IEEE TRANSACTIONS ON MAGNETICS, VOL. 53, NO.

5, MAY 2017 6201210

Modeling of the Change of Impedance of an Eddy Current

Probe Due to Small Changes in Host Conductivity
Matthew R. Cherry1 , Shamachary Sathish2 , Ryan D. Mooers1 , Adam L. Pilchak1 , and Ramana Grandhi3
1 Materials and Manufacturing Directorate, Air Force Research Laboratory, Wright-Patterson Air Force Base, OH 45433 USA
2 Structural Integrity Division, University of Dayton Research Institute, Dayton, OH 45469 USA
3 Wright State University, Dayton, OH 45435 USA

Two different approximation techniques for predicting the response of an eddy current coil in the presence of small changes in
conductivity were developed. The small changes in conductivity are the result of changes in the orientation of individual anisotropic
crystals in a polycrystalline aggregate. Orientation information from electron backscatter diffraction was imported directly into
the modeling domain and the simulations were run to map orientations into an approximated eddy current response. These
approximated responses were compared with experimental data obtained with commercially available eddy current equipment, and
the approximations were found to be in good agreement with experiment. Further verification was performed with other existing
numerical and analytical models to demonstrate the accuracy of the approximations made in deriving the eddy current response.
This paper shows these results and demonstrates the viability of using low-fidelity approximations in predicting the eddy current
response when the change in conductivity is low.
Index Terms— Eddy current, microstructure, nondestructive testing.

I. I NTRODUCTION to high temperature exposure and changes in wear resistance

due to changes in chemistry of the material [9]–[11]. Recently,
E DDY current testing (ECT) is a nondestructive testing
method used in many different industries for detection
of damage in structural components made from conductive
ECT bulk electrical conductivity measurements were used to
estimate the volume fraction of different phases and β transus
materials. ECT probes are sensitive to the electrical and mag- temperature in titanium alloys [12]. A physics-based rule
netic properties of the host material, and hence, many authors of mixtures was used to relate the microstructure to the
have also shown the applicability of ECT in microstructure ECT response, which enabled quantification of β transus and
characterization. ECT is used extensively in characterization composition on more than an empirical basis.
of steels and steel alloys, because the various constituent Yin et al. [13] used effective medium theory and finite-
phases in steels have very different magnetic and electrical element method (FEM) to estimate the electromagnetic prop-
properties [1]–[6]. In addition, the work in characterization of erties and its relation to ferrite fraction and morphology in
microstructure in nickel-based superalloys has been performed steel. Both of these studies use bulk measurement techniques,
where the various precipitates that form in the material during but they show the value of relating the estimated material
manufacturing and thermal aging contribute differently to the parameters to the microstructure through physical arguments
conductivity [7], [8]. These characterization techniques are so more properties can be quantified using the eddy current
all empirical by nature and use large coils (1–10 mm) and signal.
low frequencies (10–1000 Hz) to obtain a homogenized bulk Eddy current methods have been used in the past to charac-
conductivity aggregated from small microstructure features terize spatial inhomogeneity of conductivity on an extremely
of interest (1–40 mm). These empirical studies provide the fine scale (50–100 nm) [14]. This method achieves excellent
estimates of volume fractions of electrically and magnetically contrast with changing conductivity at different spatial loca-
different phases that constitute an aggregate and do not provide tions below the sensor but requires a very precise experimen-
spatially dependent information. tal setup. No previous work has addressed characterization
Investigation of microstructure characterization with ECT in of spatial features of microstructure with ECT coils in the
titanium alloys has been very limited. Most of the studies have resolution range between 1 mm and 100 nm. In titanium, it has
focused on using ECT as a tool to measure the bulk conduc- been shown [15] that the spatial variations in microstructure
tivity of a material without considering local perturbations of can have dramatic effects on ECT probes due to the average
the conductivity due to small coil geometries. Bulk electrical sizes of the grains (10–20 μm) and ensembles of similarly
conductivity measured in titanium alloys with eddy current oriented grains (100–5000 μm) relative to the size of probes
instruments has been used to detect oxygen contamination due (0.05–1 mm). While this paper clearly showed contrast with
changing orientation of the grains, the hypothesis was never
Manuscript received May 5, 2016; revised December 12, 2016; accepted tested by comparison of the ECT images with data from
December 22, 2016. Date of publication January 9, 2017; date of cur- orientation imaging. Furthermore, no theoretical justification
rent version April 17, 2017. Corresponding author: M. Cherry (e-mail: was made to establish the limits of orientation imaging with
[Link].2@[Link]).
Color versions of one or more of the figures in this paper are available ECT based on percent change in conductivity. In this paper,
online at [Link] electron backscatter diffraction (EBSD) was used to obtain
Digital Object Identifier 10.1109/TMAG.2017.2650148 realistic microstructure to data to import into the eddy current

0018-9464 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See [Link] for more information.
6201210 IEEE TRANSACTIONS ON MAGNETICS, VOL. 53, NO. 5, MAY 2017

models. The conductivity maps provided by ECT coils from used [24], but again, this method relies on the calculation of
experiments and models were then directly compared. The Greens functions that might not be available for complicated
theoretical models show clear limits on imaging conductivity geometries. A more general way of solving this problem is
with ECT methods. These models will provide the basis for with an FEM scheme. FEM is computationally expensive, but
microstructure characterization with ECT coils in the currently the breadth of problems that it can handle makes it a useful
commercially available size range of 0.25–1 mm. method in its own right. FEM for eddy current problems
has been studied by various recent authors [25]–[27]. The
A. Modeling in ECT examples of other modeling methods that have been explored
Traditional ECT models calculate the impedance change of include distributed point source method [28] and Nyström
an idealized coil in the presence of local conductivity changes discretization [29].
in a host material so that the response of an eddy current coil These modeling methods are all discretization based numer-
due to damage can be predicted. In [16], a relationship is given ical algorithms. Since they are intended to solve for the
for the impedance change due to a local discrepancy in the response of a coil over a discrete flaw in the material, they rely
conductivity of a host material. The expression is derived using on having fine meshes in the area of the flaw. In the case of
the principles of conservation of energy of an electromagnetic microstructure simulations, the flaw is the size of the material
field given by Poynting’s theorem to arrive at a surface integral with which the coil interacts. This implies that for any of
equation the previously discussed methods to be successful, they would
 have to allow for a relatively fine mesh over the entirety of
1
Z = 2 (Eb × Ha − Ea × Hb ) · n d x (1) the sample. This makes microstructure simulations with these
I S methods computationally intensive. For the sake of microstruc-
Z is the change of impedance due to the presence of the ture characterization with inverse techniques, a lower cost
damage, I is the input current, E is an electric field intensity, alternative is needed. A quick model for prediction of ECT
H is a magnetic field intensity, the subscript b represents the signals above anisotropic polycrystalline materials does not
field due to an unflawed part, the subscript a is a field due currently exist.
to the flawed part, S is a surface encompassing the notch or To address this need, this paper will focus on development
flawed region, and n is an inward pointing unit vector normal of two low-fidelity, fast-solving forward models to predict
to the surface S. This equation can also be expressed as a the change in impedance of an eddy current probe in the
volume integral (as will be shown in Section IV-C). These presence of heterogeneous materials that have features slightly
equations give a solution for the change of impedance of a smaller than the size of the ECT probe. The two models are
coil due to an arbitrarily shaped flaw for any coil provided based on the Born approximation, which is considered valid
the flawed and unflawed fields are known at all points on the within a certain percent change in the material properties
surface of the flaw. of the scatterer, as well as discontinuity of the fields at the
The unflawed field can be calculated for a number of grain boundaries. The validity of the approximations will
different scenarios described by the geometry of the con- be shown based on analytical calculation of the impedance
ductor and the coil. For instance, Dodd and Deeds [17] change on both isotropic and anisotropic homogeneous mate-
provided analytical solutions for flat pancake coils over mul- rials. Orientation imaging microscopy data with μm step size
tilayered samples and cylindrical coils surrounding tubes. collected to determine crystallographic orientation (as Euler
These results were extended by many authors [18]–[21] to angles) of every point in the spatial grid on the surface of
include several more complex situations such as varying a beta annealed Ti64 sample will be used in the model to
probe and sample geometries, as well as deviations from predict the eddy current signal. These predictions will be
simple media. compared with experimental eddy current data collected with
The second aspect of calculating the impedance change a commercial ECT probe to validate the predictions in the case
with (1) is the calculation of the flawed field quantities. This is of heterogeneous, anisotropic materials.
an issue of great interest and constitutes significant motivation
for the development of eddy current forward models. One
II. E LECTRICAL C ONDUCTIVITY AND M ICROSTRUCTURE
method to accomplish this is to solve for a set of dipole
OF T I 64
point sources distributed in the region of the crack that
act as a barrier to current flow [22]. A popular method to A majority of titanium alloys used in industry are mul-
accomplish this is by the volume integral method (VIM) as tiphase systems in which both body centered cubic (BCC)
described in [23]. Essentially, the field due to the dipole and hexagonal close packed (hcp) phases exist in the same
source is calculated as a Greens function and the integral sample. The composition and the processing determine the
along the entire flaw region is calculated with the method of ratio of the volume fraction of the two phases. In Ti64, there
moments. This results in a full matrix calculation, but since can be upward of 93% hcp (α-)phase. Furthermore, the hcp
the flaw region is the only region that needs to be meshed, structure is anisotropic in conductivity, so spatial variations in
the matrix sizes are relatively small. The obvious attractive the orientation of this phase will show a large contrast in ECT
part of this method is that it is quick, but it is restricted images, while the BCC phase is electrically isotropic. For these
to the problems for which the analytical Greens functions reasons, the relatively small volume fraction of BCC (β)-phase
can be calculated. Boundary element methods have also been will be ignored in this analysis.
CHERRY et al.: MODELING OF THE CHANGE OF IMPEDANCE OF AN EDDY CURRENT PROBE 6201210

Fig. 2. Inverse pole figure map of the OIM data collected from the
perpendicular grain boundary sample. This image was generated with MTEX
Fig. 1. Subsequent rotations of the crystal from the global frame of reference Toolbox [31].
to the crystal frame of reference [30].

is then given as
In Fig. 1, the hexagonal prism represents the crystal struc- ⎡ ⎤
ture of α-phase Ti64. According to the standard Miller–Bravis σ11 σ12 σ13
indices for hcp crystals, the X-axis lies along the [21̄1̄0] σ̄¯ R = R σ̄¯ R T = ⎣σ12 σ22 σ23 ⎦ (4)
direction, the Y -axis lies along the [011̄0] direction, and the σ13 σ23 σ33
Z -axis lies along the [0001] direction. The plane spanned by where
the X and Y (the (0001) plane) is called the Basal plane, and  
the axis collinear with the Z -axis is called the c-axis of the σ11 = σXX 1 − s12 s22 + σZZ s12 s22
crystal. The conductivity of α-phase titanium is different along σ12 = 2s1 c1 s22 (σXX − σZZ )/4
the c-axis than it is along any direction within the Basal plane, σ13 = −s1 c2 s2 (σXX − σZZ )
but it is purely isotropic along any direction in the Basal plane.    
σ22 = σXX 1 − s22 + s12 s22 + σZZ s22 1 − s12
This implies that the conductivity tensor of a single crystal of
the material can be expressed as σ23 = c1 c2 s2 (σXX − σZZ )
 
⎡ ⎤ σ33 = σZZ 1 − s22 + σXX s22 . (5)
σXX 0 0
σ̄¯ = ⎣ 0 σXX 0 ⎦. (2) An important part of these equations is that there is no
0 0 σZZ c3 or s3 in any of them. This agrees with intuition in that
the conductivity is isotropic in the basal plane of the hcp unit
Conductivity in any direction other than the principal axes cell. This implies that the electromagnetic response that relies
can be described by an ellipsoid with semiaxes equal to the on the conductivity is invariant under rotations about the c-axis
diagonal terms shown in this tensor. For the purpose of this of the unit cell, which is well documented in the literature [15].
paper, the specific values of conductivity that will be used in The third angle ψ2 is simply one final rotation about the crystal
this tensor are σXX = 5.8e5 Sm−1 and σZZ = 5.4e5 Sm−1 . c-axis, and thus, the eddy current response is not sensitive to it.
These values are only assumed and are not the actual values For experimental validation of the techniques developed
of the material tested, which are not currently available. in this paper, the Euler angles at each point in a large
spatial grid on a β-annealed Ti64 sample were collected using
III. M ICROSTRUCTURE DATA EBSD methods. This sample was prepared to ensure that the
grain boundaries were perpendicular to the cut faces. This
In a pollycrystalline material, the hexagonal prism repre- assumption is made in the model derivations, and thus, its
senting the single crystal could be arbitrarily oriented for each effect should be determined on the final response. However, for
crystallite, and thus, the orientation needs to be defined as an validation purposes, it was desirable to have the experimental
input to the computational routines. The orientation is defined data free from this factor. The map of orientations for the
by a set of Euler angles. In this paper, the Z -X  -Z  convention sample is shown in Fig. 2. This map relates the Euler angles,
is used to define these angles, z meaning a rotation about the measured at each spatial scan point, to an orientation relative
first Z -axis or the c-axis of the crystal, X  a rotation about to basic crystallographic planes of the hexagonal unit cell. The
the new X  -axis, and Z  a rotation about the new Z  -axis or angles and the spatial information are stored in large ASCII
again the c-axis. These rotations are shown in Fig. 1. In this text files that serve as the inputs to the computational routines
paper, these angles are referred to as (ψ1 , θ , ψ2 ), respectively. that map the crystal structure to the predicted eddy current
These angles are used to define a rotation matrix to rotate the response.
original conductivity tensor, σ̄¯ to the global reference frame,
referred to with lower case letters in this paper as x yz. The IV. ECT A PPROXIMATIONS
rotation matrix is expressed as
Two different methods were developed to approximate
⎡ ⎤
c1 c3 − s1 c2 s3 −c1 s3 − s1 c2 c3 s1 s2 the eddy current signal above a heterogeneous, anisotropic
R = ⎣s1 c3 + c1 c2 s3 c1 c2 c3 − s1 s3 −c1 s2 ⎦. (3) material. Both rely on a Born-like approximation, as will
s2 s3 s2 c3 c2 be discussed in this section. They are not equivalent in
derivation, though they produce a similar response to one
Here, ci is the cosine of the i t h angle, i ∈ {1, 2, 3} correspond- another. A summary of each method will be provided at the
ing to the index of the respective Euler angles (ψ1 , θ , ψ2 ). beginning, followed by a detailed discussion. Convergence to
si is the sine of the i t h angle. The rotated conductivity tensor a solution will be shown, followed by a discussion on the
6201210 IEEE TRANSACTIONS ON MAGNETICS, VOL. 53, NO. 5, MAY 2017

the induced currents given the known electric field. rJ is not

necessarily parallel to rE in an anisotropic material. Taking
the magnitude of this expression gives

J = E riT σ̄¯ RT σ̄¯ R ri. (8)

From this equation, the directional conductivity at the i t h point

is defined as

σi = riT σ̄¯ RT σ̄¯ R ri. (9)
Fig. 3. Image of a ficticious coil above a random patch of grains. A vector
from the center of the coil (di ) is drawn to each point in the OIM data set ( pi )
and the unit normal to that vector (ri ) is used as re in computing directional
Averaging each of the N conductivities below the coil gives
conductivity [30]. the reduced scalar conductivity as

N
convergence to the correct solution. The conductivity ranges
σr = σi. (10)
in which they are valid will be discussed as well. N
i=1

A. Reduced Conductivity Approximation At this point, any numerical algorithm that solves the one
This method was originally shown in [30], but the details are layer Dodd and Deeds problem in [17] gives an estimate
restated here for ease of reference and consistence of notation. of the impedance of the coil above this particular patch of
In this approximation, the spatial grid of points is filtered to grains. These computations are performed with σr at each coil
those just below a fictitious coil in a particular scan position. position to give the approximate ECT response.
A directional conductivity is defined for each point based on One of the key assumptions made in the process of deriving
the position relative to the center of the coil. In Fig. 3, a coil this expression for reduced conductivity was that the electric
can be seen above a random patch of grains from which EBSD field was known. If Maxwell’s equations are solved with a
data has been collected in a regular grid. The entire grid of coil above an isotropic half-space, the field within the half-
data has been filtered down to this subset that corresponds to space has an analytical expression given by Dodd and Deeds.
this particular coil position. At a point in the grid ( pi ), a line is A change in the conductivity is introduced as a subspace of
drawn from the center of the coil to the point (di ), and then, the half-space, but in the case of heterogeneous microstructure
a tangential unit vector to that line is drawn (ri ). This unit features, the change is small relative to the typical case in
vector is the direction in which the directional conductivity for which a void is introduced in the sample. Though this is a low
this point will be calculated. These directional conductivities frequency solution to Maxwell’s equations in the near field,
are calculated for each point below the coil and then averaged the disturbances to the secondary field can be considered a
to get a total reduced conductivity for the coil at that scan scattered field. As the change in the material properties of the
position. scatterer is low, the transmitted field inside the scatterer can
The directional conductivity is calculated with an applica- be loosely approximated as the impending field, in a similar
tion of the constitutive relation for induced current density manner as an application of the Born approximation in typical
given by scattering theory [32]. This provides a potential limit to the
applicability of this algorithm based on the percent change
J = σ̄¯ R · E. (6) in the conductivity. If the change is too high, this algorithm
should break down at some point producing inaccurate results.
In this equation, J is the electric current density, E is the elec-
tric field, and the rotated conductivity tensor was introduced
B. Approximate Impedance Integrals
in Section II. This is typically substituted into Maxwell’s
equations, for instance to derive the wave equation in a The second approximation was originally shown in [33]
conductive media. However, if it is assumed that the electric and gives similar results to the reduced conductivity approx-
field is known, the induced current density can then be derived imation (RCA) method, but has a much stronger basis in
from a simple application of (6). fundamentals of electromagnetic wave propagation. Maxwell’s
The directional conductivity can then be determined accord- equations, in the absence of applied fields in a linear, homo-
ing to the following procedure. The unit vector, rE , pointed in geneous media that is anisotropic only in conductivity can be
the direction of the known electric field is determined. In the written as
case of the eddy current coil, the electric field can be assumed ∇ × H = j ωεE + σ̄¯ R · E
to be directed along the tangential direction to the coil itself
∇ × E = − j ωμH. (11)
everywhere below the coil. This implies that rE coincides
with ri , and the constitutive relation at a point below the coil In these expressions, ω is the angular frequency, μ is the
can be written as permeability, and ε is the permittivity. The assumptions about
the media in which these fields exist are that it is linear,
J = J rJ = E σ̄¯ R ri (7)
homogeneous, and anisotropic in conductivity, isotropic in all
where J is the magnitude of the induced current density, E is other properties. An expression relating the Poynting vector
the magnitude of the electric field, and rJ is the direction of to the power density in the electric, magnetic, and dissipated
CHERRY et al.: MODELING OF THE CHANGE OF IMPEDANCE OF AN EDDY CURRENT PROBE 6201210

components of the field due to induced currents derived TABLE I

from (11) can then be written as [32] D IMENSIONS OF THE C OIL U SED FOR F INDING THE A CCURACY
 OF THE A NISOTROPIC C ALCULATIONS
(E × H) · n d S
S 
  
= j ωμH · H + j ωεE · E + E · σ̄¯ R · E d V. (12)
V
In [16], this expression is related to the change of impedance
of an eddy current coil to give

1
Z = 2 (Eb × Ha − Ea × Hb ) · n d S
I S

1  However, difficulty could arise at even modest frequency
= 2 j ωμ(Ha ·Ha −Hb ·Hb )+ j ωε(Ea ·Ea −Eb ·Eb ) ranges in the 100 kHz range. This effect should be considered
I V
  carefully when using the approximations in this paper.
+ Ea · (σa I3×3 · Ea ) − Eb · σ̄¯ b · Eb d V. (13)
The first part of this expression is the same as (1), σa is V. V ERIFICATION AND VALIDATION
a reference conductivity of the isotropic half-space, σ̄¯ b is Convergence and verification results for the approximate
the anisotropic conductivity tensor that has been rotated by impedance integral (AII) method have been shown previously
the Euler angles in V , I3×3 is a 3×3 identity tensor, and in [33]. These results indicated that in an isotropic test case,
permeability and permittivity remain constant from flawed to the approximation remained valid for changes of up to 50% in
unflawed configuration. conductivity. To perform further verification, a similar analysis
Since the conductivity changes should be low in a rotated was performed, but with an analytically derived anisotropic
grain, the Born approximation can be used to derive an half-space model. Furthermore, both methods were compared
expression for the impedance change given by against multiple numerical simulations of an anisotropic patch
 in an isotropic half-space. Finally, the models were both run
1   
Z = 2 Ea · σa I3×3 − σ̄¯ b · Ea d V. (14) for the map of orientations collected from the sample described
I V in Section III and were found to be in good agreement with
This equation can then be used to calculate the impedance experimental eddy current results. This section documents
response of an eddy current coil above a heterogeneous these results.
material. The quantities inside the integrand are all known,
which implies that this calculation is simply a numerical A. Verification With Anisotropic Analytical Model
integration. Since no system of equations need be solved, this The model from the study in [21] was used to determine
method can be considered mesh free, and the solution time accuracy of the approximations over multiple frequency ranges
is strictly limited to the amount of sample points needed for for calculating the impedance of a coil over an anisotropic
convergence of the integral. This was discussed further in [33]. sample. The coil in this verification study has dimensions
Furthermore, since each scan point is essentially a summation shown in Table I. To show the accuracy of the approximations,
and is independent of all other scan points, the simulation is the models were used to calculate the total impedance change
highly parallel in nature. from that of an isotropic half-space. The impedance in an
The reference electric field Ea can be calculated using either isotropic reference state was calculated using the analytical
the analytical solution or by using a numerical algorithm. expressions of Dodd and Deeds. The AII approximation was
In this paper, a linear FEM simulation in COMSOL Multi- used to calculate the changes from this reference isotopric
physics was used for the calculation of the reference electric state to the anisotropic state for varying degrees of anisotropy.
field values at the surface of the sample when the refer- Last, the actual impedance in the anisotropic configuration
ence material is an isotropic, homogeneous half-space. The was calculated using the analytical expressions of Burke.
second-order triangular elements with Lagrange interpolation Comparing the values from the approximation to the values
functions were used with an element size of δ/3, resulting from the analytical model gives a sense of how accurate the
in adequate convergence for this problem. The simulation approximate model is as a function of increasing degree of
requires ∼4 s to perform on a laptop PC, making it a relatively anisotropy.
small contribution on the overall simulation time. According to the coordinates used in [21], the conductivity
in the x-direction was varied from 10% of the value
C. Frequency Dependence of the Approximations used in the paper to 100%, or from 1.4706 × 106 Sm−1
The models previously discussed are only accurate above a to 1.4706 × 107 Sm−1 . The conductivity in the y-direction
certain low-frequency limit. The accuracy degrades relative was set to 1.4706 × 107 Sm−1 . This implies that
to the nonapproximation-based models as the frequency is when the conductivity in the x-direction is at 100%
decreased due to lowering fidelity in the approximations (1.4706 × 107 Sm-1 ), the sample is an isotropic half-space.
made during the derivation. In this paper, the frequency The reference electric field over the isotropic half-space was
of 2 MHz was seen to be accurate enough for application calculated at a conductivity of 1.4706 × 107 Sm−1 using
to microstructure characterization with eddy current coils. the 2-D FEM model discussed previously.
6201210 IEEE TRANSACTIONS ON MAGNETICS, VOL. 53, NO. 5, MAY 2017

Fig. 4. (a) Resistance change and (b) reactance change as a function of changing conductivity in one principal axis of an anisotropic half-space. In these
results, the conductivity of the x-axis is changing from 10% of the paper value up to 100% of the paper value.

Fig. 5. (a) Resistance change and (b) reactance change as a function of changing conductivity in one principal axis of an anisotropic half-space at a lower
frequency. The resistance plot clearly shows the degradation of the fidelity of the approximations as the frequency is lowered.

The real and imaginary components of impedance were dependent on frequency. Even though the reactance seems to
calculated using both the Burke model and the AII, and the have improved in accuracy, the expression for resistance is
change in resistance and reactance from the isotropic half- nearly unusable for any range of anisotropy.
space to the anisotropic half-space is plotted. The results
are shown in Fig. 4. These plots show the change from the B. Verification With Other Numerical Models
isotropic impedance calculated analytically to the anisotroipic Several computational methods such as FEM and VIM were
impedance calculated with both models as the conductivity discussed in Section I-A for solving ECT problems numeri-
in the x-direction is decreased (i.e., the anisotropy of the cally for small flaws. It was stated that these methods would
sample is increased). At 100%, both models give zero change require meshes of significant size to solve microstructure
as expected, since this state is isotropic. As the conductivity in problems. This is true for realistic microstructures, but in a
the x-direction is decreased, the AII approximation maintains simplified case, such as that shown in Fig. 6, FEM and VIM
reasonable accuracy relative to the analytical model in both can calculate solutions with reasonable mesh discretization
resistance and reactance until about 70%. This is a positive levels. This problem was solved with the parameters shown
result as the change in conductivity from one direction to in Table II with FEM, VIM, the AII approximation, and the
another in Ti64 crystals is typically on the order of 6%. RCA approximation. Brief introductions to the FEM and VIM
This analysis was also performed at 7 KHz to judge the models are given in Sections V-B.1 and V-B.2, respectively,
accuracy of the model as frequency decreases. The results and the results from each model are shown.
for resistance and reactance change with respect to level of 1) FEM Model Setup: The full 3-D FEM model was set
anisotropy are shown in Fig. 5(a) and (b), respectively. These up according to the procedure in [34]. The theory of FEM
results clearly demonstrate that the accuracy of the model is simulation of eddy current problems has been extensively
CHERRY et al.: MODELING OF THE CHANGE OF IMPEDANCE OF AN EDDY CURRENT PROBE 6201210

Fig. 7. 3-D mesh for the FEM test problem. The isotropic domain is shown
in green, the anisotropic patch in blue, and the coil in gray. The air domain
surrounding the coil is hidden.
Fig. 6. Problem setup for verification with numerical codes. The red patch
is a square of anisotropic material and the gray area is isotropic. The coil
beginning and end positions are shown. change given in (1) where the unflawed part is an FEM simu-
TABLE II lation with the patch domain set to the reference conductivity
PARAMETERS OF M ODEL FOR V ERIFICATION A GAINST and the flawed part is a simulation with the patch domain
N UMERICAL C ODES set to the anisotropic conductivity tensor. Two simulations are
performed at each coil position and the results are used to
calculate the change in impedance. Both simulations are run at
each position to reduce mesh noise from remeshing at each coil
position. This results in a simulation that takes well over 1 h
to solve on a workstation with dual hex-core processors, each
clocked at a nominal 2.7 GHz, and 48 GB of RAM.
2) VIM Model Setup: The theory of the VIM is discussed
in several references including [22] and [23]. In this paper,
a software program called VIC-3-D was used to solve the
model shown in Fig. 6. The first layer was set to the properties
of air, and the second layer was set to the isotropic reference
conductivity. The anisotropic patch was meshed with a regular
grid that was discretized with 16 × 16 × 4 elements. Solving
for air core coils with the VIM requires only the flaw domain
to be meshed. Furthermore, since the computer on which the
reviewed in [35]. A full 3-D model was used in this paper due simulations were run had sufficient memory to solve the sys-
to the lack of axial symmetry induced by the anisotropic patch. tem of equations with direct LU decomposition, the simulation
The mesh consisted of irregular tetrahedral elements with the ran in ∼4 s. This is efficient for the simplified problem, but
second-order Lagrange interpolation functions, and is shown in reality, the grid would likely have to be much finer if true
in Fig. 7. The mesh on the top surface of the anisotropic patch microstructure detail were included in the simulation, such as
consisted of triangles whose individual dimensions were no that shown in Fig. 2. Still, VIC-3-D was very efficient for the
more than half the skin depth of the material calculated with large-scale microstructure problem, especially when compared
conductivity σXX . This ensured adequate refinement in depth, with the traditional finite-element approach.
such that two elements per skin depth were achieved. The 3) RCA and AII Model Setup: A microstructure file was
mesh was relatively coarse around the coil, air, and isotropic prepared for input to the RCA and AII methods for this
material domains. A simple mesh convergence study was problem. The Euler angles were set to zero in the isotropic
performed to verify accuracy of the computations. Refining the domain, and the second Euler angle θ was set to 90 in the
mesh further produced changes on the order of 1/20th of anisotropic domain. The reference conductivity used for the
the total signal level, which was deemed accurate enough for AII method was set to half-way between the conductivities
the purposes of verification of the approximations. along the two principal axes. An element size of 200 μm was
The FEM model solves for the magnetic vector potential A used which resulted in a total simulation time of 8 s. This
and the electric scalar potential v at every point in space. is substantially higher than the VIM, but the refinement of
These quantities can be used to calculate the physical field the grid used in this paper is sufficient for any microstructure
values E and B using the relations detail needed in ECT scans. Furthermore, this method runs
in MATLAB and optimizations of the computations, such as
∂A
E=− − ∇v parallelization and precompilation, have not been performed.
∂t
B = ∇ × A. (15) 4) Discussion of Results: The results of this scan for all
of the simulations are shown in Fig. 8. The data clearly
The induced current density is calculated by substituting the shows that the results for all methods are somewhat close.
resulting A and v from the FEM model into (6) and (15). The FEM results are slightly shifted from the VIM results
These various results can be used to give the impedance over the patch, which is likely due to poor mesh refinement in
6201210 IEEE TRANSACTIONS ON MAGNETICS, VOL. 53, NO. 5, MAY 2017

Fig. 9. Experimental data collected from the eddy current scope [33].

Fig. 10. Numerical approximation of eddy current scans using the RCA
Fig. 8. Comparison of results from different numerical methods. method.

the FEM model. The AII method performs the poorest of the
low-fidelity models when attempting to predict the impedance
directly over the patch. However, it seems to perform better
as the conductivity change is lowered, moving closer to the
edge of the sample. The data are only for one test problem
and the accuracy of both methods and the differences between
them should be more thoroughly analyzed in the future work. Fig. 11. Numerical approximation of eddy current scans using the AII
The plot indicates both methods predict the response relatively method [33].
well when compared with the widely accepted numerical TABLE III
solutions to the problem. This builds confidence in their D IMENSIONS OF THE C OIL U SED IN THE VALIDATION S TUDIES W ITH
use. Furthermore, the simulation times for the methods are P ERPENDICULAR G RAIN B OUNDARIES
significantly lower than the FEM method, and are comparable
with the VIM solution. The primary disadvantage of the
approximations relative to the discretization-based simulations
is that the models are not useful at lower frequencies where
the assumptions break down.

C. Validation With Experimental Data

The sample shown in Fig. 2 was scanned with an eddy cur- compare with the model results. The frequency of operation
rent probe and these data were compared with the predictions of the probe was shown to be in an accurate range for the
from the computational routines to validate the simulations. models in the study in [33]. The experimental data are shown
An absolute, 2 MHz center frequency Uniwest pencil probe in Fig. 9. The approximate dimensions of the coil used in the
(model# P-.187 2 MHz) was used. The data were collected fem simulation for each of the approximation techniques are
with a Nortec 19eII eddy current scope [33]. The advantage shown in Table III.
of collecting data with an eddy current scope is that the noise Both approximations were run using this information and
levels are generally lower than in an impedance analyzer, the OIM data, and the results are shown in Figs. 10 and 11.
allowing the grain structure to be seen in the data. In fact, The scales of both approximation techniques as well as the
in this experiment, the noise in the large single crystal sections experimental data have been normalized for direct comparison.
had standard deviation roughly on the order of 1/100th of the Several things can be noted from these results. First, both
max signal change from one grain to another. The drawback model approximations seem to be able to predict the experi-
to using the eddy current scope is that the measurements mental results fairly accurately, qualitatively. The AII approx-
only produce a voltage which coincides with a change in imation appears to have predicted the experiments slightly
impedance from one area to another. There was no calibration better, as the contrast in Fig. 11 is closer to the experimental
block with known parameters to directly compare the signals data (this is very apparent in region 4 of the images).
taken from the experiments with the model results due to The models are somewhat noisy due to the coarseness of
the calibration blocks saturating the instrument. Because of the OIM data. The integrations and averages have not quite
this, the experiments could not be used for direct quantitative converged, resulting in numerical noise. This is especially
comparison of impedance values from the numerical methods. true around the grain boundaries themselves. Still, there are
Still, the experimental data give a representation of the changes aspects of the solutions that agree with intuition around
relative to changing orientation from one grain to another, the boundaries. For instance, Fig. 12 shows the tilt of the
and the contrast in the normalized images can be used to c-axis of each crystal relative to the z-axis (i.e., the second
CHERRY et al.: MODELING OF THE CHANGE OF IMPEDANCE OF AN EDDY CURRENT PROBE 6201210

R EFERENCES
[1] S. Ghanei, M. Kashefi, and M. Mazinani, “Eddy current nondestructive
evaluation of dual phase steel,” Mater. Design, vol. 50, pp. 491–496,
Sep. 2013.
[2] M. Kashefi, S. Kahrobaee, and M. H. Nateq, “On the relationship
of magnetic response to microstructure in cast iron and steel parts,”
J. Mater. Eng. Perform., vol. 21, no. 7, pp. 1520–1525, 2012.
Fig. 12. Tilt of the c-axis of the perpendicular grain boundary sample [33]. [3] S. Konoplyuk, “Estimation of pearlite fraction in ductile cast irons
by eddy current method,” NDT E Int., vol. 43, no. 4, pp. 360–364,
Euler angle, θ ). This is the primary source of sensitivity in 2010.
[4] S. Khan, A. N. Khan, F. Ali, M. Iqbal, and H. Shukaib, “Study
ECT measurements due to isotropy of the conductivity in the of precipitation behavior at moderate temperatures in 350 maraging
basal plane. In moving from regions 1 to 2 in this image, there steel by eddy current method,” J. Alloys Comp., vol. 474, no. 1,
is no discernable difference in the tilt of the c-axis between the pp. 254–256, 2009.
[5] K. Liu, Z. Zhao, and Z. Zhang, “Characterization of early fatigue
two regions. However, the IPF map in Fig. 2 shows that these microstructure in AISI 321 steel using eddy current non-destructive
are two distinct grains. Since their c-axis tilt is very similar, methodology,” J. Wuhan Univ. Technol-Mater. Sci. Ed., vol. 28, no. 6,
the regions appear very close in color scale in the experimental pp. 1201–1206, 2013.
[6] S. Kahrobaee and M. Kashefi, “Microstructural characterization of
data as well as the model results. The slight perturbation along quenched AISI D2 tool steel using magnetic/electromagnetic nonde-
the boundary is due to the fact that the first Euler angle is not structive techniques,” IEEE Trans. Magn., vol. 51, no. 9, pp. 1–7,
identical in the two grains. This essentially tilts the in-plane Sep. 2015.
[7] D. Pereira, T. Clarke, R. Menezes, and T. Hirsch, “Effect of microstruc-
conductivity ellipses relative to one another, causing a change ture on electrical conductivity of Inconel 718 alloys,” Mater. Sci.
in the conductivity in the direction normal to the boundary. Technol., vol. 31, no. 6, pp. 669–676, 2015.
However, since the coil is circular, the scale of the response [8] T. Parthasarathy, S. Boone, S. Rao, P. Wang, P. Nagy, and M. Blodgett,
“Investigation of microstructural effects on electrical resistivity and their
in each grain is identical. An oblique coil would be able to implications for eddy current methods in measuring residual stresses in
see the two grains distinctly due essentially to polarization IN718,” in Electromagnetic Nondestructive Evaluation. Amsterdam, The
of the incident field. The other area of note is region 3. The Netherland: IOS Press BV, vol. 13. 2010, pp. 165–172.
[9] A. Rosen, S. Nadiv, and M. Bohrer, “Effect of high temperature exposure
OIM data are slightly noisy in this region, which result in on mechanical and electrical properties of Ti-6Al-4V sheet,” Mater.
noise in the simulations that is not present in the experiments. Eval., vol. 33, no. 6, 1975.
Otherwise, the simulations are in satisfactory agreement with [10] D. J. Hagemaier, “Nondestructive detection of hydrides and alpha-case
in titanium alloys,” in Titanium Science and Technology. New York, NY,
the experimental data. USA: Springer, 1973, pp. 755–765.
[11] A. Makarov, E. Gorkunov, and L. K. Kogan, “Application of the eddy-
VI. C ONCLUSION current method for estimating the wear resistance of hydrogen-alloyed
In this paper, two different numerical approximations were β-titanium alloy BT35,” Russian J. Nondestruct. Test., vol. 43, no. 1,
developed to predict the results of an eddy current scan pp. 21–26, 2007.
[12] B. Sasi, B. Rao, T. Jayakumar, and B. Raj, “Characterisation of solution
above a heterogeneous material with small spatial variations in annealed VT-14 titanium alloy using eddy current based electrical
conductivity. The methods were tested against other models of resistivity measurements,” Trans. Indian Inst. Metals, vol. 63, no. 5,
changing homogeneous material properties as well as numer- pp. 773–777, 2010.
[13] W. Yin, A. J. Peyton, M. Strangwood, and C. L. Davis, “Explor-
ical models with patches of anisotropic conductivity, and then ing the relationship between ferrite fraction and morphology and the
against experimental data taken from an eddy current scan of electromagnetic properties of steel,” J. Mater. Sci., vol. 42, no. 16,
a polycrystalline titanium alloy. The results from verification pp. 6854–6861, 2007.
[14] V. Nalladega, S. Sathish, K. Jata, and M. Blodgett, “Development
indicate that assumptions made during the derivation of the of eddy current microscopy for high resolution electrical conductivity
approximation methods are valid within the small conductivity imaging using atomic force microscopy,” Rev. Sci. Instrum., vol. 79,
change seen in anisotropic heterogeneous materials, provided no. 7, p. 073705, 2008.
[15] M. Blodgett and P. B. Nagy, “Anisotropic grain noise in eddy current
the frequency of operation of the probe is high enough to be in inspection of noncubic polycrystalline metals,” Appl. Phys. Lett., vol. 72,
the accurate range. Furthermore, the results from experimental no. 9, pp. 1045–1047, 1998.
validation indicate that these models are able to predict the [16] B. Auld and J. Moulder, “Review of advances in quantitative eddy
current nondestructive evaluation,” J. Nondestruct. Eval., vol. 18, no. 1,
eddy current response above heterogeneous materials relatively pp. 3–36, 1999.
well. The two methods were shown to not give identical [17] C. V. Dodd and W. E. Deeds, “Analytical solutions to eddy-current
results, though the difference between the two is minimal. probe-coil problems,” J. Appl. Phys., vol. 39, no. 6, pp. 2829–2838,
1968.
However, the AII approximation has a much stronger basis [18] T. P. Theodoulidis and E. E. Kriezis, “Impedance evaluation of rectangu-
in the fundamentals of electromagnetic wave propagation and lar coils for eddy current testing of planar media,” NDT E Int., vol. 35,
was also slightly more accurate in experimental validation. no. 6, pp. 407–414, 2002.
[19] E. Uzal, J. C. Moulder, S. Mitra, and J. H. Rose, “Impedance of
This indicates that these results are likely to be more accurate coils over layered metals with continuously variable conductivity and
in the context of uncertainty quantification and characteriza- permeability: Theory and experiment,” J. Appl. Phys., vol. 74, no. 3,
tion of microstructure with ECT methods. pp. 2076–2089, 1993.
[20] R. Ditchburn and S. Burke, “Planar rectangular spiral coils in eddy-
ACKNOWLEDGMENT current non-destructive inspection,” NDT E Int., vol. 38, no. 8,
pp. 690–700, 2005.
The work of S. Sathish was supported by the Materials [21] S. Burke, “Eddy-current induction in a uniaxially anisotropic plate,”
& Manufacturing Directorate, Air Force Research Laboratory J. Appl. Phys., vol. 68, no. 7, pp. 3080–3090, 1990.
[22] J. Bowler, S. Jenkins, L. Sabbagh, and H. Sabbagh, “Eddy-current probe
under Contract FA8650-14-D-5224. The authors would like to impedance due to a volumetric flaw,” J. Appl. Phys., vol. 70, no. 3,
thank S. L. Semiatin for providing samples. pp. 1107–1114, 1991.
 6201210 IEEE TRANSACTIONS ON MAGNETICS, VOL. 53, NO. 5, MAY 2017

[23] J. R. Bowler et al., “A theoretical and computational model of eddy- Ryan D. Mooers is currently an Associate Materials Research Engineer 28
current probes incorporating volume integral and conjugate gradient with the Materials State Awareness Branch, Structural Materials Division, 29
methods,” IEEE Trans. Magn., vol. 25, no. 3, pp. 2650–2664, Mar. 1989. Materials and Manufacturing Directorate, Air Force Research Laboratory, Air 30
[24] T. Theodoulidis, H. Wang, and G. Y. Tian, “Extension of a model for Force Materiel Command, Wright Patterson Air Force Base, OH, USA, in the 31
eddy current inspection of cracks to pulsed excitations,” NDT E Int., Area of Nondestructive Evaluation. He also assists on various programs that 32
vol. 47, pp. 144–149, Apr. 2012. are developing next generation field and depot level eddy current inspection 33
[25] A. Rosell and G. Persson, “Finite element modelling of closed cracks equipment. He has personally assisted in the beta testing of the next generation 34
in eddy current testing,” Int. J. Fatigue, vol. 41, pp. 30–38, Aug. 2012. advance bolt hole scanning system at each of the three air logistics centers as 35
[26] Z. Zeng, T. Wang, L. Sun, R. He, and J. Chen, “A domain decomposition well as in the research labs. He recently undertook the position of technical 36
finite-element method for eddy-current testing simulation,” IEEE Trans. point of contact on an Air Force SBIR topic concerning subsurface corrosion 37
Magn., vol. 51, no. 10, pp. 1–9, Oct. 2015. detection, and will be leading a Program looking at determining the probability 38
[27] Z. Zeng, L. Udpa, S. S. Udpa, and M. S. C. Chan, “Reduced magnetic of detection for a newly transitioned bolt hole eddy current inspection system. 39
vector potential formulation in the finite element analysis of eddy current His current research interests include the area of eddy current forward model 40
nondestructive testing,” IEEE Trans. Magn., vol. 45, no. 3, pp. 964–967, verification and validation, and eddy current experimentation. 41
Mar. 2009.
[28] D. Placko and T. Kundu, DPSM for Modeling Engineering Problems.
Hoboken, NJ, USA: Wiley, 2007.
[29] J. C. Young, S. D. Gedney, and R. J. Adams, “Eddy current analysis
using a Nyström-discretization of the volume integral equation,” IEEE
Trans. Magn., vol. 49, no. 12, pp. 5675–5681, Dec. 2013.
[30] M. R. Cherry, S. Sathish, A. L. Pilchak, A. J. Cherry, and M. P. Blodgett,
“Characterization of microstructure with low frequency electromagnetic
techniques,” in Proc. 40th Annu. Rev. Prog. Quant. Nondestruct. Eval.,
Incorp. 10th Int. Conf. Barkhausen Noise Micromagn. Test., 2014, Adam L. Pilchak is currently a Senior Research Materials Engi- 42

vol. 1581. no. 1, pp. 1456–1462. neer and a Research Lead for the Metallic Materials and Processes 43

[31] G. Nolze and R. Hielscher, “Orientations-perfectly colored,” J. Appl. Team with the Metals Branch, Structural Materials Division, Materi- 44

Crystallogr., vol. 49, no. 5, pp. 1786–1802, 2016. als and Manufacturing Directorate, Air Force Research Laboratory, Air 45

[32] C. A. Balanis, Advanced Engineering Electromagnetics, vol. 111. Force Materiel Command, Wright-Patterson Air Force Base, OH, USA. 46

Hoboken, NJ, USA: Wiley, 2012. He has 28 first author articles in peer reviewed journals focusing on 47

[33] M. Cherry, S. Sathish, R. Mooers, and A. Pilchak, “Progress in processing-microstructure-property relationships in titanium alloys and is co- 48

model development for eddy current response in the presence of author of another 18 papers. He has given over 30 presentations at national and 49

small conductivity changes,” in Proc. 42nd Annu. Rev. Prog. Quant. international conferences. He is internationally recognized for his work on the 50

Nondestruct. Eval., Incorp. 6th Eur.-Amer. Workshop Rel. NDE, dwell fatigue response of titanium alloys, where he is making contributions 51

vol. 1706. 2016, p. 090019. on all aspects of the problem, including developing new processing routes 52

[34] M. Cherry, R. Mooers, J. Knopp, J. Aldrin, H. Sabbagh, and to mitigate the problem in future systems, microstructure quantification using 53

T. Boehnlein, “Low frequency eddy current finite element model vali- destructive and nondestructive inspection protocols, failure analysis, fleet risk 54

dation and benchmark studies,” Rev. Prog. Quant. Nondestruct. Eval., mitigation, and probabilistic fatigue lifetime prediction. He also leads a Team 55

vol. 1335, no. 1, pp. 357–364, 2011. of 25 government engineers and contractors, who maintain technical depth 56

[35] R. Albanese and G. Rubinacci, “Finite element methods for the solution in key critical Air Force technologies to respond to current and future needs, 57

of 3D eddy current problems,” Adv. Imag. Electron Phys., vol. 102, such as single crystal blade casting and additive manufacturing. 58

pp. 2–86, Dec. 1997.

1 Matthew R. Cherry has performed research in new sensors and methods

2 for over seven years. He is currently a Materials Research Engineer with the
3 Materials State Awareness Branch, Air Force Research Laboratory, WPAFB,
4 OH, USA, in the area of modeling and simulation of nondestructive evaluation
5 (NDE) methods. He has authored papers and given presentations on forward
6 uncertainty propagation through forward models, the development of forward
7 models, and the motivation for developing probabilistic models. This includes Ramana Grandhi is currently the Executive Director of International Col- 59
8 an invited lecture on reliability method in NDE. He developed different laborations and Graduate Programs and a Distinguished Research Professor 60
9 physics-based simulations currently in active use and has shaped and defended with the College of Engineering and Computer Science, Wright State Uni- 61
10 basic research Funded by AFOSR and other internal Funding sources. versity, Dayton, OH, USA. He also serves as the Director of Engineering 62
Ph.D. Program. In his 30-year academic career, he has supervised 25 Ph.D. 63
dissertations, 50 M.S. theses, and 30 Post-Doctoral Fellows. He is the author of 64
11 Shamachary Sathish was with the Nondestructive Evaluation Science over180 journal papers, 210 conference articles, and one textbook. His current 65
12 Branch, NASA Langley Research Center, Hampton, VA, USA, with the research interests include multidisciplinary analysis and optimization, aircraft 66
13 Department of Physical Metallurgy, École Polytechnique Fédérale de Lau- and automotive structures, risk-based design, and advanced manufacturing 67
14 sanne, Lausanne, Switzerland, and with the Department of Physics, Norwe- processes. His extensive research has included partnerships with the U.S. Air 68
15 gian Institute of Technology, Trondheim, Norway, where he was involved Force, the U.S. Navy, NASA, NSF, DARPA, GE, Boeing, Lockheed, Pratt & 69
16 in the extensive research on Scanning Acoustic Microscopy and micro- Whitney, Caterpillar, Ford, GM, and small businesses. 70
17 nondestructive evaluation (NDE) characterization of materials. He has over Dr. Grandhi has been honored with numerous awards, including the 71
18 25 years of research experience in nondestructive materials characterization American Society of Mechanical Engineers Fellow Award, the American 72
19 using several NDE techniques at multiple length scales in a wide variety Institute of Aeronautics and Astronautics (AIAA) Fellow Award, the AIAA 73
20 of materials. He is currently a Distinguished Research Engineer with the Multidisciplinary Design Optimization Award, the Solberg Award from the 74
21 University of Dayton Research Institute, Dayton, OH, USA, and holds a joint American Society of Naval Engineers, the Telugu Association of North 75
22 appointment as a Professor with the Department of Materials Engineering, America Award for Excellence in Engineering, the Outstanding Engineers 76
23 University of Dayton. He is also the Principal Investigator of an on-site NDE and Scientists Award from the Engineering and Science Foundation of Dayton, 77
24 research contract with the Air Force Research Laboratory, Wright-Patterson the Brage Golding Distinguished Professor Award, the Up and Comers Award 78
25 Air Force Base, Dayton. His current research interests include the development from Price Waterhouse, the Caterpillar Quality Improvement Award, the Dow 79
26 of new and innovative NDE techniques for degradation and the early detection Outstanding Faculty Award from American Society of Engineering Education, 80
27 of damage in aerospace metallic and composite materials. and the Ralph R. Teetor Educational Award from SAE. 81