IEEE SENSORS JOURNAL, VOL. 12, NO.

6, JUNE 2012 2113

Steel Corrosion Characterization Using

Pulsed Eddy Current Systems
Yunze He, Student Member, IEEE, Guiyun Tian, Senior Member, IEEE, Hong Zhang, Student Member, IEEE,
Mohammed Alamin, Student Member, IEEE, Anthony Simm, Member, IEEE, and Paul Jackson

Abstract— A pulsed eddy current (PEC) system has been Atmospheric corrosion is perhaps the most widely recog-
used to characterize atmospheric corrosion on steel samples. nised form of environmental degradation and accounts for
International Paint has supplied coated and uncoated mild steel considerable losses to many structural and industrial compo-
(S275) samples with marine atmospheric corrosion (exposure
time between 1 and 10 months). The PEC response due to cor- nents. Numerous works have studied atmospheric corrosion
rosion is a complex mix of many factors, including conductivity, and most of such publications concern long-term exposure
permeability and material thickness variation, which are all taken [4]. The simplest corrosion tests (such as those defined by the
into account through experimental studies and the extraction of ISO-DIS 9223 standard) are measurements of the mass loss
signal features. Considering the conductivity and permeability (g/(m2·a)) or thickness loss (µm/a) [5]. The variation of long-
variation in the corrosion layer or the actual rust region, two
time-domain features, each representing the conductivity and per- term atmospheric corrosion over time is known to conform to
meability, are extracted and used to characterize corrosion. The a power function:C = At n based on mass or thickness loss
relationship between PEC features and exposure time has been tests. Where, t is the exposure time in years [6], C is corrosion
derived, which can be useful for corrosion rate measurement and depth [7] or sample thickness loss [8] after a period t, A is
early-stage corrosion evaluation and prediction. In conclusion, unchanged in the first year, and n is constant. In the model,
PEC testing has the potential for corrosion characterization and
monitoring in areas such as the marine industry. corrosion A in the first year is an important parameter as it
allows not only the level of corrosion of a given environment to
Index Terms— Corrosion characterization, feature extraction, be calibrated, but also long-term corrosion to be predicted [9].
pulsed eddy current, steel.
However, in practice these tests consist of preparing metallic
coupons, cleaning them before testing, weighing them before
I. I NTRODUCTION exposure, exposing them to the corrosion media, post-test

S TEEL components are used in many industries, such as removal of visible corrosion products, and reweighing [10].
shipping, off-shore oil and gas production and power The process of these tests takes a long time and cannot provide
plants. Corrosion is the destructive attack of a metal and its real-time measurement.
properties by chemical or electrochemical reaction with its Several methods can be used for corrosion characterisation.
environment [1]. The direct cost of corrosion in countries Electrochemical impedance spectroscopy (EIS) is an elec-
throughout the world is estimated to be between 3% and 4% trochemical technique, which can be used to test corrosion
of the Gross Domestic Product (GDP) of each country [2]. rate [11] and to investigate the protective properties of a
Corrosion testing and detection is an increasingly important coating [12]. Firstly, the electrochemical tests provide a real-
area in industry. Recent developments in metallic materials time measurement of the corrosion rate. Secondly, they can
have heightened the need for reliable corrosion detection and provide time-corrosion rate data on a single coupon. Thirdly,
monitoring methods [3]. they are quick to perform. Most of the work using this
technique to investigate atmospheric corrosion is focused on
Manuscript received October 14, 2011; revised December 14, 2011; the corrosion behaviour and mechanisms [13], [14]. However,
accepted January 1, 2012. Date of publication January 13, 2012; date of
current version April 26, 2012. This work was supported in part by the Engi- these electrochemical tests include the need for comparatively
neering and Physical Sciences Research Council under Grant EP/E005071, expensive equipment and high levels of technical expertise for
U.K. The associate editor coordinating the review of this paper and approving data analysis [15]. It is also not ideal tool for in-situ corrosion
it for publication was Dr. Patrick Ruther.
Y. He is with Newcastle University, Newcastle upon Tyne NE1 7RU, U.K., detection and monitoring.
and also with the National University of Defense Technology, Hunan 410073, In addition to this method, Non-destructive Testing and
China (e-mail: [email protected]). Evaluation (NDT&E) techniques based on acoustic, thermal,
G. Tian, H. Zhang, and M. Alamin are with Newcastle Univer-
sity, Newcastle upon Tyne NE1 7RU, U.K. (e-mail: [email protected]; electromagnetic radiation and other effects, to monitor the
[email protected]; [email protected]). inherent properties of materials are also used to characterise
A. Simm is with the School of Electrical, Electronic and Computer corrosion. The acoustic emission (AE) technique, based on
Engineering, Newcastle University, Newcastle upon Tyne NE1 7RU, U.K.
(e-mail: [email protected]). the rapid release of energy within a material generating a
P. Jackson is with International Paint Ltd., Tyne and Wear NE10 0JY, U.K. transient elastic wave propagation, is used for monitoring
(e-mail: [email protected]). corrosion and pitting corrosion [16], [17]. Thermography
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. has the ability to detect corrosion under paint [18]. Radi-
Digital Object Identifier 10.1109/JSEN.2012.2184280 ographic methods provide images of the variation in density
1530–437X/$31.00 © 2012 IEEE

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.
2114 IEEE SENSORS JOURNAL, VOL. 12, NO. 6, JUNE 2012

of metallic components, however its use has safety issues

[19]. In addition, some electromagnetic NDT&E methods
have been investigated for corrosion detection and monitoring
[20], including eddy current based structural health monitoring
(SHM) [21].
The early stages of corrosion (sometimes occurring under
coatings) should be detected and evaluated, which is important
in understanding the progression of corrosion, corrosion pro-
tection, lifetime extension and condition based maintenance.
However, most studies of atmospheric corrosion using mass or
thickness loss methods are used for the long-term corrosion
measurement and analysis. The electrochemical technique is
usually applied in the laboratory investigations to understand
corrosion mechanisms [22]. Pulsed eddy current (PEC) has
Fig. 1. The schematics of different steel samples: (a) un-corroded steel,
demonstrated its detection capabilities in stress measurement (b) corrosion with thickness increase, (c) corrosion with metal loss, and
[23], flaw detection [24], [25], and corrosion characterisation (d) corrosion replaced by air.
[26], [27]. From transient response analysis in PEC, several
features can be extracted to identify and classify the defects
in time-domain, as well as frequency-domain features [34]– hydroxide Fe(OH)3 , goethite α-FeOOH). These iron oxides
[36]. The RTD-Incotest [28], is a PEC system which measures and iron hydroxides have different electromagnetic proper-
induction current, rather than the magnetic field, has been used ties. For example, hematite is a semiconducting mineral [37]
for corrosion under insulation (CUI) detection. The system, and goethite shows lower conductivity than hematite [38].
developed by Röntgen Technische Dienst from the Nether- However, the conductivity of magnetite is significantly higher
lands, is based on the diffusive behaviour of eddy currents (×106) than hematite, due to electron exchange between the
related to the material properties and the wall thickness of the Fe2+ and Fe3+ ions [39]. Hematite exists weakly ferromag-
object. The eddy current signal is processed and compared to a netic above the Morin temperature (TM ≈260K). Goethite is
reference signal, which eliminates the material properties and anti-ferromagnetic at room temperature. However, maghemite
reads the average wall thickness within magnetic field area is ferromagnetic [38] and magnetite exhibits permanent mag-
results [29]. The presence of corrosion has been simplified netism and is ferrimagnetic [39].
using a “metal loss” type of defect or notch in the sample [30], Over time only the proportions of the corrosion constituents
as shown in Figs. 1(c) and (d). This approach has established change, there is little effect on the composition [40]. Because
the link between PEC differential peak-value feature and the the density of iron oxides and hydroxides is less than pure
dimensions of the metal loss [31], [32]. steel, the volume of corroded steel will change. The thickness
However, corrosion is more complex than this, within the of corrosion at the early stage will increase but the thickness of
corrosion layer or region, the conductivity and permeability long-term corrosion will decrease due to metal loss. Fig. 1(a)
are reduced [33], rather than being metal loss. In this paper, is the schematic of un-corroded steel sample. Fig. 1(b) shows
the conductivity and permeability variation in the corrosion the schematic of early-stage corrosion with thickness increase.
layer or the actual rust region and lift-off variation of coating d1 is corrosion height, d2 is corrosion depth, and d1 + d2 is
thickness are considered in the PEC responses as illustrated the corrosion layer thickness. Fig. 1(c) is the schematic of
in the Fig. 1(b). Normalised and non-normalised differential long-term corrosion (metal loss). Fig. 1(d) is the schematic
peak-values are investigated and compared for the variation of corrosion replaced by air, which is used in previous
of conductivity and permeability and applied for prediction work [31].
of corrosion progression over time, which will be reported in The work in this paper is focused on the characterisation
Section IV. of early-stage corrosion with an increase in thickness. Addi-
The paper is organised as follows. Firstly, the corrosion and tionally, many steel components include a coating material
sample design is introduced in Section II and the PEC experi- and corrosion occurs under the coating. Based on these
mental set-up is described in Section III. Next, the PEC feature considerations, uncoated and coated steel samples provided
extraction and optimisation are introduced in Section IV. Then, by International Paint, with corrosion produced in a marine
coated and uncoated steel corrosion with different exposure atmosphere are evaluated. Firstly, mild steel (S275) plate is cut
times are characterised and the fitted line and formula of into samples with length×width×height of 300×150×3 mm,
corrosion development are obtained in Section V. Finally, whose composition (wt %) is < 0.22 C, 0.05-0.15 Si, <0.65
conclusions are outlined in Section VI. Mn, <0.3 Ni, <0.05 S, <0.04 P, <0.3 Cr, <0.012 N, and
<0.3 Cu. Secondly, the mild steel samples are covered by
plastic tape except for a central area (30 mm×30 mm) to keep
II. C ORROSION AND S AMPLE D ESIGN
the steel beneath tape clean and dry. Thirdly, the samples are
Corrosion in steel is a general term for a series of iron exposed to a marine atmosphere for a certain time (t = 1,
oxides (hematite α-Fe2 O3 , magnetite Fe3 O4 , and maghemite 3, 6, and 10 months) to form the uncoated samples with
γ -Fe2 O3 ), and hydroxides (ferrous hydroxide Fe(OH), ferric atmospheric corrosion. Then, some of the uncoated samples

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.
HE et al.: STEEL CORROSION CHARACTERIZATION USING PEC SYSTEMS 2115

Fig. 2. Photographs of (a) uncoated and (b) coated sample.

TABLE I
S PECIFIC D ETAILS OF S AMPLE N UMBERING Fig. 3. Photograph of PEC set-up system.

Exposure time Uncoated Uncoated Coated

1 month 1 2 3
3 months 4 5 6
6 months 7 8 9
10 months 10 11 –

are covered using a non-conductive paint with a thickness of

∼100µm. Fig. 2 shows photographs of a typical uncoated
sample (a) and a coated sample (b). Details of the samples
are given in Table I. Fig. 4. (a) One half period of the nonnormalized PEC response. (b) One
half period of the normalized PEC response.

III. E XPERIMENTAL PEC S YSTEM

Fig. 3 shows the PEC experimental system [41]. The X-Y studies [23], [24], [42]. Then, their sensitivities are compared
scanning machine is controlled via a PC parallel interface. The and analysed in the terms of corrosion characterisation.
QinetiQ TRECSCAN system is used for PEC measurements.
A single period of an excitation waveform is created in Matlab A. PEC Response and Feature Extraction
and converted to an analog voltage signal by the Analog A square-wave contains rich frequency components. Each
Output (AO) subsystem of a DAQ board (NI PCI-6255). The frequency component relates to the information at different
voltage signal is converted to the excitation current using penetration depths. The penetration depth of every frequency
TRECSCAN. TRECSCAN operates in current excitation mode in the conductive material is governed by the skin effect,
with an exponentially damped square wave of duty cycle 50%, shown in equation (1)
repetition frequency of 200Hz and time constant Tc of 100µs. !
The excitation current is sent to the PEC probe. The probe δ = 1/ f σ πµ (1)
consists of a ferrite core, excitation coil with mean diameter
of 11 mm and a Hall sensor. The PEC response measured Where f is the excitation frequency, σ is the electrical
by the Hall sensor is low-pass filtered (1 kHz cut-off) and conductivity, and µ is the permeability of the material under
amplified by the TRECSCAN. This signal is digitised by the inspection. We can see from equation (1), as the frequency
Analog Input (AI) subsystem of the DAQ board. The PEC increases the skin depth will decrease.
response is acquired using a sampling rate of 500 kHz. The In this work, the excitation repetition frequency is 200 Hz.
excitation wave form control, data acquisition, analysis and According to [33], the electrical conductivity and relative
feature extraction are performed in Matlab. In the tests, the permeability of steel is 4.68×106 S/m and 60, respectively.
PEC probe is in contact with the sample surface (steel and The electrical conductivity and relative permeability of rust is
corrosion area) to eliminate lift-off effects. about 0.75×106 S/m and 4, respectively. So, the skin depth is
2.1 mm in steel, which is smaller than the thickness of sample
(3 mm), and the skin depth is approximately 20 mm in rust,
IV. F EATURE E XTRACTION AND O PTIMIZATION much larger than the rust thickness (d1 + d2 in Fig. 1(b)).
Various time-domain features, namely time to peak, peak When probe scans over the corroded area, both the corrosion
height and the more recently reported, rising point and zero and the steel can influence the PEC response.
crossing, extracted from the PEC response are used to identify Fig. 4 shows the typical PEC responses where the time is
and characterise defects in metallic targets [34], [35], [42]. In normalised to the repetition period T of the excitation. As
this section, two time domain features are extracted from PEC shown in Fig. 4(a), B R E F is the reference signal obtained
response and used to characterise the conductivity and perme- on non-corroded material; B is the total signal on the cor-
ability of the corroded area, which are extended from previous roded area. In order to increase the signal resolution, it is

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.
2116 IEEE SENSORS JOURNAL, VOL. 12, NO. 6, JUNE 2012

convenient to quantify the differential signal (&B) as shown

in equation (2)
&B = B − B R E F (2)
When testing ferrous materials, the maximum value of the
measured magnetic field can vary significantly due to the
magnetisation effect. Owing to the ferromagnetic properties
of the steel sample, the magnitude of the PEC signal is scaled
proportionally to the magnetic permeability of the steel. The
feature characterising the magnetic permeability variation is
Fig. 5. (a) Differential signal of total nonnormalized signal: ∂B/∂t. (b)
the maximum value of &B, named Max(&B) in this work. Differential signal of difference nonnormalized signal: ∂(&Bnorm)/∂t.
As shown in Fig. 4(b), it is possible to normalise the total
signal B and reference signal B R E F to their respective maxima TABLE II
and then to get the differential normalised signal by S EVEN F EATURES FOR C ORROSION ON U NCOATED S AMPLES 8 AND 2

&B norm = B/max(B) − B R E F /max(B R E F ) (3) Feature name sam 8 sam 2 & ε (%)
PV(&Bnorm) 0.0549 0.0259 0.029 52.82
Where B/max(B) is the normalised total signal ∂(&Bnorm)/∂t
andB R E F /max(B R E F ) is the normalised reference signal. 0.0016 6.70E-04 9.30E-04 58.11
rising
The peak value of differential normalised signal &Bnorm PV(∂B/∂t) 0.1672 0.1667 0.0005 0.30
is called PV(&Bnorm). T2P(&Bnorm) represents the time Max(&B) −0.7922 −0.3436 −0.4486 130.56
to PV(&Bnorm). Under such a convention, a positive pulse ∂(&Bnorm)/∂t
−4.40E-04 −1.99E-04 −2.41E-04 120.98
of &Bnorm due to rising edge of rectangular excitation falling
current means that time constant Tc of B decreases than T2P(∂B/∂t) 1.00E-05 2.00E-05 −1.00E-05 −50.00
that of B R E F , which can correspond to decreasing electrical T2P(&Bnorm) 1.28E-04 1.38E-04 −1.00E-05 −7.25
conductivity according to
Tc = L/R (4) and ∂(&Bnorm)/∂t falling. All of features are extracted from
the differential response (equation (2)), which is zero when the
Where, L is inductance and Ris resistance of the electric probe is on the steel. In feature optimisation, we did not use the
circuit including the EC probe and a test piece. In that, values of these features when probe is on the steel. Extremum
the positive pulse of &Bnorm will result in the increase of values of the seven features when the probe is scanned over
PV(&Bnorm) and decrease of T2P(&Bnorm), which means the the corrosion on uncoated samples 8 (6 months exposure) and
electrical conductivity of the corrosion decreases. Therefore, 2 (1 month exposure) are shown in Table II. Extremum values
the PV(&Bnorm) after normalisation and differential process of the features when the probe is scanned over the corrosion
can reflect electrical conductivity change of detected area of on coated samples 9 (6 months exposure) and 3 (1 month
material under test. Previous work on stress measurement exposure) are shown in Table III. In the tables, the specific
in aluminium alloys shows that the positive pulse of the sample number is simplified as ‘sam no’. In Table II, absolute
difference normalised signal is indicative of a decrease in variation & between samples 8 and 2 is calculated by
electrical conductivity [23], [41].
& = value of sam 8 − value of sam 2 (5)
B. Feature Comparison Relative variation ε (%) is calculated by
In the last sub-section, two time features are extracted and ε = (&/value of sam 8) × 100 (6)
linked to changes in the electrical conductivity and permeabil-
ity of the detected area. To evaluate and compare the sensitivity From Tables II and III, it is found that PV(&Bnorm),
of the features, other features are extracted and optimisation ∂(&Bnorm)/∂t rising, Max(&B), and ∂(&Bnorm)/∂t falling
is carried out through comparison studies. have good sensitivity in the seven features due to large values
Fig. 5(a) shows the differential signal of the total non- in relative variation, ε. Additionally, the measured values of
normalised signal ∂B/∂t, where two features PV(∂B/∂t) and PV(&Bnorm) and Max(&B) for the same sample are larger
T2P(∂(B)/∂t) are extracted. PV(∂B/∂t) represents the peak than ∂(&Bnorm)/∂t rising and ∂(&Bnorm)/∂t falling in the
value of ∂B/∂t and T2P(∂B/∂t) represents the time to peak order of magnitude.
value of ∂B/∂t. Fig. 5(b) shows the differential signal of
the difference normalised signal: ∂(&Bnorm)/∂t, where two V. C ORROSION C HARACTERIZATION
features ∂(&Bnorm)/∂t rising and ∂(&Bnorm)/∂t falling are Based on the feature extraction and comparison studies
extracted. ∂(&Bnorm)/∂t rising represents the maximum value in Section IV, it is concluded that Max(&B) feature can
of ∂(&Bnorm)/∂t and ∂(&Bnorm)/∂t falling represents the characterise permeability change, PV(&Bnorm) feature can
minimum value of ∂(&Bnorm)/∂t. characterise conductivity change, and both features have good
Now, we have seven features: Max(&B), PV(&Bnorm), sensitivity. In this section, they are used for corrosion charac-
T2P(&Bnorm), PV(∂B/∂t), T2P(∂B/∂t), ∂(&Bnorm)/∂t rising terisation, in particular monitoring corrosion over time.

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.
HE et al.: STEEL CORROSION CHARACTERIZATION USING PEC SYSTEMS 2117

TABLE III TABLE IV

S EVEN F EATURES FOR C ORROSION ON C OATED S AMPLES 9 AND 3 F ITTED VALUE OF A, n, AND B

Feature name sam 9 sam 3 & ε (%) Sample type Feature A n B

PV(&Bnorm) 0.0449 0.0191 0.0258 135.08 PV(&Bnorm) 0.0230 0.3989 0.0621
Uncoated corrosion
Max(&B) −0.2879 0.4695 −0.9209
∂(&Bnorm)/∂t
0.0013 5.40E-04 7.60E-04 140.76 PV(&Bnorm) 0.0192 0.4679 0.0615
rising Coated corrosion
Max(&B) −0.2588 0.5338 −0.9751
PV(∂B/∂t) 0.1675 0.1665 0.001 0.60
Max(&B) −0.7147 −0.2641 −0.4506 170.62
∂(&Bnorm)/∂t
−4.36E-04 −1.52E-04 −2.84E-04 186.85
falling
T2P(∂B/∂t) 8.00E-06 1.60E-05 −8.00E-06 −50.00
T2P(&Bnorm) 1.22E-04 1.28E-04 −6.00E-06 −4.69

Fig. 7. Measured values and fitted lines of Max(&B) for uncoated and coated
corrosion.

where, t y is the exposure time in years, C is PEC feature

after t, n are constant, which is the same value of n in (7),
B = 12n A is PEC feature in the first year (12 months in (7)).
Fig. 6. Measured values and fitted lines of PV(&Bnorm) for uncoated and
The value of B is shown in Table IV. The equations (7) and (8)
coated corrosion. can be used to predict the exposure time and PEC feature of
corrosion. The PV(&Bnorm) of uncoated and coated corrosion
with 12 months (1 year) exposure time are predicted as 0.0621
Measurements on coated and uncoated samples were per- and 0.0615.
formed using PEC system. The scan step size is 1 mm The measured values and fitted lines of Max(&B) of
and the size of the corroded area on all of the samples uncoated and coated samples are shown in Fig. 7. The red
is 30 mm×30 mm. For each corrosion area with different circles are measured values of uncoated corrosion and the
exposure time, 16 measurements are conducted under the same green squares in are measured values of coated corrosion.
excitation. The average value of 16 measurements is used The blue solid line and black double “−” line are fitted lines
as measured value. The measured values and fitted lines of for uncoated and coated corrosion, which are also fitted using
PV(&Bnorm) for uncoated and coated samples are shown the power function. The fitted A, n and B are also listed in
in Fig. 6. The red circles indicate the measured values of Table IV. Except the sign of A and B is negative, the power
uncoated corrosion and the green squares indicate measured function is also suitable for the fitting line of Max(&B).
values of coated corrosion. The blue solid line and black dou- As the coating layer on the coated samples is ∼100 µm
ble “-” line are fitted lines for uncoated and coated corrosion, thick, the effect on PEC characterisation using normalised
which are fitted according to the power function: and non-normalised differential features is not significant.
As illustrated in Fig. 6 and Fig. 7, the measured values of
C = At n (7)
uncoated and coated corrosion are close. In Fig. 6, there is a
where, t is the exposure time in months, C is PEC feature good agreement with the fitted lines between the uncoated
after t, A is that in the first month, and n are constant. and coated corrosion: PV(&Bnorm) has an increase with
As we mentioned in Section I, the power function is widely exposure time t. According to section IV, the increase in
used in long-term corrosion (t > 1 year) monitoring [7]–[9]. PV(&Bnorm) means the average electrical conductivity of the
From Fig. 6, it is concluded that the power function is also detected area decreases [23]. In Fig. 7, the fitted lines between
suitable for the early stage corrosion monitoring (t < 1 year). uncoated and coated corrosion also have a good agreement.
The fitted values of A and n are listed in Table IV. If we The feature Max(&B) has a monotonic decrease with exposure
change the unit of t from month to year, the function is time t. According to section IV, the decrease in Max(&B)
" #n means that average permeability of the detected area decrease.
C = A 12t y = 12n At yn = Bt yn (8) Because the skin depth of PEC excitation is larger than the

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.
2118 IEEE SENSORS JOURNAL, VOL. 12, NO. 6, JUNE 2012

Fig. 9. Photographs of corrosion with exposure time of (a) 3 months,

(b) 6 months, and (c) 10 months.

Fig. 9(c). This shows that later corrosion will spread rather
than increase in thickness. Many studies have shown that
atmospheric corrosion development in long-term exposure is
different from that in short-term exposure [7], in terms of
the increase in corrosion layer depth and corrosion area. As
Fig. 8. Measured values of corrosion heights versus explosure time and their the approach in this work concentrates on the characterisation
fitted line. of corrosion layer depths, further investigation will consider
the increase in corrosion area for comprehensive corrosion
prediction and monitoring.
rust thickness, the PEC responses are a combination of the
responses due to corrosion and steel. Therefore, the total
VI. C ONCLUSION
corrosion layer thickness will affect the PEC response. Over
time, an increase in the corrosion thickness means an increase In this paper, both uncoated and coated mild steel samples
of the contribution of corrosion to the PEC responses; this will with atmospheric corrosion (exposure times between 1 month
lead to a decrease of the average conductivity and permeability and 10 months) are measured using a PEC system and laser
of detected area. profilometry for corrosion heights. Through normalisation
Corrosion rate (PEC feature/year) can be derived from the and differential operations, PEC features are extracted and
first derivative of equation (8) compared to characterise the average electrical conductivity
and permeability changes of the corroded steel area. Corrosion
v = n Bt yn−1 (9) characterisation over time is studied and the results show
that:
It is noticed n is less than one in Table IV, which means a 1) Both electrical conductivity and permeability of the
larger decrease in corrosion rate as exposure time t increases atmospheric corrosion are less than mild steel. As expo-
[43]. As shown in Fig. 6 and 7, the slope of the fitted lines sure time increases, the decrease of average electrical
decreases over exposure time. This indicates the atmospheric conductivity and permeability means an increase of
corrosion in this work is a convergent process. Over time, corrosion thickness.
the rate of supply of oxygen to the corroding surface will 2) The power function is suitable for characterisation of
decline as the corrosion product layer builds up [44]. The early stage (t < 1 year) atmospheric corrosion, whose
corrosion rate will decrease and the rate of increase of the development is a convergent process. The corrosion rate
thickness of the corrosion layer (the slope of the fitted line and thickness increase ratio decrease over time.
in Fig. 8) will decrease. The values of corrosion height in 3) Based on the relationship between the PEC features and
Fig. 8 are measured using laser profilometry for verification, exposure time, the fitted line and formula can be used
which reflect the corrosion layer depths. The average value of for exposure time evaluation of coated and uncoated
corrosion heights for 1, 3, and 6 months exposed corrosion corrosion. Further monitoring and prediction of corrison
are 43.06, 77.65, 108.28µm, respectively, which has good progression will measure not only the corrosion layer
progression pattern with the results from PEC features in depths vs. conductivity and permability varation, but also
Figs. 6 and 7, which reflect the variation of conductivity and/or the spread of corrosion.
permeability.
Future work will include further investigation using PEC
As the exposure time increases, the area of corrosion will
for specific conductivity and permability variation and their
be increased. Figs. 9 (a)–(c) show corrosion after 3, 6 and
feature extraction and modelling for corrosion monitoring
10 months in exposure time, respectively. As the corrosion
under coating.
expands, there will be a gap between the plastic tape and
the steel. The oxygen and water can reach to the steel under
the tape then the rust will spread to these areas. As shown ACKNOWLEDGMENT
in Fig. 9 (b), after 6 months exposure, the corrosion begins The authors would like to thank the International Paint
to spread. In Fig. 9(c), the 10 months exposed corrosion Gateshead, U.K., for providing the experimental samples,
continues to spread to a larger area. In addition, the rust and X. Zuo from Mechanical Engineering College, Hebei
will become loose and flake off as it expands, as shown in Province, China, for assistance with the experimental studies

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.
HE et al.: STEEL CORROSION CHARACTERIZATION USING PEC SYSTEMS 2119

during his visiting research at Newcastle University. Mr. He [22] A. Nishikata, Y. Yamashita, H. Katayama, T. Tsuru, A. Usami, K.
would like to thank the China Scholarship Council for spon- Tanabe, and H. Mabuchi, “An electrochemical impedance study on
atmospheric corrosion of steels in a cyclic wet-dry condition,” Corrosion
soring his visiting study to Newcastle University, Ministry of Sci., vol. 37, no. 12, pp. 2059–2069, 1995.
Education for Scholarship Award for Excellent Doctoral Stu- [23] M. Morozov, G. Y. Tian, and P. J. Withers, “Noncontact evaluation of
dent (No.: JY20101), Hunan province for Graduate Scientific the dependency of electrical conductivity on stress for various Al alloys
as a function of plastic deformation and annealing,” J. Appl. Phys., vol.
Research Innovation Fund (No.: CX2010B010), and National 108, no. 2, pp. 024909-1–024909-9, 2010.
University of Defense Technology for Outstanding Graduate [24] G. Y. Tian, A. Sophian, D. Taylor, and J. Rudlin, “Multiple sensors
Innovation Fund (No.: YY201032). on pulsed eddy-current detection for 3-D subsurface crack assessment,”
IEEE Sensors J., vol. 5, no. 1, pp. 90–96, Feb. 2005.
[25] B. Lebrun, Y. Jayet, and J.-C. Baboux, “Pulsed eddy current signal
R EFERENCES analysis: Application to the experimental detection and characterization
of deep flaws in highly conductive materials,” NDT E Int., vol. 30, no.
[1] P. Kritzer, “Corrosion in high-temperature and supercritical water and 3, pp. 163–170, 1997.
aqueous solutions: A review,” J. Supercritical Fluids, vol. 29, nos. 1–2, [26] T. W. Krause, D. Harlley, V. K. Babbar, and K. Wannamaker, “Pulsed
pp. 1–29, 2004. eddy current thickness measurement of selective phase corrosion on
[2] R. Baboian, Corrosion Tests and Standards: Application and Interpre- nickel aluminum bronze valves,” in Proc. AIP Conf., 2010, pp. 401–
tation, 2nd ed. West Conshohocken, PA: ASTM International, 2005. 408.
[3] P. A. Schweitzer, Metallic Materials:Physical, Mechanical, and Corro- [27] Y. A. Plotnikov, W. J. Bantz, and J. P. Hansen, “Enhanced corrosion
sion Properties. New York: Marcel Dekker, 2003. detection in airframe structures using pulsed eddy current and advanced
[4] J. B. Johnson, P. Elliott, M. A. Winterbottom, and G. C. Wood, “Short processing,” Mater. Evaluat., vol. 65, no. 65, pp. 403–410, 2007.
term atmospheric corrosion of mild steel at two weather and pollution [28] M. A. Robers and R. Scottini, “Pulsed eddy current in corrosion
monitored sites,” Corrosion Sci., vol. 17, no. 8, pp. 691–700, 1977. detection,” presented at the 8th European Conference on Nondestructive
[5] Corrosion of Metals and Alloys-Corrosivity of Atmospheres- Testing Barcelona, 2002.
Classification, ISO Standard 9223:2012, 1992. [29] C. H. P. Wassink, M. A. Robers, J. A. Raad, and T. Bouma, “Con-
[6] S. Feliu, M. Morcillo, and S. Feliu, “The prediction of atmospheric dition monitoring of inaccessible piping,” presented at the 15th World
corrosion from meteorological and pollution parameters-II. Long-term Conference on Non-Destructive Testing Roma, 2000.
forecasts,” Corrosion Sci., vol. 34, no. 3, pp. 415–422, 1993. [30] R. A. Smith and G. R. Hugo, “Transient eddy current NDE for ageing
[7] W. Hou and C. Liang, “Atmospheric corrosion prediction of steels,” aircraft capabilities and limitations,” Insight: Non-Destruct. Test. Condit.
Corrosion, vol. 60, no. 3, pp. 313–322, 2004. Monit., vol. 43, no. 1, pp. 14–25, Jan. 2001.
[8] Y. Ma, Y. Li, and F. Wang, “The atmospheric corrosion kinetics of low [31] Y. He, F. Luo, and M. Pan, “Defect characterisation based on pulsed
carbon steel in a tropical marine environment,” Corrosion Sci., vol. 52, eddy current imaging technique,” Sensors Actuat. A: Phys., vol. 164,
no. 5, pp. 1796–1800, 2010. pp. 1–7, Sep. 2010.
[9] S. Feliu, M. Morcillo, and S. Feliu, “The prediction of atmospheric [32] R. A. Smith, D. Edgar, J. Skramstad, and J. Buckley, “Enhanced transient
corrosion from meteorological and pollution parameters-I. Annual cor- eddy current detection of deep corrosion,” Insight: Non-Destruct. Test.
rosion,” Corrosion Sci., vol. 34, no. 3, pp. 403–414, 1993. Condit. Monit., vol. 46, no. 2, pp. 88–91, 2004.
[10] G. Malumbela, M. Alexander, and P. Moyo, “Steel corrosion on RC [33] Y. Gotoh, H. Hirano, M. Nakano, K. Fujiwara, and N. Takahashi,
structures under sustained service loads a critical review,” Eng. Struct., “Electromagnetic nondestructive testing of rust region in steel,” IEEE
vol. 31, no. 11, pp. 2518–2525, 2009. Trans. Magnet., vol. 41, no. 10, pp. 3616–3618, Oct. 2005.
[11] A. Nishikata, Y. Ichihara, and T. Tsuru, “An application of electrochem- [34] Y. He, F. Luo, M. Pan, X. Hu, J. Gao, and B. Liu, “Defect classification
ical impedance spectroscopy to atmospheric corrosion study,” Corrosion based on rectangular pulsed eddy current sensor in different directions,”
Sci., vol. 37, no. 6, pp. 897–911, 1995. Sensors Actuat. A: Phys., vol. 157, no. 1, pp. 26–31, 2010.
[12] P. L. Bonora, F. Deflorian, and L. Fedrizzi, “Electrochemical impedance [35] G. Y. Tian and A. Sophian, “Defect classification using a new feature
spectroscopy as a tool for investigating underpaint corrosion,” Elec- for pulsed eddy current sensors,” NDT E Int., vol. 38, no. 1, pp. 77–82,
trochim. Acta, vol. 41, nos. 7–8, pp. 1073–1082, 1996. 2005.
[13] J. H. Wang, F. I. Wei, Y. S. Chang, and H. C. Shih, “The corrosion [36] Y. He, M. Pan, and F. Luo, “Reduction of lift-off effects in pulsed eddy
mechanisms of carbon steel and weathering steel in SO2 polluted current for defect classification,” IEEE Magnet. Trans., vol. 47, no. 12,
atmospheres,” Mater. Chem. Phys., vol. 47, no. 1, pp. 1–8, 1997. pp. 4753–4760, Dec. 2011.
[14] Y. Ma, Y. Li, and F. Wang, “Corrosion of low carbon steel in atmospheric [37] N. Canter, “Tech beat: Conductivity in minerals,” Tribology Lubricat.
environments of different chloride content,” Corrosion Sci., vol. 51, no. Technol., vol. 64, no. 7, pp. 14–15, 2008.
5, pp. 997–1006, 2009. [38] N. Guskos, G. J. Papadopoulos, V. Likodimos, S. Patapis, D. Yarmis,
[15] G. Salvago, G. Bollini, and P. L. Cavallotti, “Electrochemical evaluation A. Przepiera, K. Przepiera, J. Majszczyk, J. Typek, M. Wabia, K.
of localized corrosion probability in stainless steel,” Mater. Sci. Forum, Aidinis, and Z. Drazek, “Photoacoustic, EPR and electrical conductivity
vols. 289–292, no. 2, pp. 933–953, 1998. investigations of three synthetic mineral pigments: Hematite, goethite
[16] M. Fregonese, H. Idrissi, H. Mazille, L. Renaud, and Y. Cetre, “Initiation and magnetite,” Mater. Res. Bulletin, vol. 37, no. 6, pp. 1051–1061,
and propagation steps in pitting corrosion of austenitic stainless steels: 2002.
Monitoring by acoustic emission,” Corrosion Sci., vol. 43, no. 4, pp. [39] N. N. Greenwood and A. Earnshaw, Chemistry of the Elements. New
627–641, 2001. York: Oxford, 1997.
[17] H. Mazille, R. Rothea, and C. Tronel, “An acoustic emission technique [40] D. Fuente, I. Díaz, J. Simancas, B. Chico, and M. Morcillo, “Long-term
for monitoring pitting corrosion of austenitic stainless steels,” Corrosion atmospheric corrosion of mild steel,” Corrosion Sci., vol. 53, no. 2, pp.
Sci., vol. 37, no. 9, pp. 1365–1375, 1995. 604–617, 2011.
[18] G. A. Matzkanin and H. T. Yolken, “Detecting hidden corrosion,” Pract. [41] M. Morozov, G. Y. Tian, and P. J. Withers, “The pulsed eddy current
Oil Anal., vol. 10, pp. 1–42, Aug. 2008. response to applied loading of various aluminium alloys,” NDT E Int.,
[19] J. K. Paik, “Condition assessment of aged ships,” presented at the vol. 43, no. 6, pp. 493–500, 2010.
16th International Ship and Offshore Structures Congress, Southampton, [42] D. Zhou, G. Y. Tian, B. Zhang, M. Morozov, and H. Wang, “Optimal
U.K., 2006. features combination for pulsed eddy current NDT,” Nondestruct. Test.
[20] M. E. Davoust, L. Le Brusquet, and G. Fleury, “Robust estimation Evaluat., vol. 25, no. 2, pp. 133–143, 2010.
of hidden corrosion parameters using an eddy current technique,” J. [43] H. E. Townsend, “Effects of alloying elements on the corrosion of
Nondestruct. Evaluat., vol. 29, no. 3, pp. 155–167, 2010. steel in industrial atmospheres,” Corrosion, vol. 57, no. 6, pp. 497–501,
[21] H. A. Sodano, “Development of an automated eddy current structural 2001.
health monitoring technique with an extended sensing region for cor- [44] R. E. Melchers, “Transition from marine immersion to coastal
rosion detection,” Struct. Health Monit., vol. 6, no. 2, pp. 111–119, atmospheric corrosion for structural steels,” Corrosion, vol. 63, no. 6,
2007. pp. 500–514, 2007.

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.
2120 IEEE SENSORS JOURNAL, VOL. 12, NO. 6, JUNE 2012

Yunze He (S’11) received the Bachelors degree from Xi’an Jiaotong Mohammed Alamin (S’11) received the B.E. degree in electrical and
University, Xi’an, China, and the Masters degree in instrument science and electronic engineering and the M.Sc. degree in communications and signal
technology from the National University of Defense Technology, Changsha, processing from Newcastle University, Newcastle upon Tyne, U.K., in 2008
China, in 2006 and 2008, respectively. He is currently pursuing the joint Ph.D. and 2009, respectively. He is currently pursuing the Ph.D. degree with
degree with the School of Electrical, Electronic and Computer Engineering, the School of Electrical, Electronic and Computer Engineering, Newcastle
Newcastle University, Newcastle upon Tyne, U.K., and the College of Mecha- University.
tronics and Automation, National University of Defense Technology. His current research interests include electromagnetic sensing, signal
His current research interests include electromagnetic nondestructive test- processing and analysis, and nondestructive testing.
ing, digital signal processing, and instrumental science and technology.

Guiyun Tian (SM’03) received the B.Sc. degree in metrology and instrumen- Anthony Simm (S’01–M’00) received the B.Sc. degree in computing science
tation and the M.Sc. degree in precision engineering from the University of from Sunderland University, Sunderland, U.K., the M.Sc. degree in process
Sichuan, Chengdu, China, in 1985 and 1988, respectively. analytics from Newcastle University, Newcastle upon Tyne, U.K., and the
He is currently a Chair Professor of sensor technologies with the School M.Phil. degree in multiobjective optimization of industrial control systems
of Electrical, Electronic and Computer Engineering, Newcastle University, from Nottingham University, Nottingham, U.K., in 1996, 2001, and 2004,
Newcastle upon Tyne, U.K., and the Group Head of Communications and respectively. He is currently pursuing the Ph.D. degree with Newcastle Uni-
Sensor Systems. He has a joint background in engineering and computer versity, where his research focuses on the areas of theoretical and experimental
science. He has published over 150 books and papers in English and Chinese eddy current nondestructive testing (NDT).
in the above areas. His current research interests include the areas of He is currently a Research Assistant with the School of Electrical,
sensors and instrumentation, signal processing, electromagnetic nondestructive Electronic and Computer Engineering, Newcastle University, where he is
evaluation, and structural health monitoring in particular. investigating the use of electromagnetic NDT methods for monitoring the
Prof. Tian is a regular reviewer for international journals and conferences. performance of coatings.

Hong Zhang (S’11) received the Bachelors degree from the joint program Paul Jackson received the B.Sc. degree in chemistry and the Ph.D. degree
between Nanjing Normal University, Nanjing, China, and Northumbria Uni- in chemical physics from Nottingham University, Nottingham, U.K., in 1981
versity, Newcastle, U.K., in 2009, and the Masters degree in communication and 1985, respectively.
and signal processing from Newcastle University, Newcastle upon Tyne, U.K., He is currently a Principal Research Technologist with International Paint,
in 2010. He is currently pursuing the Ph.D. degree with the School of Gateshead, U.K., part of the Akzo Nobel Group, where he leads several of the
Electrical, Electronic and Computer Engineering, Newcastle University. company’s longer term research initiatives focused on coatings for the marine
His current research interests include electromagnetic nondestructive test- and protective coating market sectors. His current research interests include
ing, digital signal processing, and communication science and technology. the areas of corrosion protection, materials science, and effect coatings.

Authorized licensed use limited to: b-on: UNIVERSIDADE DO MINHO. Downloaded on October 16,2023 at 16:12:21 UTC from IEEE Xplore. Restrictions apply.