NDT&E International 79 (2016) 142–149

Contents lists available at ScienceDirect

NDT&E International
journal homepage: www.elsevier.com/locate/ndteint

Pulse-modulation eddy current inspection of subsurface corrosion

in conductive structures
Yong Li n, Bei Yan, Da Li, Haoqing Jing, Yili Li, Zhenmao Chen
State Key Laboratory for Strength and Vibration of Mechanical Structures, Shaanxi Engineering Research Center of NDT and Structural Integrity Evaluation,
School of Aerospace, Xi'an Jiaotong University, ShaanXi, Xi'an 710049, China

art ic l e i nf o a b s t r a c t

Article history: Due to corrosive and hostile environment, in-service conductive structures are prone to subsurface
Received 7 September 2015 corrosion which has posed a severe threat to structural integrity and safety. Although Pulsed eddy
Received in revised form current testing (PEC) has been found advantageous over other Electromagnetic Non-destructive Eva-
19 January 2016
luation (ENDE) techniques particularly in detection and characterisation of subsurface defects in con-
Accepted 2 February 2016
ductive structures, it is subject to technical drawbacks. In light of this, in this paper, Pulse-modulation
Available online 8 February 2016
eddy current technique (PMEC) is proposed in an effort to enhance the inspection sensitivity to sub-
Keywords: surface corrosion and quality of corrosion imaging. Closed-form expressions of PMEC responses to
Electromagnetic non-destructive evaluation subsurface corrosion are formulated via the Extended Truncated Region Eigenfunction Expansion
Pulse-modulation eddy current
(ETREE) modelling. A series of simulations are subsequently conducted to analyse the characteristics of
Corrosion
PMEC signals and inspection sensitivity. Following this, experiments of PMEC for evaluation and imaging
Analytical modelling
Signal processing of subsurface corrosion are carried out. Through theoretical and experimental investigation, it has been
Sensitivity analysis found that PMEC is advantageous over PEC in terms of evaluation sensitivity and quality of corrosion
imaging.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction frequency harmonics) to infinite depth (due to DC). Whereas, close-up

Compared with traditional Eddy Current (EC) techniques including
Single- or Multi-frequency Eddy Current [1] and Sweep-frequency
Eddy Current [2,3], Pulsed Eddy Current technique (PEC) has been
found to be advantageous particularly in evaluation, characterisation
and identification of subsurface defects in in-service conductive
structures [4–7]. The significant difference between PEC and other EC
lies in the waveform of the excitation current which drives the exci-
tation coil for generation of the incident field. In lieu of sinusoidal
waveform for EC, PEC utilises the excitation current in a waveform
similar to the rectangular waveform, which is presented in Fig. 1(a). As
shown in Fig. 1(a), the PEC current signal can be discretised into an
infinite train of sinusoidal waveforms. This indicates that the infor-
mation regarding structural integrity, which could be extracted from
EC with a large number of sinusoidal excitations with various fre-
quencies, can be acquired by using PEC with a single excitation pro-
cess [8].
In consideration of skin effect of electromagnetic field and spectral
analysis of the PEC excitation current, it is noticeable that PEC actually
implements the inspection from the conductor surface (due to high-
n frequencies of the carrier and modulation signals, whilst the char-
Corresponding author at: School of Aerospace, Xi'an Jiaotong University, No. 28
Xianning West Road, Shaanxi, Xi'an 710049, China. Tel.: þ 86 29 82665721.
E-mail address: yong.li@mail.xjtu.edu.cn (Y. Li).
investigation of Fig. 1(a) indicates the technical drawback which
undermines PEC. Even though the rising and falling parts of the cur-
rent signal influence the harmonics in excitation, a large amount of
excitation energy is allocated to DC component which is barely useful
for generation of eddy currents, whilst the energy allocated to exci-
tation harmonics corresponding to eddy-current penetration depths
up to a conductor thickness is considerably lower. Such technical
drawback hinders the enhancement of inspection sensitivity and
evaluation accuracy of PEC, even though researchers have proposed a
number of counter-measures in probes and signal processing, etc.
[4,9,10].
The demand regarding mitigation of the technical drawback opens
up the pursuit of an appropriate waveform for the excitation current,
which allocates sufficient excitation energy to harmonics inducing
eddy currents within the conductor effective for evaluation of defects
at the depth of interest. In light of this, this paper proposes Pulse-
modulation eddy current technique (PMEC) which employs the exci-
tation current in Pulse Modulation Waveform (PMW) normally
adopted in millimetre wave testing [11] and radar technologies [12].
From Fig. 1(b), it can be seen that the excitation energy could be
allocated to the effective harmonics by adjusting the fundamental
acteristics similar to PEC in terms of broad-band and low-power-
consumption excitation are secured.

http://dx.doi.org/10.1016/j.ndteint.2016.02.001
0963-8695/& 2016 Elsevier Ltd. All rights reserved.
 Y. Li et al. / NDT&E International 79 (2016) 142–149 143

Fig. 2. A cylindrical probe of PMEC above a stratified conductor in the truncated

region.

probe is placed over a stratified conductor, which is shown in

1
Fig. 2. The probe consists of: (1) an excitation coil generating the
primary/incident field to induce eddy currents in the conductor
Amplitude (arb)

0.5
and (2) a solid-state magnetic field sensor placed at the centre of
0 the excitation coil for picking up transient signals of the net field
which is the superposition of the primary field and secondary field
-0.5
induced by eddy currents in the conductor.
-1 Through ETREE modelling, the closed-form expressions of
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
transient signals B(t) acquired from the magnetic field sensor
Time (s)
when the excitation coil is driven by an excitation current in
Normalised Amplitude (arb)

1 arbitrary waveform can be written as [14,15]

8
>
> B ðt Þ ¼ B 1 ðt Þ þ B 2 ðt Þ
>
> X 1
>
>   
>
< B 1 ðt Þ ¼ M N υ1 e  ai z1  e  ai z2 eai zs2  eai zs1 Primary field
0.5
i¼1 ; ð1Þ
>
> X
1
>
>   
>
> B 2 ðt Þ ¼ M N υ2 e  ai z2  e  ai z1 e  ai zs2  e  ai zs1 Secondary field
>
:
i¼1

0
0 200 400 600 800 1000
Frequency (Hz)
Fig. 1. Temporal signals and spectra of: (a) PEC excitation current with the unit
amplitude and fundamental frequency of 20 Hz and (b) the unit-amplitude pulse
modulation waveform with frequencies of the carrier and modulation signals at
100 Hz and 20 Hz, respectively.
In this paper, PMEC for detection, evaluation and imaging of sub-
surface corrosion which has been posing a severe threat to integ-
rity and safety of conductive structures is intensively investigated
via theory and experiments. The advantage of PMEC over PEC is
identified. The rest of the paper is organised as follows: Section 2
presents theoretical investigation of PMEC based on the Extended
Truncated Region Eigenfunction Expansion (ETREE) modelling [13].
The expressions of the PMEC signal and its response to initial sub-
surface corrosion are formulated. The comparison of inspection sen-
sitivity between PMEC and PEC is conducted via a series of simula-
tions. The experimental investigation regarding PMEC of localised
subsurface corrosion and comparison of PMEC with PEC in terms of
evaluation sensitivity and imaging quality are elaborated in Section 3.
2. Theoretical investigation
2.1. Formulation of PMEC signals
Since both PMEC and PEC are related to the transient eddy
current problem, ETREE modelling previously utilised for PEC
modelling could be applicable for PMEC. Suppose that a cylindrical
1200 1400 1600 1800 2000 where,
8 2πμ N J ða r Þχ ða r ;a r Þ
< M ¼ rs ðr2  r1 Þðz2 0z1 Þðzs2  zs1 Þ; N ¼ 1 i s i 12 5i 2
½hJ 0 ðai hÞ ai
: ð2Þ
: υ ¼ 1 R þ 1 I ðωÞejωt dω; υ ¼ 1 R þ 1 I ðωÞη ðωÞejωt dω
1 π 1 2 π 1 i
In Eq. (2), ω is the angular frequency of each harmonic within
the pulse excitation. m0 is the permeability of vacuum. Jn denotes
the Bessel function. N is the number of turns of the coil. h stands
for the radial distance of truncated solution region. ai is the posi-
tive root of J1(aih)¼0. ηi(ω) is the conductor reflection coefficient
corresponding to each harmonic, which can be calculated by using
the equations presented in [14,15]. The coil coefficient χ(air1, air2)
can be computed by referring to the identity in [1].
It is noted that the harmonics I(ω) within the excitation current
are readily computed by means of Fourier Transform (FT) of the
temporal current signal, and thus the transient field responses are
recovered by using Inverse Fourier Transform (IFT) of spectral field
signals over entire harmonics. Whereas, further analysis has
revealed that the computation of PMEC signals takes time as FT of
the excitation current needs to be computed and a number of
excitation harmonics should be taken into account.
In consideration of Fourier theorem, the derivation of I(ω) by using
FT of the excitation current could be neglected by using convolution of
time-domain signals [16]. Therefore, the expressions of signals regar-
ding primary and secondary fields in Eq. (1) are modified as
" #
X
1   
B1 ðt Þ ¼ I ðt Þ U ψ ¼ I ðt Þ U M N e  ai z1  e  ai z2 eai zs2  eai zs1 ð3Þ
i¼1
144 Y. Li et al. / NDT&E International 79 (2016) 142–149

" #
X
1    Table 1
B 2 ðt Þ ¼ I ðt Þ  ξ ðt Þ ¼ I ðt Þ  M N e  ai z2  e  ai z1 e  ai zs2  e  ai zs1 ηi ðt Þ ð4Þ Parameters of the simulation model.
i¼1

where  denotes convolution. In each expansion term, ηi(t) is derived r2 (mm) r1 (mm) H (mm) z1 (mm) z2 (mm) N
from IFT of ηi(ω). It is noticeable from Eqs. (3) and (4) that ψ and ξ(t)
are independent of the excitation current. ψ depends on the para- 5.5 4.6 26.3 1.5 27.8 1175
h (mm) d1 (mm) σ1 (MS/m) rs (mm) zs1 (mm) zs2 (mm)
meters of the probe, and keeps constant regardless of the excitation 500 10.0 24.7 0.9 0.5 1.0
current. Since ηi(t) is only governed by the frequency of each harmonic
with unit amplitude, therefore ξ(t) depicts the temporal response of
eddy currents in the conductor under illumination of the primary field
which is generated by the probe driven by an unit-step current. The 1
expressions of the PMEC signal can be formulated by substituting the
0.8
analytical expression of the excitation current of PMEC into Eqs.
(3) and (4), and written as 0.6
8 " #
> X1
cosðω1 t Þ 2 X 1
cosðω2 t Þ  
< Bðt Þ ¼ A sinðω t Þ þ 2   ψ þ ξ ðt Þ
0.4
2
c
πn ¼ 1 n πn ¼ 1 n
>
: 0.2
ω1 ¼ ωc  nωm ; ω2 ¼ ωc þ nωm

I(t) (A)
0
ð5Þ
where n are odd integers, n¼1, 3, 5….ωc and ωm are angular fre- -0.2

quencies of the carrier and modulation signals, respectively. A denotes -0.4

the amplitude of the carrier signal.
-0.6

2.2. Computation of PMEC signals and corroboration -0.8

-1
Based on Eqs. (1)–(5), there are basically 3 approaches applicable 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
for predicting the transient signals of PMEC inspection of a stratified Time (s)
conductor: (1) the FT-based method; (2) superposition method; and
(3) convolution-based method. 0.04
The FT-based and convolution-based methods are based on Eqs.
(1) and (5), respectively. In contrast, regarding the superposition 0.03
method [17], the excitation current in PMW is firstly discretised into a
series of sinusoidal signals by referring to the analytical expression of I(t) 0.02
indicated in Eq. (5). The PMEC signal is subsequently recovered by
superimposing the multiple temporal signals in sinusoidal waveform 0.01
corresponding to computed B(ωc), B(ω1) and B(ω2). The number of
Bz (T)

sinusoidal signals for field excitation is determined by comparison of I(t) 0

obtained by the analytical expression with that acquired by direct
multiplication of the carrier and modulation signals, and thus set as -0.01

2001 due to sufficient accuracy reached. It is noted particularly for the

-0.02
convolution-based method that 2001 harmonics with unit amplitude TSFEM
are also adopted for computation of ηi(t). FT and IFT of signals are Convolution-based
-0.03
computed numerically and efficiently by using Fast FT (FFT) and Inverse Superposition
FT-based
Fast FT (IFFT) which have been realised in MATLAB [8,14]. The com-
-0.04
putation of convolution is readily implemented by using the routine 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
‘conv’ in MATLAB. Time (s)
In a bid to verify and evaluate the established PMEC model, cross Fig. 3. (a) The PMW current and (b) computed PMEC signals.
comparison of the predicted PMEC response to a non-magnetic con-
ductive slab acquired from 3 approaches and Time-Stepping Finite The comparison of the PMEC signals from 3 approaches and TSFEM
Element Modelling (TSFEM) [18] is carried out. The Normalised Root is shown in Fig. 3(b). The computing time along with NRMSD of the
Mean Squared Deviations (NRMSD) between the 3-approach results approaches are tabulated in Table 2. It can be seen from Fig. 3(b) that
and those from TSFEM are worked out for comparison of accuracy,
the computed PMEC signals by means of 3 approaches have good
whilst the computing time is obtained to evaluate the computation
agreement with the result from TSFEM. From Table 2 it is noticeable
efficiency.
that the convolution-based method has highest computational accu-
The parameters of the probe and conductor are listed in Table 1. A
racy and efficiency. As a result, the convolution-based method is
PMW current is supplied with the probe, whose maximum amplitude
preferred for the following theoretical investigation.
is 1A. The frequency of the PMW carrier signal is selected in con-
sideration of the Standard Penetration Depth (SPD) of electromagnetic
field and the rule of thumb regarding PMW: at least 1 cycle of the 2.3. Sensitivity analysis
carrier signal within the pulse width of the modulation signal (othe-
rwise, the PMW current is analogous to the typical excitation current In the presence of subsurface corrosion, the apparent electrical
used in PEC). In simulations, the fundamental frequencies of the car- properties of the conductor vary, whilst from the view of sizing, the
rier and modulation signals are set as fc ¼100 Hz (preferred integer thickness of the conductor decreases from the bottom surface.
corresponding to approximately 10 mm in SPD with the conductor) Therefore, Material Degradation (MD) and Wall-thinning flaw (WT)
and fm ¼20 Hz. The PMW current is shown in Fig. 3(a). are introduced to PMEC sensitivity analysis complementary to
 Y. Li et al. / NDT&E International 79 (2016) 142–149 145

experiments with localised surface corrosion in an Aluminium slab. facilitate the extraction of PVs from the differential signals with
Comparison of sensitivity between PMEC and PEC to MD and WT is absolute amplitudes. In light of the fact that the magnitudes of signals
conducted in an attempt to identify the advantage of PMEC over PEC. of PMEC and PEC are in different scales, PVs are normalised to the
Regarding MD, σ1 varies from 100% to 50% of the base conductivity extrema of reference signals Bmax of PMEC and PEC for sensitivity
(34 MS/m), whilst d1 ¼10 mm. In contrast, d1 varies from 10 mm comparison. For various defect scenarios, the sensitivities to MD and
down to 1 mm while σ1 ¼34 MS/m for WT. The other modelling WT are thus calculated by using the equation written as:
parameters are listed in Table 1. Differential signals of PMEC and PEC    
PV σ PV d
are calculated to acquire the responses of PMEC and PEC to either MD Sσ ¼ Δσ 1 ; Sd ¼ Δd 1 : ð6Þ
Bmax Bmax
or WT by subtracting the reference/defect-free signals into those with
defects. Peak Values (PVs) are extracted from the differential signals. In order to analyse the entire sensitivity of either PMEC or PEC
Particularly for PMEC, envelope-detection algorithm is employed to when MD and WT develop, it is imperative to compute the signals
for initial MD (with Δσ1 ¼0) and WT (with Δd1 ¼0). In such case,
Table 2 the initial sensitivity Sini is expressed as
Cross comparison between 3 approaches and TSFEM.
" ! # " ! #
PV ini
σ PV ini
Convolution- Superposition FT-based TSFEMa Sini
σ ¼ lim Δσ 1 ; S ini
d ¼ lim d
Δ d1 : ð7Þ
based method method method Δσ 1 -0 Bmax Δd1 -0 Bmax

NRMSD (%) 0.7 1.5 1.8 N/A Since σ1 and d1 influence the conductor reflection coefficient,
Computing 0.2 1.2 0.6 857.6
consequently the expression of the initial differential signal where
timeb (s)
PVini is extracted can be formulated by substituting the first-order
a derivative of ηi(t) with respect to either σ1 or d1 for the original
Number of elements: 207,337.
b
Computation implemented in a computer with Intels Core™2 Duo 2.93 GHz ηi(t). For a non-magnetic conductive slab, the first-order derivative
CPU, 4 GB RAM. of the conductor refection coefficient corresponding to each

Fig. 4. Predicted differential signals vs. MD and WT for (a) PMEC and (b) PEC.
146 Y. Li et al. / NDT&E International 79 (2016) 142–149

harmonic within the excitation current is written as where,

8
8  h i >
> p ¼ 2a1 ai cosh ða1 d1 Þ
< jωμ0 ai ð2a1 m1  m2 Þ = 2a1 ðp þ qÞ2 for σ 1 > q ¼ a2 þ a2 sinh ða d Þ
>
>
dηi ðωÞ ¼  ð8Þ >
> 1 1
: 2a a2 a2  a2 =ðp þ qÞ2
>
<
i
 1   
for d1 m1 ¼ ai 1  cosh ð2a1 d1 Þ þ d1 a2i  a21
i 1 i 1
  ð9Þ
>
>
> m2 ¼qaffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i þa1 sinh ð2a1 d1 Þ
2 2
>
>
> ffi
>
>
: a1 ¼ a2i þ jωμ0 σ 1

The first-order derivative of ηi(t) with respect to either σ1 or d1

is thus readily computed with IFT which is implemented by using
the routine ‘ifft’ in MATLAB. Following this, PVini is extracted from
the calculated initial differential signal, and thus the initial sensi-
tivity is computed via PVini/Bmax.
The predicted differential signals of PMEC and PEC against MD and
WT are presented in Fig. 4 whilst Fig. 5 shows the sensitivity curves of
PMEC and PEC to MD and WT during their development. Note that in
order for better comparison, the excitation current of PEC is same as
the modulation signal of PMW. Its maximum amplitude is 1 A. Further
investigation has revealed that the fundamental frequency of the
excitation current varying from 10 Hz up to 500 Hz, which is preferred
especially for PEC, barely affects PVs. Therefore, the fundamental fre-
quency and duty cycle of the PEC excitation current are set as 20 Hz
and 50%, respectively.
It can be found from Fig. 4 that the magnitude of the differential
signal of PMEC decreases with degree of MD as well as depth of WT
increased. This agrees with that of PEC, and is due to suppression of
eddy currents in the slab when either MD or WT develops. From
Fig. 4, it can be seen that the sensitivities of PMEC and PEC have
monotonic relationships with the degradation degree of MD and
depth of WT. They rise when degradation degree of MD as well as
depth of WT increases. Even though the sensitivities of PMEC and PEC
vary when MD and WT develop, it is noteworthy from Fig. 5 that the
sensitivity of PMEC is higher than that of PEC. This indicates that
PMEC is more sensitive to subsurface corrosion in conductive struc-
tures than PEC, which could benefit the evaluation and imaging of
localised subsurface corrosion.

3. Experiments

In an effort to further investigate PMEC for evaluation and imaging

of subsurface corrosion, a series of experiments have been carried out
by using a customised system implementing not only PEC but also
PMEC. Localised subsurface corrosion in different sizes and shapes are
introduced to experimental study.

3.1. Experimental setup

An experimental system integrating PEC and PMEC has been built

Fig. 5. Sensitivity curves of PMEC and PEC vs. (a) MD and (b) WT. up. The schematic illustration of the system is presented in Fig. 6. A

Fig. 6. Schematic illustration of the experimental system.

 Y. Li et al. / NDT&E International 79 (2016) 142–149 147

net magnetic field, and deployed at the bottom centre of the excita-
tion coil. This complies with the model setup in the theoretical study.
The stand-off distance between the sensor and the surface of the
sample is 1 mm (zs1 ¼1 mm). It is noted that the waveforms of exci-
tation currents of PMEC and PEC are generated by the signal gen-
erator, and the signal from the power amplifier which drives the
excitation coil has the maximum amplitude of 0.5 V.
An aluminium alloy plate (length  width  thickness: 500 
500  6 mm3) is utilised in experiments. By referring to the freq-
uency-selection strategy of PMEC, the frequencies of the carrier and
modulation signals are set as 300 Hz and 20 Hz, respectively. In con-
trast, the frequency and duty cycle of the PEC excitation current are set
as 20 Hz and 50%, respectively. In a bid to simulate subsurface cor-
rosion, different localised corrosions are fabricated on the bottom
surface of the sample: (1) Circle-shaped Corrosion (CSC) with depths
of 4 mm and 5 mm, and diameters of 20 mm and 30 mm; and (2)
Square-shaped Corrosion (SSC) with depths of 4 mm and 5 mm, and
width of 20 mm.

3.2. Experimental signals and signal processing

The experimental investigation starts with the analysis of acquired

signals and comparison of inspection sensitivity between PMEC and
PEC. The measured reference signals of PMEC and PEC, and the defect
signals when the probe is deployed right over the SSC (depth  width:
4  20 mm2) are shown in Fig. 7(a). The differential signals derived
from subtraction of reference signals from defect signals are computed
and presented in Fig. 7(b).
It can be seen from Fig. 7 that the signal trajectories acquired from
experiments agree with those from simulations. The signal amplitude
increases in the presence of the corrosion whilst the magnitude of the
PEC signal is higher than that of the PMEC signal. This is due to the
lower-amplitude excitation current of PMEC, which results from
(1) frequency harmonics of the PMEC signal are higher than those of
the PEC signal and (2) the constant-voltage instead of constant-
current excitation source is used in experiments. Whereas, it is
noticeable from Fig. 7(b) that not only the amplitude of the PMEC
differential signal but particularly its extremum is larger than that of
the PEC differential signal. This also applies in further investigation
with other subsurface corrosion. As a result, it implies that for indi-
vidual corrosion the inspection sensitivity of PMEC is higher than that
of PEC, which agrees with the conclusion drawn from theoretical inve-
stigation.

3.3. Imaging of subsurface corrosion

The imaging of subsurface corrosion by using PMEC and PEC

individually is carried out by scanning the probe over the corrosion
region. The schematic illustration of the 2D scanning is shown in
Fig. 8. The coordinate of the corrosion centre is X¼0, Y¼0 within the
probe scanning zone. The scanning curves depicting PVs extracted

Fig. 7. Experimental signals of PMEC and PEC: (a) measured signals of z-compo-
nent of the net magnetic field and (b) differential signals.

probe is fabricated in order to realise PEC and PMEC. The parameters

of the excitation coil are listed in Table 1. As one of the solid-state
magnetic field sensors, the Tunnel Magneto-resistance (TMR) sensor
(MultiDimension TMR-MMLP57F) is used to sense z-component of the Fig. 8. Schematic illustration of the probe scanning for corrosion imaging.
148 Y. Li et al. / NDT&E International 79 (2016) 142–149

Corrosion images via PMEC

-30 -30 -20 -20 -20 0.09

Normalised PV (arb)
-20 -20 0.075

Y poistion (mm)
Y poistion (mm)

Y poistion (mm)

Y poistion (mm)

Y poistion (mm)
-10 -10 -10
-10 -10 0.06
0 0 0 0 0 0.045
10 10 0.03
10 10 10
20 20 0.015
30 30 20 20 20
-20 0 20 -20 0 20 -20 0 20 -20 0 20 -20 0 20
X poistion (mm) X poistion (mm) X poistion (mm) X poistion (mm) X poistion (mm)

Corrosion images via PEC

-30 -30 -20 -20 -20 0.09

Normalised PV (arb)
0.075
Y poistion (mm)

Y poistion (mm)

Y poistion (mm)

Y poistion (mm)

Y poistion (mm)
-15 -15 -10 -10 -10
0.06
0 0 0 0 0 0.045
0.03
15 15 10 10 10
0.015
30 30 20 20 20
-30 -15 0 15 30 -30 -15 0 15 30 -20 0 20 -20 0 20 -20 0 20
X poistion (mm) X poistion (mm) X poistion (mm) X poistion (mm) X poistion (mm)

Fig. 9. Images of various subsurface corrosion vs. PMEC and PEC: (a) CSC in 30 mm  5 mm; (b) CSC in 30 mm  4 mm; (c) CSC in 20 mm  5 mm; (d) CSC in 20 mm  4 mm;
and (e) SSC in 20 mm  5 mm.

Table 3
Computed SSIMs of acquired corrosion images.

Corrosion CSC (diameter  depth) SSC (length  depth)

2 2 2 2
30  5 mm 30  4 mm 20  5 mm 20  4 mm 20  5 mm2

PMEC 0.602 0.593 0.589 0.581 0.614

PEC 0.592 0.584 0.579 0.573 0.608

from differential signals vs. probe positions are obtained to construct higher than those of PEC. This indicates that for imaging of localised
the images of the subsurface corrosion. It is noted that PV at every subsurface corrosion PMEC provides corrosion images with higher
probe position is normalised to the extrema of reference signals Bmax quality than PEC due to its characteristics of dedicated inspection of
in order for further comparison. conductors. From the aspect of corrosion imaging the advantage of
The acquired corrosion images of PEC and PMEC are exhibited PMEC over PEC in terms of inspection sensitivity is further identified.
in Fig. 9 with the true profile (top view) of each corrosion indi- Therefore, corrosion imaging via PMEC is more beneficial to detection,
cated by the region with the solid-line boundary. Every image is characterisation and quantitative evaluation of localised subsurface
presented in grey scale with black and white corresponding to the corrosion in conductive structures.
minimum and maximum values of all processed PVs, respectively.
It is found from Fig. 9 that even though each corrosion image is
‘blurred’ due to the shielding effect of the conductive region above 4. Conclusion
corrosion against electromagnetic field, the corrosion profile could
be indicated by using the acquired corrosion image via both PMEC Following the advancement of PEC, this paper proposes a new
and PEC. Whereas, the intuitive comparison of the corrosion image ENDE technique, namely PMEC in an effort to enhance the
between PMEC and PEC reveals that the image quality of PMEC is inspection sensitivity and image quality of subsurface corrosion
higher than that of PEC. The Structural Similarity Index (SSIM) [19] within conductors. ETREE modelling of PEC is extended and made
of corrosion images of PMEC and PEC are subsequently computed applicable for PMEC modelling. Through ETREE modelling,
in a bid to evaluate and compare the image quality. The computed expressions of PMEC responses to layered conductors and its
SSIMs are listed in Table 3. Note that the reference image of true initial sensitivity to a conductive slab with varying thickness and
corrosion profile for calculating SSIM comprises the corrosion area conductivity are formulated. Based on the cross-comparison
set in white and remaining region (out of the corrosion boundary) regarding computation of PMEC signals among the convolution-
in black. based method, superposition method and FT-based approach, the
It can be noticeable from Table 3 that the calculated SSIM former is preferred in theoretical study of the PMEC response and
decreases when the volume of CSC is decreased. The reasoning is that its sensitivity to MD and WT. From simulations, PMEC is found
the perturbation of magnetic field from CSC becomes weak with its superior to PEC particularly in evaluation sensitivity to MD and
volume dropping, and in other words the shielding effect from con- WT when they develop.
ductive region above CSC is enhanced. Interestingly, the image quality In parallel to the theoretical investigation, a series of experi-
of SSC is found better than that of CSC. This could be due to the ments are conducted to further analyse PMEC for evaluation and
characteristics of SSC profile particularly its corners which could be imaging localised subsurface corrosion in an Aluminium alloy and
taken as the scattering points under electromagnetic illumination, and its comparison with PEC. It can be found from experimental results
thus enhances the imaging sensitivity to corrosion edges. It can also be that the PV extracted from the differential signal of PMEC for
found that the computed SSIMs of corrosion images using PMEC are subsurface corrosion has larger magnitude than that of PEC, which
 Y. Li et al. / NDT&E International 79 (2016) 142–149
indicates the advantage of PMEC in evaluation sensitivity to
localised corrosion over PEC. Through investigation of corrosion
imaging, the quality of the corrosion image acquired using PMEC is
found better than that using PEC. This further implies the super-
iority of PMEC to PEC, and potential of PMEC in evaluation and
imaging of subsurface corrosion with higher sensitivity and
accuracy.
Acknowledgements
The authors would like to thank the National Natural Science
Foundation of China (Grant no. 51477127) and National Magnetic
Confinement Fusion Program of China (Grant no. 2013GB113005)
for funding this research.
References
[1] Li Y, Chen ZM, Mao Y, Qi Y. Quantitative evaluation of thermal barrier coating
based on eddy current technique. NDTE Int 2012;50:29–35.
[2] Betta G, Ferrigno L, Laracca M, Burrascano P, Ricci M, Silipigni G. An experi-
mental comparison of multi-frequency and chirp excitations for eddy current
testing on thin defects. Measurement 2015;63:207–20.
[3] Diraison YL, Joubert PY, Placko D. Characterisation of subsurface detects in
aeronautical riveted lap-joints using multi-frequency eddy current imaging.
NDTE Int 2009;42:133–40.
[4] Hosseini S, Lakis AA. Application of time-frequency analysis for automatic
hidden corrosion detection in a multilayer aluminium structure using pulsed
eddy current. NDTE Int 2012;47:70–9.
[5] He YZ, Pan MC, Luo FL, Tian GY. Pulsed eddy current imaging and frequency
spectrum analysis for hidden defect non-destructive testing and evaluation.
NDTE Int 2011;44:344–52.
149
[6] Adewale ID, Tian GY. Decoupling the influence of permeability and conductivity in
pulsed eddy-current measurements. IEEE Trans Magn 2013;49:1119–27.
[7] Xie SJ, Chen ZM, Takagi T, Uchimoto T. Quantitative non-destructive evaluation
of wall thinning defect in double-layer pipe of nuclear power plants using
pulsed ECT method. NDTE Int 2015;75:87–95.
[8] Tian GY, Li Y, Mandache C. Study of lift-off invariance for pulsed eddy current
signals. IEEE Trans Magn 2009;45:184–91.
[9] Arjun V, Sasi B, Purna B, Rao C, Mukhopadhyay CK, Jayakumar T. Optimisation
of pulsed eddy current probe for detection of sub-surface defects in stainless
steel plates. Sens Actuators A: Phys 2015;226:69–75.
[10] Espina-Hernández JH, Ramírez-Pacheco E, Caleyo F, Pérez-Benitez JA, Hallen
JM. Rapid estimation of artificial near-side crack dimensions in aluminium
using a GMR-based eddy current sensor. NDTE Int 2012;51:94–100.
[11] Xu QF, Liu Q, Ye Q, Pan ZQ, Cai HW, Qu RH, Fang ZJ. Millimeter-wave pulse
generation based on pulse reshaping using superstructure fiber bragg grating.
Optik – Int J Light Electron Opt 2010;121:1853–8.
[12] Seyfried D, Schoebel J. Detection capability of a pulsed ground penetrating
radar utilizing an oscilloscope and radargram fusion approach for optimal
signal quality. J Appl Geophys 2015;118:167–74.
[13] Li Y, Theodoulidis T, Tian GY. Magnetic field-based eddy-current modeling for
multilayered specimens. IEEE Trans Magn 2007;43:4010–5.
[14] Li Y, Tian GY, Simm A. Fast analytical modelling for pulsed eddy current
evaluation. NDTE Int 2008;41:477–83.
[15] Li Y, Chen ZM, Qi Y. Generalized analytical expressions of liftoff intersection in
PEC and a liftoff-intersection-based fast inverse model. IEEE Trans Magn
2011;47:2931–4.
[16] Chen XL, Lei YZ. Electrical conductivity measurement of ferromagnetic
metallic material using pulsed eddy current method. NDTE Int 2015;75:33–8.
[17] Li Y, Liu XB, Chen ZM, Zhao HD, Cai WL. A fast forward model of pulsed eddy
current inspection of multilayered tubular structures. Int J Appl Electromagn
Mech 2014;45:417–23.
[18] Xie SJ, Chen ZM, Takagi T, Uchimoto T. Efficient numerical solver for simula-
tion of pulsed eddy current testing signals. IEEE Trans Magn 2011;47:4582–91.
[19] Wang Z, Bovik A, Sheikh H, Simoncelli E. Image quality assessment: from error
visibility to structural similarity. IEEE Trans Image Process 2004;13:600–12.