Collins, D. R. Parramore, M. Aldoujailly, and 121. S. M. Nair and J. H. Rose. Inverse Problems 6:1007-1030 (1990).
P. R. Travis. Rev. Prog. QNDE 7A:191-197 (1988). 122. J. H. Rose and S. M. Nair. Inverse Problems 7:131-136 (1991).
91. B. A. Auld, S. R. Jefferies, and J. C. Moulder. J. Nondestr. Eval. 123. A. J. Bahr and B. A. Auld. J. Nondestr. Eval. 7:71-77 (1988).
7:79-84 (1988). 124. B. R. Groshong, G. L. Bilbro, and W. E. Synder. J. Nondestr.
92. T. Takagi, M. Hashimoto, T. Jugiura, S. Norimatsu, S. Arita, and Eval. 10:127-137 (1991).
K. Miya. Rev. Prog. QNDE 9A:327-334 (1990). 125. L. D. Sabbagh and H. A. Sabbagh. Rev. Prog. QNDE 28:155-
93. Z. Badics, H. Komatsu, H. Motosuji, K. Aoki, and F. Nakayasu. 157 (1983).
Int. J. Electromag. Mater. 4:357-362 (1994). 126. L. D. Sabbagh and H. A. Sabbagh. Rev. Prog. QNDE 4A:635-
94. Z. Badics, Y. Matsumoto, K. Aoki, F. Nakayasu, M. Uesaka, and 643 (1985).
K. Miya. J. Nondestr. Eval. 14:181-192 (1995). 127. L. Udpa and S. S. Udpa. J. Nondestr. Eval. 7:111-120 (1988).
95. J. R. Bowler. Electrosoft 2:142-156 (1991). 128. M. P. Connolly, D. H. Michael, and R. Collins. J. Appl. Phys.
96. R. E. Beissner. Electrosoft 2:122-141 (1991). 64:2638-2647 (1988).
97. D. Lether, J. Chao, N. Nakagawa, and J. C. Moulder. Rev. Prog. 129. S. J. Norton and J. R. Bowler. J Appl. Phys. 73:501-512 (1993).
QNDE 15A:361-368 (1996). 130. J. R. Bowler, S. J. Norton, and D. J. Harrison. J. Appl. Phys.
98. M. Riaziat and B. A. Auld. Rev. Prog. QNDE 3A:511-521 (1984). 75:8138-8144 (1994).
99. M. Riaziat. Analytic Methods in Electromagnetic Nondestructive 131. S. K. Burke. J. Appl. Phys. 76:3072-3080 (1994).
Evaluation, Ph.D. thesis, Department of Electrical Engineering, 132. B. A. Auld, S. Jefferies, J. C. Moulder, and J. C. Gerlitz. Rev.
Stanford University, Stanford, CA (1984). Prog. QNDE 5A:383-393 (1986).
100. N. Harfield and J. R. Bowler. Rev. Prog. QNDE 13A:279-286 133. J. C. Moulder, E. Uzal, and J. H. Rose. Rev. Sci. Instrum. 63:3455-
(1994). 3465 (1992).
101. W. Lord and R. Palanisamy. In Eddy Current Characterization 134. E. Uzal, J. C. Moulder, S. Mitra, and J. H. Rose. J. Appl. Phys.
of Materials and Structures, ASTM STP 722, G. Birnbaum and 74:2076-2089 (1993).
G. Free, eds. (1981), pp. 5-21. 135. A. Sethuraman and J. H. Rose. J. Nondestr. Eval. 14:39-46 (1995).
102. J. C. Treece, K. Murphy, and H. A. Sabbagh. Rev. Prog. QNDE 136. J. H. Rose, E. Uzal, and J. C. Moulder. SPIE2160:164-176 (1994).
13A:319-326 (1994). 137. D. J. Harrison, L. D. Jones, and S. K. Burke. J. Nondestr. Eval.
103. B. A. Auld and M. Riaziat. J. Appl. Phys. 54:3509-3517 (1983). 15:21-34 (1996).
104. R. Collins, D. Mirshekar-Syakahl, and D. H. Michael. Proc. Roy. 138. R. E. Beissner, M. J. Sablik, K. J. Krzywosz, and J. E. Doherty.
Soc. London A393:159-170 (1984). Rev. Prog. QNDE 2B:1159-1286 (1983).
105. D. McA. McKirdy. Rev. Prog. QNDE 12A:265-270 (1993). 139. R. E. Beissner and J. L. Fisher. Rev. Prog. QNDE 5A:189-197
106. N. Ida, H. Hoshikawa, and W. Lord. NDE Int. 18:331-338 (1985). (1985).