Construction and Building Materials 152 (2017) 777–793

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier.com/locate/conbuildmat

Predicting the residual strength and deformability of corroded steel plate

based on the corrosion morphology
Youde Wang ⇑, Shanhua Xu ⇑, Hao Wang, Anbang Li
School of Civil Engineering, Xi’an University of Architecture and Technology, Xi’an 710055, China

h i g h l i g h t s

Conducted 3D profile measurement and tensile test to evaluate the corrosion effects.

Quantitatively characterized the corrosion profile and developed a dedicated program.

Summarized the empirical formulas used to predict residual performance.

Established a morphology based numerical method to predict residual performance and fracture.

Surface strain of corroded specimen was measured by using DIC technology.

a r t i c l e i n f o a b s t r a c t

Article history: The effects of corrosion on the mechanical performance of steel were investigated by conducting surface
Received 2 March 2017 measurements and tensile tests on naturally corroded steel plates. With increasing corrosion damage,
Received in revised form 10 June 2017 significant degradations in strength and deformation can be observed. We summarized the existing
Accepted 4 July 2017
empirical formulas and elaborated a morphology based numerical method to predict the residual perfor-
Available online 12 July 2017
mance of corroded steel. With the aid of specially developed programs, it is easy to extract the pertinent
evaluation indicators from the corrosion morphology and generate the finite element model of corroded
Keywords:
surfaces. By conducting error analysis of predicted results, we found that most empirical formulas could
Steel plate
Corrosion
achieve satisfying predictions in tensile strength but were unworkable in predicting the deformability.
Morphology The profile based numerical method could not only achieve more reliable predictions in strength but also
Tensile test provide the degraded load-displacement curve of corroded steel plates.
Performance prediction Ó 2017 Elsevier Ltd. All rights reserved.
Numerical simulation

1. Introduction The plate thickness and mechanical properties of structural

Steel structures including transmission towers, steel bridges,
workshops, and offshore platforms exposed to harsh environment
such as urban, industrial and offshore atmosphere for a long time
often show signs of apparently severe surface corrosion, when
the protective coatings and cathodic protection systems either
are non-existent or have become ineffective [1,2]. Corrosion has
been considered to be the predominate factor leading to age-
related structural degradation of steel structures due to the
harmful consequences including strength degradation, thickness
penetration, fatigue cracks, brittle fracture and unstable failure
[3]. These forms of damage can give rise to significant issues in
terms of safety, health and environment.
⇑ Corresponding authors.
E-mail addresses: yord.w@sohu.com, yord_w@sina.com (Y. Wang), xushanhua@
163.com (S. Xu), wh19881218@126.com (H. Wang), lianbangjdtm@163.com (A. Li).
steel such as yield strength (fyi), ultimate strength (fu), elongation
at failure (d), as well as yield ratio (fyi/fu) are important in struc-
tural design according to the present standards. However, during
the actual service, these properties are considered to be constant,
neglecting the corrosion-induced deteriorations that the steel
structures may suffer. Although the design of steel structures are
required to contain a certain allowance for corrosion wastage
and the safety factor in initial design may delay the occurrence
of structural failure, the priori negligence of corrosion in structural
integrity analysis may lead to significant overestimation of the
damage tolerance ability of existing steel structures [4,5].
Conventionally, the residual bearing capacity of corroded com-
ponent is assessed using the required effective thickness, aiming to
meet the minimum strength requirement. This method is
convenient and simple to apply. However, up to now, this method
has not been developed systematically and the associated evalua-
tion indicators are still less than satisfactory, for instance, the
determination of effective thickness. Moreover, how to determine

http://dx.doi.org/10.1016/j.conbuildmat.2017.07.035
0950-0618/Ó 2017 Elsevier Ltd. All rights reserved.
778 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793
the influence of localized corrosion using the required effective
thickness approach is still full of controversy [5], because corrosion
discontinuities (pits) can lead to stress/strain concentration [1,6–9]
and even premature fracture.
The method of introducing degradation mechanism into mate-
rial constitutive relation may be a near-perfect approach to con-
sider the corrosion effects, which can achieve a good balance in
predicting precision and computational efficiency. The standard
tensile test provides an intuitive and convenient way to evaluate
the influence of corrosion on the mechanical properties of steel
plate and have been successfully conducted in a laboratory setting
[10–15]. However, for existing structures, this approach is destruc-
tive in nature. For this reason, many empirical formulas used to
predict the residual performance of corroded steel were proposed
based on the experimental results.
A good prediction of the corrosion effects by using the empirical
formula is largely dependent on the quantification of corrosion
itself. Recent years, several physical probe techniques (SEM [16],
AFM [6], X-ray tomography [17–19], white light interferometry
(WLI) [20], white light axial chromatism [21,22], etc.) have been
proposed and applied to characterize the corrosion features and
quantify the corrosion level. All these investigations promote the
innovation of corrosion measurements and also provide a new
way for non-destructive evaluation (NDE). Extensive literatures
and standards [21,23–28] have presented numerous parameters
to characterize the corrosion morphology and quantify its severity.
Many empirical formulas based on the morphological parameters
were also proposed by researchers to estimate the degradation of
mechanical properties. However, these formulas were established
by making some simplifications, and their application conditions
were not described in detail by researchers. Therefore, the practi-
cality have not been widely verified and recognized.
More recently, due to improved computational capabilities, the
numerical method based on corrosion morphology has gained
greater emphasis. Initially, researchers typically focused on the
simplification of corrosion morphology [29–32] and forming the
surface degradation by utilizing the empirical corrosion models
[3,13,33–35]. A disadvantage of this simplification is that the pre-
dicted mechanical properties are dependent on how the actual cor-
rosion is interpreted or the corrosion model behavior [5]. After
that, the direct scanning [1,6,9,10,22,28] from a corroded surface
was also applied in modeling. Pidaparti [1,6] and Kainuma [9]
developed a numerical method based on the scanned corrosion
morphology to evaluate the stress concentration level around pits.
A similar method was also employed in our previous work [22] to
predict the possible nucleation site of pitting induced crack and
then predict the fatigue life of corroded steel plate. Appuhamy
[10] discussed the feasibility of establishing a numerical methodol-
ogy to preliminary predict the residual strength of corroded steel
plates using fewer number of scanning points. In the work of Ahm-
mad [13], the quasi-static tensile tests and numerical simulations
were both conducted on replica specimens.
The corrosion morphology is identified as the most intuitive
data because it can comprehensively reflect the corrosion levels
without any simplification or quantification. Although a few liter-
atures have utilized the corrosion morphology to evaluate the
effects of corrosion on material properties, it has not been paid
enough attention in the field of NDE of existing corroded steel
structures. There is almost no literature involved the prediction
of loading and deformation behavior of corroded steel plates dur-
ing the whole process of stretching, especially for the large defor-
mation behavior after necking. In addition, most conclusions,
methods and empirical formulas in previous studies were obtained
or proposed under artificial or specific corrosion conditions, their
applicability still needed to be verified. In the present study, we
summarized the existing empirical formulas and elaborated a cor-
rosion morphology based numerical method to predict the residual
performance of corroded steel. Their effectiveness and applicability
were also investigated by conducting a series of experiments
including corrosion morphology measurement and characteriza-
tion, uniaxial tension and surface strain measurement on the nat-
urally corroded steel plates. Special programs were also developed
to extract the morphological parameters and generate the finite
element (FE) models based on the surface data. The research
results would provide technical support for the residual perfor-
mance evaluation of existing corroded steel structures.
2. Experimental procedure
2.1. Material and specimen preparation
The test specimens were all cut out from an H-beam steel truss (shown in
Fig. 1a) with 8 years nature expose at Xi’an (a city in western China). The appear-
ance after removing the rust is shown in Fig. 1d. The tested material was a normal-
ized 0.2% low-carbon steel (named Q235 in China), which is widely used in
infrastructure and civil engineering. Its chemical compositions are presented in
Table 1. According to the different position and obliquity (see Fig. 1a), the truncated
components were categorized into three general types: horizontal (H), vertical (V)
and sloping (S) members. One specimen was cut from the severely rusted top
flanges (TF), bottom flanges (BF) and webs (W) of each selected member. In
addition, one corrosion-free specimen (NC) was fabricated to act as a control and
determine the material properties of flanges and webs by removing the corrosion
features using a milling machine.
2.2. 3D corrosion profile measurements
The dimensions of specimen for profile measurement are shown in Fig. 2a.
Before conducting the surface measurements, corrosion products were removed
carefully by using the electric wire brushes. Note that the two corroded surfaces
of each steel plate were scanned separately, therefore the relative position of the
two surfaces need to be determined before conducting the quantitative character-
ization and establishing the finite element (FE) model. To solve this problem, a
reference zone with the width of 7.5 mm shown in Fig. 2a was prefabricated at
the edge of specimen using a milling machine.
A PS50 non-contact 3D profiler produced by NANOVER was employed to obtain
the 3D morphology of corroded surface. The vertical and horizontal resolutions
were 280 nm and 8 lm, respectively. Each corroded surface had one measurement
region of 32.5 mm  50 mm, which covered the whole parallel part of tensile test
specimen and the reference plane. The equipment and the schematic of measure-
ment area are shown in Fig. 2. The longitudinal and transversal scanning steps (S)
were both 50 lm. Thus, a regular grid with 651  1001 points was generated by
this arrangement. After the profile tests, the thickness of reference zone (Tsa) was
measured with an ultrasonic thickness gauge or vernier caliper.
2.3. Tensile test procedure
The geometry of dog-bone specimens (see Fig. 2c) for standard tensile test was
designed based on GB/T 228.1–2010 [36]. The distance between the two black
markers in Fig. 2c denotes the initial gauge length. All specimens should be
machined with no residual stress, surface damage, and deformation.
All specimens were tested according to GB/T 228.1–2010 [36] at room temper-
ature in laboratory air. The uniaxial tension was performed by an Instron Model
1341 electro hydraulic test frame with a 200 kN load cell. The temperature and
humidity were 25 ± 3 °C and 65 ± 5%, respectively. In each test, the coupon was fully
clamped at both ends and stretched up to fracture under strain control at a constant
strain rate of 2.5  104/s. An extensometer was utilized to track the elongation of
parallel part in real time and control the crosshead speed.
2.4. Surface strain measurement
A stereo digital image correlation (SDIC) system produced by XTOP was
employed here to measure the in-plane displacements at the specimen surface. In
this system, two 2448  2048 pixels CCD cameras and a REGER-100 electromechan-
ical materials testing machine were used. Before the measurements, a fine speckle
pattern was required on the area of interest (AOI) where the full-field displace-
ments were measured (see Fig. 3b). The initial size of AOI was 25 mm (width) 
50 mm (initial gauge length). During the tensile test, a series of digital images
corresponding to AOI were firstly captured and then processed by SDIC analysis
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793 779

Fig. 1. The corrosion state of tested H-beam steel truss.

Table 1
Chemical compositions of Q235 steel (wt%).

Fe C Si Mn P S Cr Ni Cu N
96.8 0.2 0.35 1.4 0.045 0.045 0.3 0.3 0.3 0.008

Fig. 2. (a) Specimen for surface measurement, (b) equipment for surface measurement and (c) specimen for uniaxial tensile test.
780 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793

Fig. 3. Experimental implementation of the SDIC system for surface strain measurement.

program to generate the local distribution of in-plane strain components. In this
paper, a NC and a SW specimens were used to measure the surface strain fields,
respectively.
3. Experimental results and discussion
3.1. Morphology characterization and parameter extraction
Fig. 4 clearly shows the corrosion morphology of both surfaces
of all test specimens. It can be seen the corrosion form may be
described as a rather non-uniform general corrosion, and the
effects of pitting and general corrosion are combined. Before con-
ducting morphology analysis, the centimeter scale waves were fil-
tered out to eliminate the effect of noises. It should be noted
especially that the original surface was not preserved during the
atmospheric exposure. Therefore, the corrosion depth in this sec-
tion refers in particular to the perpendicular distance between
the vertex and each scanning point (see Fig. 7a). Table 2 lists the
maximum (Dmax) and average (Dave) corrosion depth of each
specimen. Apparently, varying corrosion morphologies appear on

Fig. 4. Corrosion morphology of the both surfaces of all test specimens (A and B represent the sky-ward and ground-ward surface in H and S members).
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793 781

Fig. 5. (a) Principle diagram of CroEva for extracting the pitting features, and the extracted results of (b) HTF-A, (c) VW-A and (d) SW-B.

Fig. 6. Aspect ratio (R/h) of pits as a function of pit depth (h) in all specimens.
782 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793

Fig. 7. Procedures for extracting the cross-sectional parameters and volume loss ratio.

Table 2
Results summary for corrosion depth and pitting shapes of all corroded specimens.

Member Specimens Dmax/um Dave/um Pd(pits/cm2) have/um Rave/um (R/h)ave

(A/B)
H HTF 331/1253 113/515 20.2/6.4 169/713 1293/2249 7.3/3.3
HBF 945/1106 361/399 6.9/7.5 508/536 2231/1976 4.0/3.2
HW 868/692 286/298 9.4/11.5 413/391 1772/1690 3.9/3.9
S STF 437/1282 134/466 15.7/7.6 202/586 1345/1981 6.2/3.2
SBF 737/910 228/341 13.4/10.7 309/473 1499/1863 4.9/3.9
SW 888/1097 247/385 8.6/6.8 375/576 2183/2272 5.3/3.6
V VTF 466/751 152/192 15.3/14.6 269/301 1337/1314 4.8/4.3
VBF 810/730 279/171 11.0/13.3 433/282 1552/1410 3.4/4.8
VW 593/487 163/179 12.2/14.5 319/293 1679/1473 5.3/5.1

different specimens. Great majority of large corrosion pits dis-
tribute on the bottom flanges and webs of H and S members, par-
ticularly the ground-ward surface. The evolution of corrosion in
these specimens can be partly ascribed to the extension of wetness
time induced by gravity and accumulation of raindrops. In the
work of [9,11], the same phenomenon was also observed.
Some studies [22,23,27,28,37–41] exhibited the statistical anal-
ysis associated with pitting corrosion. In this paper, the pit density
and shapes including depth (h), diameter (R) and aspect ratio (R/h)
are also investigated. From Fig. 4, it can be seen the corrosion-
induced pits are distinctly separated from the matrix and the
geometry is essentially irregular. Consequently, a special program
named CorEva written in Matlab code was developed by us to
address the particular requirements of parameters acquisition. This
program created an image slice-by-slice in height, and then
employed a Floodfill algorithm to extract the manually picked con-
nected regions (i.e., pits) as shown in Fig. 5a and calculate the asso-
ciated shape parameters. With lowing the slice plane, all connected
regions (pits) would be extracted, as shown in Fig. 5b–d. Note that
the average value of pitting length and width was employed as the
diameter R, since the pits were extremely irregular. The computed
results are summarized in Table 2. Examinations of HTF-A and
VW-B surfaces in Fig. 5b and c reveal a large number of narrow
pits, while on the surface of SW-B (see Fig. 5d), wider pits tend
to appear. Another phenomenon is that crowded pits generally dis-
tribute on the surface of V member, while the pit density (Pd) is
much smaller on the H and S members, except for the sky-ward
surface of HTF and STF. The reduced density and wider pits are
supposed to be related to the link up of adjacent pits during atmo-
spheric exposure. As for the pitting shapes, the variation of aspect
ratio (R/h) with pit depth (h) in different specimens is discussed
here and depicted in Fig. 6. It can be seen that, the aspect ratios
of pits in almost all measured surfaces are in the range of 0.5–10,
except HTF-A and STF-A that have the majority of aspect ratios
in the range of 0.5–18. That is to say the corrosion pits on the
sky-ward surface of HTF and STF tend to be ‘‘sharper”.
Many researchers confirmed that the tensile strength of cor-
roded steel plate was governed by the smallest cross-sectional area
[11,13,14] or volume loss ratio [15]. Fig. 7 illustrates the procedure
for extracting these associated parameters. When the thickness of
reference zone (Tsa) has been obtained, the vertical distance
between the vertexes (Tmax) of the two corroded surfaces can be
determined by:
T max ¼ Dre re
A þ DB þ T sa ; ð1Þ
where Dre re
A and DB are the relative depth of reference plane. Note
that the vertexes on the both surfaces generally do not have the
same (x, y) coordinates, as shown in Fig. 7b. Since the reference
plane was no longer useful for the subsequent analysis, only the
corroded surface with a region of 25 mm  50 mm which was cor-
responding to the parallel part, was extracted from the original
data. Then the Z coordinate of each point on the two surfaces
(ZA and ZB) can be easily determined from the measured corrosion
depth:
Z A ðx; yÞ ¼ T max  DA ðx; yÞ ð2Þ
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793 783

Z B ðx; yÞ ¼ DB ðx; yÞ
The regular grid generated by the extracted points was then uti-
lized to discretize the 3D entity bounded by the upper and lower
surfaces. Then the cross sectional area can be calculated by adding
up the areas of all squares at this section (see Fig. 7b):
X
M
SðZ A ðxi ; yÞ  Z B ðxi ; yÞ þ Z A ðxiþ1 ; yÞ  Z B ðxiþ1 ; yÞÞ
AðyÞ ¼
i¼1
2
where S is the scanning step and M is the number of squares.
Similarly, the volume of corroded steel plate can be described as
the sum of the volume of all columns generated by grid points
(see Fig. 7c). Then the volume loss ratio can be expressed as:
q ¼ qn þ qu ;
ðT 0  T max Þ
qu ¼ ; ð6Þ
T0
X
MN
TðjÞ  s2
j¼1
qn ¼ 1   qu :
W  L  T0
where T0 is the initial thickness of un-corroded steel plate,
MN is the total number of columns contained in the grid
(M  N = 500  1000), and T(j) is the average height of four vertical
sides of the jth column. qu and qn refer to the uniform and non-
uniform volume loss ratio, respectively.
The developed program CorEva was also employed here to
implement these procedures and the results are summarized in
Table 3. From Table 3 it can be seen that, the uniform thickness loss
(tu) of H and S members fluctuates around a constant value (about
1.1 mm), while the value of V members is significantly lower
(about 0.6). This indicates that the member position may play a
decisive role in uniform volume loss. On the other hand, the non-
uniform thickness loss (tn) exhibits high discreteness, which
demonstrates that the non-uniform volume loss is more random
and sensitive to the external environment. Moreover, greater value
of tn tends to appear on the bottom flanges and webs of H and S
members. Here, it is also likely the results of the accumulation of
raindrops. In order to preliminary assess the severity of corrosion,
three types of corrosion levels are categorized here in terms of the
minimum thickness ratio l [10]: 1) minor corrosion, l > 0.75; 2)
moderate corrosion, 0.75  l  0.5; 3) severe corrosion, l < 0.5. l
is defined as Tmin/T0 and the corrosion levels are listed in Table 3.
It can be seen most specimens present moderate corrosion after
8 years of exposure, except for VTF (minor corrosion) and SW
(severe corrosion). The minimum cross-sectional area (Amin) and
ð3Þ volume loss ratio (q) are also summarized in Table 3. Since the
webs are mainly used to resist shear force and generally designed
to be much thinner than flanges, relatively higher volume loss ratio
(q) can be observed. That is to say, in assessing the residual perfor-
mance of corroded steel structures, the corroded webs can to some
extent be determined as the ‘‘weak links”, especially for the webs
of horizontal components. The above results indicate that various
ð4Þ evaluation indicators can be conveniently extracted from the cor-
rosion morphology. These parameters will also provide supporting
data for Section 4 to predict the residual properties of tested plates.
3.2. Tensile test results and discussion
The tensile test results of all specimens are presented in Table 4
ð5Þ
and Fig. 8. Note that the strength (yield strength fyi and ultimate
strength fu) in Table 4 is a nominal value, which is defined as the
ratio of applied load to nominal area. To investigate the effects of
non-uniform volume loss, the maximum residual thickness (Tmax)
is employed here as the nominal thickness. This figure clearly illus-
trates that the loss of bearing capacity is directly proportional to
the volume loss ratio. A reduction of ultimate load of about
ð7Þ 18.7% and 27.9% is respectively observed for 9.0 mm flanges and
6.5 mm webs. The ultimate load of the corroded webs of HW and
SW has reduced to a value comparable to the yield load of non-
corroded webs.
The reduction of bearing force can be divided into two parts
which are induced by uniform and non-uniform volume loss (as
shown in Fig. 7), and their proportions are both listed in Table 4.
It can be seen, the uniform volume loss usually plays a main part
in the load degradation, but for the specimens with higher propor-
tion of pitting corrosion (V members), we find the exact opposite.
The reduction of bearing capacity resulting from uniform volume
loss can be assessed based on the measured residual thickness,
however, things are quite different for localized volume loss. The
degradation induced by localized corrosion could be depicted indi-
rectly by the reduction of strength or the non-uniform volume loss
ratio. Specimens with larger non-uniform volume loss ratio gener-
ally present lower yield and ultimate strength. For the specimens
with moderate corrosion, yield strength and ultimate strength
are reduced by 8.2–14.3% and 3.2–11.7%, respectively. The
decreased yield strength also results in the expected decrease of
elastic modulus (Es) and yield ratio (fyi/fu).
On the aspect of deformation, a shorter yield plateau accompa-
nied with a range of angle can be observed for corroded steel in
Fig. 8, and eventually the well-defined yield plateau even
disappears for specimens with severe corrosion (SW specimen),
which is consistent with the conclusions of previous works in
[10,12,13]. This is logically because the cross-sectional area varies

Table 3
Results summary for the cross-sectional parameters and volume loss ratio.

Specimens T0 Tmax Tmin Tma Tave Tsd tu tn l Corrosion level Amin qu qn q

2
(mm) (mm ) (%)
NC 9 4.72 4.72 – 4.72 – – – – – 118.0 – – –
HTF 9 7.98 6.40 7.22 7.35 60 1.02 1.58 0.711 Mod 180.4 11.3 9.5 20.8
HBF 9 7.93 5.88 6.95 7.17 97 1.07 2.05 0.653 Mod 173.7 11.9 11.9 23.8
HW 6.5 5.28 3.72 4.61 4.70 63 1.22 1.56 0.572 Mod 115.3 18.8 10.5 29.3
STF 9 7.86 6.14 7.11 7.26 64 1.14 1.72 0.682 Mod 177.7 12.7 10.5 23.2
SBF 9 7.80 6.15 7.09 7.24 52 1.20 1.65 0.683 Mod 177.2 13.3 8.9 22.2
SW 6.5 5.27 3.29 4.49 4.64 72 1.23 1.98 0.498 Sev 112.1 18.9 14.7 33.6
VTF 8 7.36 6.14 6.93 7.02 33 0.64 1.22 0.768 Min 173.4 8.0 8.9 16.9
VBF 8 7.30 5.76 6.74 6.85 43 0.70 1.54 0.720 Mod 168.5 8.7 10.5 19.2
VW 6 5.42 4.24 4.97 5.08 33 0.58 1.18 0.707 Mod 124.3 9.7 9.8 19.5

T0–the original thickness of intact plate; Tmax–maximum residual thickness (i.e., the vertical distance between the vertexes); Tmin–the minimum thickness: Tmin =
Tmax  (DAmax + DBmax); Tma–the average thickness of the minimum cross-section, defined by Amin/W; Tave–the average thickness; Tsd–the standard deviation of thickness; tu–the
uniform thickness loss; tn–the non-uniform thickness loss: tn = DAmax + DBmax; Amin–the minimum cross sectional area; Mod–Moderate, Min–Minor, Sev–Severe.
784 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793

Table 4
Results summary for all tensile tests.

Specimens Tmax (mm) q qn Fyi Puyi/Pnyi Fui/Fu Puu/Pnu fyi fu eu De1 d Es (GPa) fyi/fu
(%) (kN) (MPa) (%)
NC 4.72 – – 34.5 – 51.3 – 292.6 435.0 20.0 2.16 35.2 207.0 0.673
NC-F 9.0 – – 65.8 – 97.8 – 292.6 435.0 20.0 2.16 35.2 207.0 0.673
NC-W 6.5 – – 47.6 – 70.6 – 292.6 435.0 20.0 2.16 35.2 207.0 0.673
HTF 7.98 20.8 9.5 53.5 7.5/4.8 80.3 11.1/6.4 268.2 402.7 17.5 1.97 31.7 194.6 0.666
HBF 7.93 23.8 11.9 52.3 7.8/5.7 79.2 11.6/7.0 263.8 399.7 18.5 1.45 31.3 184.5 0.660
HW 5.28 29.3 10.5 33.1 8.2/6.2 50.8 13.2/6.6 250.7 384.5 17.2 1.45 24.1 189.7 0.652
STF 7.86 23.2 10.5 53.1 8.3/4.4 79.3 12.4/6.1 267.0 403.6 18.7 1.66 31.5 200.3 0.669
SBF 7.80 22.2 8.9 52.3 8.8/4.7 79.4 13.0/5.4 268.1 407.1 18.9 1.72 32.4 198.9 0.659
SW 5.27 33.6 14.7 31.9 8.3/7.3 51.1 13.3/6.2 242.4 388.2 17.2 1.39 21.6 187.3 0.624
VTF 7.36 16.9 8.9 48.7 4.7/5.1 76.5 7.0/3.5 264.8 415.6 18.3 1.41 32.3 203.9 0.637
VBF 7.30 19.2 10.5 48.3 5.1/5.1 75.2 7.6/4.1 264.5 412 19.3 1.46 31.8 200.7 0.642
VW 5.42 19.5 9.8 36.8 3.9/3.6 55.9 6.2/2.2 271.6 412.5 19.1 1.38 30.8 195.4 0.654

Puyi and Pnyi refer to the degradation of yield load induced by uniform and non-uniform volume loss. Puu and Pnu refer to the degradation of ultimate load induced by uniform and
non-uniform volume loss.

Fig. 8. Experimental load-displacement curves (F–DL) of all test specimens. (Note that the load-displacement curves of un-corroded specimens for comparison are plotted by
amplifying the load of NC coupons based on the ratio of original (T0) to actual thickness.)

along the specimens due to randomness in the corrosion attack
[10]. Since uniform volume loss does not give rise to discontinu-
ities, this degradation should be attributed to non-uniform volume
loss. From Table 4, we can find that the yield plateau after corro-
sion is shortened by 9–37%. Also shown in that Figure and Table,
the elongations d at fracture of 21.6–32.4% are obtained for rusted
coupons, which are significant less than that of non-corroded steel
(34.9–35.2%). The loss of ductility mostly can be attributed to the
premature necking and a shorter descending branch of the load-
displacement curve past the necking point. The inset picture of
Fig. 8 also reveals that corrosion has a more significant effect on
the material properties after necking (at high strain levels). Take
SW specimen for instance, a brittle fracture occurs just after
exceeding the ultimate load, accordingly, the breaking elongation
decreases by up to 38% but the breaking load is still higher than
85% of the ultimate load. This fully indicates that the ductility of
corroded steel is not sufficiently developed after necking, as is evi-
dent from the fractures of corroded steel plates with less obvious
necking in Figs. 9 and 10.
Fig. 9 illustrates the fracture position of corroded steel plates. It
shows that most of the corroded specimens break near a section
with the minimum area (Amin). The fracture morphologies of two
types of specimens (un-corroded and corroded) are also shown in
Fig. 10. It can be seen the fracture morphology changes signifi-
cantly with the increasing corrosion degree. For corrosion-free
specimen, a typical necking with a cushion-like shape and a dimple
in the center of the fracture surface are observed, indicating that
crack initiates from the center and the fracture is ductile. While
for corroded specimens, the shear zone on fracture section
increases due to the pitting damage, and crack generally initiates
from the surface or root of corrosion pits, so that an obvious tearing
crack is observed. The visual evidence of the shift of crack initiation
site can be seen by comparing the scanning electron micrographs
of fracture surfaces [42]. For internal initiation, the fracture surface
is relatively rough and characterized by fairly equiaxed dimples
(see Fig. 10e), while for surface initiation, the fracture surface is
smoother and characterized by asymmetric dimples representative
of shear dominated deformation (see Fig. 10f).
4. Residual performance prediction
4.1. Summary of empirical formulas
In this section, we summarized the existing empirical formulas
used for predicting the residual performance of corroded steel.
Most of these formulas involved the yield load, ultimate load,
and total elongation.
4.1.1. Residual strength
Matsumoto [43] and Nakai [11] confirmed that the ultimate
load of a corroded steel plate was governed by the smallest
cross-section, i.e.:
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793 785

Fig. 9. Variation of thickness along Y direction and the fracture position: (a) HTF, (b) HBF, (c) HW.

Fig. 10. Fracture morphology of (a) NC, (b) HTF, (c) HBF, (d) HW and the SEM images of fracture that initiated from the center of intact specimen (e) and the surface of
corroded specimen (f).

F u ¼ WT ma f u0 ; ð8Þ T 0  T ma
Rm ¼ : ð11Þ
T0
where Tma refers to the average thickness of minimum cross-
section, defined by Amin/W. fu0 is the ultimate strength of
F u ¼ WT 0 f u0 ð1  Rm Þ0:73 : ð12Þ
corrosion-free plate. Sheng [14] found that ultimate load (or yield
load) and minimum cross-sectional area exhibited a good linear Muranaka [44] and Kariya [45] suggested using the average
relationship for Q235 steel: thickness (Tave) and standard deviation of thickness (Tsd) to predict
the ultimate load:
F u ¼ 486:7WT ma  5245; ð9Þ
F u ¼ WðT av e  aT sd Þf u0 ; ð13Þ
F y ¼ 185WT ma þ 24528: ð10Þ vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u k
u1 X
Ahmmad [13] introduced a damage index Rm, which depended T sd ¼t ðT i  T av e Þ2 ; ð14Þ
on the average thickness of minimum cross-section, to predict the k i¼1
ultimate load:
786 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793
where a is a coefficient, 0.7 and 1.3 were adopt in [44] and [45],
respectively. Ti is the average thickness of the ith cross-section, k
is the number of cross-sections, and Tave is the average value of
all Ti. Based on the previous researches, Appuhamy [10] proposed
a universal formula to predict the yield load and ultimate load of
corroded steel plates:
F y;u ¼ Wð0:2T 0 þ 0:8T min Þf yi0;u0 :
Here Tmin refers to the minimum thickness. In the work of
Garbatov [15], the volume loss ratio q was also used to quantify
the corrosion severity. Through regression analysis, he established
the empirical formulas for predicting the yield and ultimate load of
corroded steel (235 MPa):
F u ¼ ð0:0068q  2:3599q þ 400ÞWT 0 ;
2
F y ¼ ð0:0229q2 þ 0:5551q þ 235ÞWT 0 :
Since the corrosion studied in the present paper is a rather non-
uniform general corrosion, the reduction of bearing force can be
divided into two parts which are induced by uniform and non-
uniform volume loss:
F yi;u ¼ WðT 0  t u  ðten Þyi;u Þf yi0;u0 :
In this expression, tu and (ten)yi,u
refer to the (effective) thickness
loss induced by uniform and non-uniform volume loss, respec-
tively. Once (ten)yi,u is determined, the bearing force can be easily
calculated according to Eq. (18). However, it seems that the influ-
ence of non-uniform volume loss was never investigated sepa-
rately. As a matter of fact, the damage mechanisms of these two
parts on the bearing capacity are entirely different. In general,
the uniform thickness reduction will only cause the reduction of
bearing load without affecting the strength and ductility. Localized
volume loss can cause not only thickness loss but also stress/strain
concentration which will further lead to premature hardening and
strength degradation. In the work of Flaks [46], an approximate
method was proposed to calculate the effective thickness loss
(Ten)y,u induced by localized corrosion according to the features
of pits:
ðT en Þyi;u ¼ Q yi;u MW yi;u ;
qffiffiffiffiffiffi qffiffiffiffiffi
A 2 B 2
M¼ PdA ðhav e Þ þ PBd ðhav e Þ :
where Qyi,u and Wyi,u are empirical coefficients determined from
Table 5. have and Pd are average pitting depth (mm) and average
pitting distribution density (pits/cm2), respectively. Then the yield
and ultimate load of corroded steel plate can be further predicted
by Eq. (18) .
4.1.2. Deformability
The cross-sectional parameters, such as Tma, Rm, and Tmin may be
useful for estimating the residual strength. However, these param-
eters are completely unworkable in predicting the degradation of
deformability, because deformation is the behavior of whole paral-
lel part rather than a single cross-section. It has also been proved by
Ahmmad [13] that the deformability does not have a good correla-
tion with the maximum pit depth or damage parameter Rm directly.
Table 5
The values of Qyi,u and Wyi,u [46].
Index M < 0.3
Qyi 0.15
Qu 0.21
Wyi,u 0.448
Rather, it seems have a good correlation with the surface roughness
parameter Rs:
T av e  T ma
Rs ¼ ; ð21Þ
T0
d0 ¼ ð1  8:14Rs þ 26:4R2s Þd0 : ð22Þ
ð15Þ
In the work of Garbatov [15], the volume loss ratio q was also
utilized to predict the degradation of total elongation induced by
corrosion:
d0 ¼ ð0:0015q2  0:35051q þ 22Þ%: ð23Þ
Sheng [14] believed that the deformability of corroded steel
plate was mainly affected by the pit depth and denseness, and pro-
ð16Þ posed an empirical formula based on a multivariate regression
analysis:
ð17Þ
d0 ¼ 1:589Rt  0:3Ra þ 0:32; ð24Þ
T 0  T min
Rt ¼ ; ð25Þ
T ma
Ap
ð18Þ Ra ¼ : ð26Þ
WL
where Rt and Ra refer to the pit depth factor and pit denseness fac-
tor, respectively. Ap is the area that pits occupy the steel surface.
With the aid of corrosion morphology measurements and anal-
ysis, all evaluation indicators mentioned in this Section have been
extracted by CroEva in Section 3.2. The accuracy of the summa-
rized empirical formulas would be compared and discussed in
Section 4.3. Through the above summary, we found that the cur-
rent empirical formulas rarely involved the other important
parameters in the constitutive models of corroded steel such as
elastic modulus, yield strain, yield platform, ultimate strain and
hardening coefficient. Therefore, further more studies still need
to be carried out to establish more comprehensive and systematic
empirical formulas which can be applied to existing corroded
steel structures.
4.2. Numerical predictions based on corrosion morphology
ð19Þ
4.2.1. Material properties
ð20Þ Before conducting numerical analysis, the material true consti-
tutive relationship and fracture toughness of Q235 steel need to be
calibrated firstly.
In structural analysis, a power law constitutive relationship was
usually employed. However, it has been demonstrated that this
constitutive relationship may underestimate the actual stress-
strain curve beyond the uniform elongation (after necking) if we
simply extrapolate the measured stress-strain data [47–49]. In this
study, a recently developed advanced technique making use of DIC
technique and FE simulations was used to determine the true
stress-strain relationship [50]. Fig. 11 shows the digital images of
the strain field of tested NC coupon under different loads. The
DIC technique measured average axial strain-average axial stress
curve (ey  ry ) was employed as an initial guess of material true
constitutive relation and imported in the first FE analysis (FEA) ses-
sion. Then an iterative procedure was used to implement the suc-
cessive corrections on the initially assumed true stress-strain curve
based on the ratio between the experimentally measured (DIC) and
FE computed average axial stress-strain curves. Fig. 12 summarizes
the FE model and results of the tested NC coupon. It can be seen
M > 0.3 four FEA sessions are finally carried out to obtain the satisfactory
0.26 true stress-strain curve.
0.34 Since corrosion is a mesoscopic damage, the microscopic
0.836
fracture models VGM (void growth model) [51] and SMCS (strain
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793 787

modified fault strain model) [52–55] are used here to predict the
fracture behavior of corroded steel:

Z etp
VGM ¼ 1 expð1:5TÞdep  gt ð27Þ
0

SMCS ¼ ep  etp ¼ ep  at expð1:5TÞ ð28Þ

where ep and ep t are current and critical equivalent plastic strain,

Fig. 11. Selected digital images of NC coupon: (a) before the application of axial
yield loading; (b) at the maximum axial load point; (c) prior to final fracture.
T = rm/re is stress triaxiality (ratio of hydrostatic stress rm to effec-
tive stress re), gt and at are material dependent toughness param-
eters determined by tensile test of intact specimens. To calibrate
toughness parameters gt and at, the FE model for calibrating the
material true constitutive in Fig. 12 was also employed here. Once
the critical plastic strain and traxiality (at the center of the cross
section) corresponding to the critical displacement at fracture were
obtained, they were substituted into VGM and SMCS criteria to
back-calculate the critical toughness parameters. The calibrated
toughness parameters of gt and at for the tested material are
2.336 and 3.461, respectively. Besides, the parameter of character-
istic length should also be determined, which means that a macro
crack is triggered once the fracture index exceeds zero over the
characteristic length l⁄ [56]. According to the test results in [57],
250 um is appropriate.

Fig. 12. Procedures for determining the material true constitutive relationship: (a) detailed mesh of 1/8 FE model, (b) assumed (or corrected) true stress-strain curves
imported in the analysis sessions; (c) comparison of measured and computed ey  ry curves; (d) comparison of measured and computed F–DL curves.
788 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793
4.2.2. Establishment of finite element model
The high variability in corrosion morphology makes traditional
FE modeling approach not well suitable, since the mesh is required
to accurately reflect the fine details distributed over the entire cor-
rosion patch [58]. In the literatures of [1,3,6,9,10,13,22,28,59–64],
many researchers have proposed their own special method. In gen-
eral, shell element is typically used for compression analysis and
solid element is usually utilized in tension simulation. However,
the existing solid models (see Fig. 13) more or less possess the fol-
lowing ‘short slab’: 1) established based on the simplified geome-
try instead of actual corroded surface [13,64], 2) overly refined
element size and large model information [1,6,28], and 3) neglect-
ing the mesh dependency at large strain [9,10].
Considering that hexahedral element has fewer nodes under the
same size condition and provides higher computation precision
especially for the position with large strain gradient, the hexahe-
dral solid element with 8 nodes were employed here. Firstly, the
Z coordinate of each point on the two corroded surfaces (ZA and
ZB) was determined by Eqs. ((2)(3)) and then a series of points
with the intervals (I) in Table 6 were equidistantly extracted from
the original 500  1000 points. Subsequently, the Z coordinate of
the ith node on the jth layer between the lower and upper surface
was calculated by using the linear interpolation method, as shown
in Fig. 14:
Z ij ðx; yÞ ¼ Z iA ðx; yÞ  ðj  1ÞðZ iA ðx; yÞ  Z iB ðx; yÞÞ=n; ðj ¼ 2 . . . nÞ:
where n is the element number along the thickness direction. Then,
the Z coordinate of every generated node is orderly assigned to the
eight nodes of solid element to generate the FE model by using an
ANSYS interface specially written in MATLAB code. An example of
the established FE model is shown in Fig. 14a, and the meshing
information of all FE models is listed in Table 6.
The mesh dependency is examined in Fig. 15 by comparing the
stress contraction factor, load-displacement curve, yield load, ulti-
mate load and ultimate displacement of 7 FE models (taking HW
specimen for instance). The shapes of load-displacement curves
before necking are roughly same for all models, and the errors of
yield load and ultimate load are always kept below 1% irrespective
of the mesh density. Although the mesh density has a significant
influence on the error of simulated ultimate deformation, it is
shown to converge at the mesh size of 750 um. Moreover, the
stress distribution remains about the same and the error of stress
concentration factor is with 1% when I  750. Naturally, the ele-
ment size of 750 um is eventually selected for the FEA before neck-
ing occurs.
Since the mesh size effect is more evident after the necking
occurs, a local refined mesh was implemented at the expected
necking region by using our ANSYS interface. As shown in
Fig. 14c and d, the element size was refined to about 250 um,
which was consistent with the characteristic length l⁄ mentioned
above. The micromechanics based ductile fracture criteria of
Fig. 13. Existing solid modeling methods for corroded plates:(a) [1,6], (b) [64], (c) [10],
VGM and SMCS were both adopt to predict the fracture, and the
damaged elements were removed from the analysis when reaching
the critical state of rupture.
4.3. Predicted results and validation
In Table 7, the yield load, ultimate load and elongation at frac-
ture predicted by empirical formulas for each test specimen are
compared with experimental values. From Table 7, it can be seen
that the equations proposed by Muranaka [44] and Kariya [45] pro-
vide relatively more precise values of ultimate load with the max-
imum errors of 2.5% and 3.3%, respectively. In addition, the
ultimate loads predicted by the empirical formulas established
with Tma [11,13,43] can also satisfy the engineering requirements,
as most of the errors are within 5% of the experimental values. This
is consistent with the observation that most of the corroded spec-
imens break near a section with the minimum area (Amin). Given
that Flask [46] focuses on the effect of corrosion pits, Eq. (18) can
also provide relatively accurate predictions with the maximum
percentage error of 3.3. Although the empirical formula proposed
by Garbatov [15] is not a dimensionless expression, the ultimate
load can be predicted by Eq. (16) with an acceptable error, as we
employed the steel with same Grade. However, a large deviation
would be predictable for other steels with different strength. The
ð29Þ same conclusion can be reached for the formula proposed by Sheng
[14]. Unexpectedly, Eq. (15) does not provide accurate and stable
predictions in this paper and the maximum error reaches 16.3%,
which is contrary to the conclusion in [10]. For yield load, the com-
parisons show that the existing empirical formulas could not work
up satisfactory in predictions. In addition, the total elongation
predicted by Eq. (22) deviates from the testing results greatly,
especially for W specimens, which indicates that this formula
may not adequately take into account the effects of plate thickness.
Regardless of the predicting accuracy, the difference in the
expression of empirical formulas indirectly demonstrates that they
may not fully consider the severe randomness of corroded surface.
The inaccurate predictions do not mean that the empirical formu-
las have no usefulness for predicting the residual performance of
corroded steel, but rather that these empirical formulas may have
specific application conditions. Moreover, the empirical formulas
are all unworkable in predicting the degradation of deformability.
Therefore the degraded constitutive relation of corroded steel
cannot be determined based on the empirical formula method.
However, the same cannot be said for the morphology based
numerical method. Although the evolution of corrosion is extre-
mely random, the mechanical properties are unique for a steel plate
with a particular corrosion morphology. This correspondence can
greatly reduce the stochastic of predicted results. Moreover, this
method can provide continuous load-displacement curve of cor-
roded steel plate, which can be further transformed into the
stress-strain curve (or constitutive relation). To verify the accuracy
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793 789

Table 6
Summary of FE models.

Specimens Convergence checking Refined mesh

Tmax Interval I, or No. of extracted Selected No. of Refined Size of refined No. of refined Total No. of
(mm) mesh size (um) points on surface interval I (um) elements zone (mm) element (um) elements elements
HTF 7.98 600, 43  85, 750 33  66  11 25–45 250 224334 238854
HBF 7.93 750, 34  67, 33  66  11 25–45 224334 238854
HW 5.28 900, 29  57, 33  66  7 20–40 142758 151998
STF 7.86 1050, 25  49, 33  66  11 25–45 224334 238854
SBF 7.80 1200, 22  43, 33  66  11 15–35 234135 248292
SW 5.27 1800, 15  29, 33  66  7 26–46 142758 151998
VTF 7.36 2500 11  21 33  66  10 20–40 203940 217140
VBF 7.30 33  66  10 15–35 212850 225720
VW 5.42 33  66  7 15–35 148995 158004

Fig. 14. (a) Principle for the establishment of corroded FE model, (b) FE model of SW specimen established with the mesh size of I = 1200 um, (c) refined FE model of SW
specimen with a local refined mesh at the necking region.

Fig. 15. Examinations of mesh dependency performed on HW specimen: (a) stress distribution under the displacement of 45 um (I = 1800, 750 and 600 um); (b) effects of
mesh size on load-displacement curves; (c) variation of stress concentration factor, yield load, ultimate load and ultimate elongation with mesh size.
790 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793

Table 7
The results predicted by empirical formulas.

Specimens Fu/kN Fyi/kN d /%

Nak. [43,11] Ahm. [13] Mur. [44] Kar. [45] App. [10] Gar. [15] Eq. (18) App. [10] Gar. [15] Eq. (18) [46] Ahm. [13]
HTF 78.5 83.3 79.5 79.1 75.3 78.3 81.8 50.6 53.2 55.8 31.1
HBF 75.6 81.0 77.2 76.6 70.7 76.5 81.2 47.6 52.9 55.4 28.7
HW 50.1 53.5 50.6 50.2 46.5 52.8 52.5 31.3 37.6 36.7 31.5
STF 77.3 82.4 78.5 78.0 73.0 76.9 81.4 49.1 53.0 55.4 30.6
SBF 77.1 82.2 78.3 77.9 73.1 77.5 80.9 49.2 53.1 55.0 30.8
SW 48.8 53.9 49.9 49.4 42.8 50.9 52.8 28.8 37.0 36.2 28.9
VTF 75.4 78.4 76.1 75.9 70.8 71.6 77.5 47.6 47.6 52.5 32.3
VBF 73.3 76.8 74.2 73.9 67.5 70.4 76.0 45.4 47.4 51.6 31.3
VW 54.0 56.9 55.0 54.8 49.9 52.7 56.3 33.6 35.6 38.3 30.3
Errors (%) 4.6 5.5 2.5 3.3 16.3 6.4 3.3 9.8 16.0 13.5 33.7

Table 8
The results predicted by the morphology based numerical method.

Specimens Fyi/kN Fu/kN E/GPa eu /% d/% (VGM) d/% (SMCS)

HTF 53.3 80.2 195.6 18.9 32.7 33.4
HBF 52.4 78.9 193.8 19.1 32.6 33.5
HW 32.9 51.3 185.7 17.8 26.2 26.8
STF 52.7 79.1 196.6 19.1 33.7 33.7
SBF 52.4 78.9 197.5 19.2 32.2 32.2
SW 32.7 51.4 183.1 19.2 24.0 24.3
VTF 50.6 75.7 200.0 19.3 32.6 33.6
VBF 49.5 74.6 199.2 19.0 33.1 34.0
VW 37.3 55.5 201.4 19.9 32.6 33.3
Errors (%) 0.2–3.8 0.1–1.0 0.5–5.0 1.7–11.6 0.6–11.0 0.6–12.3

Fig. 16. Comparison of experimentally measured and numerically predicted load-displacement curves (H member).

and effectiveness of the numerical method, the computed results
such as the yield load, ultimate load and percentage errors are sum-
marized in Table 8. Compared with the empirical formulas, the
numerical method can provide more accurate and stable results
as the errors always remain within 1%. Here, a greatly improved
accuracy of the predicted yield load is also indicated in Table 8. Fur-
thermore, the numerical method can also provide appropriate val-
ues of elastic modulus and ultimate strain based on the computed
load-displacement curves, which are essential in structural analysis
but have not been fully assessed by the empirical equations. Fig. 16,
taking H member for example, shows the predicted load versus dis-
placement curves, which are overlaid on the experimentally mea-
sured results. Good agreement can be found, especially for the
uniform deformation phase before necking.
Fig. 17, taking STF specimen for example, compares the experi-
mental fracture position and predicted critical cross-section. This
illustration further indicates that the numerical method can
precisely predict the critical section, including its location and even
the angle of shear band (15 degree). The critical elongations at frac-
ture predicted by VGM and SMCS models are also listed in Table 8
and pointed out in Fig. 16 with different marks. As can be seen,
FEM computations tend to over-predict the fracture elongation.
This can be attribute to the shift in crack initiation from internal
to surface, which may have an impact on the fracture mechanism.
The FE results of plastic strain distribution at the fracture surface in
Fig. 17c also demonstrates the existence of such shift, which is con-
sistent with the experimental observations of fractography and our
previous work in [64]. The ductile fracture criterion VGM and
SMCS are derived on the assumption that equiaxed void growth
will control the initiation of fracture, whereas the surface fracture
initiation may be controlled by the nucleation and asymmetrical
distortion of voids in a shear dominated stress state [65].
Although a more accurate model that accounted for this limita-
tion was proposed in [66,67], too many parameters in this model
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793
Fig. 17. (a) Experimental and (b) predicted fracture position of STF specimen, and (c) plastic strain distribution at fracture surface.
Fig. 18. Axial strain distributions on a corroded surface of SW specimen: exper-
imentally measured at (a) maximum load and (b) fracture; numerically simulated
at (c) maximum load and (d) fracture.
need to be calibrated. Because of the lower experimental and com-
putational costs, the VGM and SMCS models are still effective tools
in practical applications to predict the fracture of corroded steel
plates. Actually, the errors of predicted values are kept about
11%. For security, a reduction factor of 1.1 is appropriate in
engineering application. Furthermore, given that VGM takes into
account the effect of entire stress history on fracture, more
791
accurate predictions obtained by VGM criterion is visible in
Fig. 16. On the other hand, VGM and SMCS criteria are actually
uncoupled models, which only describe the critical condition for
fracture initiation according to the stress and deformation evolu-
tion, but not incorporate the material damage into the constitutive
model directly. Hence, from Fig. 16 it can be clearly seen that
several predicted load-displacement curves are slightly higher
than the experimental curves after necking, for instance, SBF and
VTF specimens.
The comparison of experimental and predicted load-
displacement curves is a straightforward and convenient way to
check the effectiveness of proposed method but not sufficient.
Fig. 18 presents the measured and simulated axial strain distribu-
tions on a corroded surface of SW specimen at the maximum load
and fracture. Higher strains are generally found in most of the val-
leys of the corroded surface or at the critical section, consistent with
the previous reports [6,68]. The comparisons illustrate that there is
a good agreement between the predictions and experiments in
terms of the surface strain distributions, despite the slightly higher
value of prediction in magnitude. This slight deviation can be attri-
bute to the working principle of DIC technology, i.e., the surface
strain in a single point is not directly obtained but calculated from
the average displacement of several neighboring subsets of pixels,
typically containing 5–15 pixels [69].
The above analysis indicates that the numerical methodology
based on corrosion morphology for residual performance predic-
tions can capture the significant influence of corrosion on mechan-
ical properties. Through this method, the residual performance of
corroded steel including strength and deformability could be
accurately predicted without destruction and experimental tests.
Once the load-displacement curve of corroded steel plate has been
obtained, the degraded constitutive relation can be easily
determined and then imported into the structural analysis to fur-
ther assess the service performance of existing corroded steel
structures.
5. Summary and conclusion
5.1. Experimental results
The objective of this investigation was to assess the effects of
corrosion on the mechanical performance of steel. For corrosion
792 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793
characterization and tensile testing, main conclusions obtained are
as follows:
(1) Most of the specimens presented moderate corrosion after
8 years of natural exposure. The member position may have
a certain effect on the corrosion morphology. Great majority
of large corrosion pits distributed on the bottom flanges and
webs, and the pit density of H and S members was much
smaller than that of V member. Almost all of the corrosion
pits were found to be ‘shallow pit’ with the aspect ratios of
0.5–18, while the sky-ward surfaces had relatively sharper
pits. A relatively higher volume loss ratio was observed for
webs, thus, the corroded webs could to some extent be
determined as the ‘‘weak links” in assessing the residual per-
formance of corroded steel structures.
(2) Corrosion would cause a significant decrease in bearing
capacity and deformability. The ultimate load was governed
mainly by thickness loss, while the degradation of deforma-
bility was attributed to the localized volume loss. With the
increasing of non-uniform volume loss, a shorter yield pla-
teau was observed and eventually the well-defined yield
plateau even disappeared. Most corroded specimens broke
near a section with the minimum area, and crack generally
initiated from the surface or root of corrosion pits instead
of the center of cross-section.
5.2. Residual performance prediction
We summarized the previously proposed empirical formulas
and elaborated the 3D corrosion morphology based numerical
method to predict the residual performance of corroded steel.
The ultimate load predicted by most empirical formulas may
satisfy the engineering requirements. However, these empirical
formulas were not universally applicable to all corrosion states
due to the extreme complexity and stochasticity of corrosion.
Moreover, existing empirical formulas were all unworkable in pre-
dicting the degradation of deformability. Compared with empirical
formulas, the 3D profile based numerical method could achieve a
more reliable, stable and comprehensive prediction. Furthermore,
this method could also provide the degraded load-displacement
curve of corroded steel plate, which can be further transformed
into the stress-strain curve.
In our future work, the in-situ acquisition of corrosion morphol-
ogy and the practical application of the predicted load-
displacement curve in structural analysis will be further studied.
Since we can only get the 3D corrosion morphology of the area
of interest through sampling detection, the stochastic of predicted
results in the structural level need further studies. In addition, the
micromechanics based fracture models of SMCS and VGM
employed in the modeling tended to over-predict the fracture
elongation. Thus, the effects of corrosion on fracture mechanism
need to be further studied.
Acknowledgement
This work is supported by the National Natural Science Founda-
tion of China (Grant No. NSC-51378417), National Key Research
and Development Program (Grant No. 2016YFC0701305) and PhD
Innovation Fund of Xi’an University of Architecture and
Technology.
References
[1] R.M. Pidaparti, A.S. Rao, Analysis of pits induced stresses due to metal
corrosion, Corros. Sci. 50 (2008) 1932–1938.
[2] R.E. Melchers, Extreme value statistics and long-term marine pitting corrosion
of steel, Probab. Eng. Mech. 23 (2008) 482–488.
[3] M.R. Khedmati, M.M. Roshanali, Z.H.M.E. Nouri, Strength of steel plates with
both-sides randomly distributed with corrosion wastage under uniaxial
compression, Thin-Walled Struct. 49 (2011) 325–342.
[4] P.V. Petroyiannis, A.T. Kermanidis, R. Akid, C.A. Rodopoulos, S.G. Pantelakis,
Analysis of the effects of exfoliation corrosion on the fatigue behavior of the
2024–T351 aluminum alloy using the fatigue damage map, Int. J. Fatigue 27
(2005) 817–827.
[5] Y. Wang, J.A. Wharton, R.A. Shenoi, Ultimate strength analysis of aged steel-
plated structures exposed to marine corrosion damage: A review, Corros. Sci.
86 (2014) 42–60.
[6] R.M. Pidaparti, R.R. Patel, Correlation between corrosion pits and stresses in Al
alloys, Mater. Lett. 62 (2008) 4497–4499.
[7] M. Cerit, K. Genel, S. Eksi, Numerical investigation on stress concentration of
corrosion pit, Eng. Fail. Anal. 16 (2009) 2467–2472.
[8] M. Cerit, Numerical investigation on torsional stress concentration factor at the
semi elliptical corrosion pit, Corros. Sci. 67 (2013) 225–232.
[9] S. Kainuma, Y. Jeong, J. Ahn, Investigation on the stress concentration effect at
the corroded surface achieved by atmospheric exposure test, Mater. Sci. Eng. A
602 (2014) 89–97.
[10] J.M.R.S. Appuhamy, T. Kaita, M. Ohga, K. Fujii, Prediction of residual strength of
corroded tensile steel plates, Int. J. Steel Struct. 11 (2011) 65–79.
[11] T. Nakai, H. Matsushita, N. Yamamoto, H. Arai, Effect of pitting corrosion on
local strength of hold frames of bulk carriers 1st report, Mar. Struct. 17 (2004)
403–432.
[12] M. Bruneau, S.M. Zahrai, Effect of severe corrosion on cyclic ductility of steel, J.
Struct. Eng. 123 (1997) 1478–1486.
[13] M.M. Ahmmad, Y. Sumi, Strength and deformability of corroded steel plates
under quasi-static tensile load, J. Mar. Sci. Technol. 15 (2010) 1–15.
[14] J. Sheng, J. Xia, Effect of simulated pitting corrosion on the tensile properties of
steel, Constr. Build. Mater. 131 (2017) 90–100.
[15] Y. Garbatov, C.G. Soares, J. Parunov, J. Kodvanj, Tensile strength assessment of
corroded small scale specimens, Corros. Sci. 85 (2014) 296–303.
[16] D.B. Reiser, R.C. Alkire, The measurement of shape change during early stages
of corrosion pit growth, Corros. Sci. 24 (1984) 579–585.
[17] A. Česen, T. Kosec, A. Legat, Characterization of steel corrosion in mortar by
various electrochemical and physical techniques, Corros. Sci. 75 (2013) 47–57.
[18] S.M. Ghahari, A.J. Davenport, T. Rayment, T. Suter, J. Tinnes, C. Padovani, J.A.
Hammons, M. Stampanoni, F. Marone, R. Mokso, In situ synchrotron X-ray
micro-tomography study of pitting corrosion in stainless steel, Corros. Sci. 53
(2011) 2684–2687.
[19] D.A. Horner, B.J. Connolly, S. Zhou, L. Crocker, A. Turnbull, Novel images of the
evolution of stress corrosion cracks from corrosion pits, Corros. Sci. 53 (2011)
3466–3485.
[20] B. Holme, O. Lunder, Characterization of pitting corrosion by white light
interferometry, Corros. Sci. 49 (2007) 391–401.
[21] S. Xu, B. Qiu, Experimental study on fatigue behavior of corroded steel, Mater.
Sci. Eng. A 584 (2013) 163–169.
[22] S. Xu, Y. Wang, Estimating the effects of corrosion pits on the fatigue life of
steel plate based on the 3D profile, Int. J. Fatigue 72 (2015) 27–41.
[23] E.N. Codaro, R.Z. Nakazato, A.L. Horovistiz, L.M.F. Ribeiro, R.B. Ribeiro, L.R.O.
Hein, An image processing method for morphology characterization and
pitting corrosion evaluation, Mater. Sci. Eng. A 334 (2002) 298–306.
[24] H.W. Pickering, The role of electrode potential distribution in corrosion
processes, Mater. Sci. Eng. A 198 (1995) 213–223.
[25] C. Guillaumin, J. Gueriau, Quantitative measurement and shape
characterization of pits with an image analyzer, Proceedings of the
International Metallography Coference-MC95 1995.
[26] ASTM G46-94, Standard Guide for Examination and Evaluation of Pitting
Corrosion, American society for testing and materials, West Conshohocken,
1999.
[27] S. Xu, Y. Wang, Q. Xue, Evaluation indicators and extraction method for pitting
corrosion of structural steel, J. Harbin Inst. Technol. 22 (2015) 15–21.
[28] S. Lv, Q. Mu, X. Gao, T.S. Srivatsan, Influence of morphology of corrosion on
fracture initiation in an aluminum alloy, Mater. Des. 45 (2013) 96–102.
[29] Y. Garbatov, C.G. Soares, Corrosion wastage modeling of deteriorated ship
structures, Int. Shipbuild. Progress 55 (2008) 109–125.
[30] A.F. Mateus, J.A. Witz, On the post-buckling of corroded steel plates used in
marine structures, RINA Trans. 140 (1998) 165–183.
[31] A. Rahbar-Ranji, Stress Analysis of a Randomly Undulated Plate due to
Corrosion in Marine Structures, Yokohama National University, 2001.
[32] B. Qiu, The Study on Surface Characteristics and Eccentric Compressive Load
Capacity of Corroded H-shape Steel Members at Neutral Salt Fog Environment,
Xi’an University of Architecture and Technology, 2014.
[33] A. Rahbar-Ranji, Ultimate strength of corroded steel plates with irregular
surfaces under in-plane compression, Ocean Eng. 54 (2012) 261–269.
[34] J.E. Silva, Y. Garbatov, C.G. Soares, Ultimate strength assessment of rectangular
steel plates subjected to a random localized corrosion degradation, Eng. Struct.
52 (2013) 295–305.
[35] A.P. Teixeira, C.G. Soares, G. Wang, Probabilistic modelling of the ultimate
strength of ship plates with non-uniform corrosion, J. Mar. Sci. Technol. 18
(2013) 115–132.
[36] GB/T 228.1-2010, Metallic materials-Tensile testing-Part 1: Method of test at
room temperature, Standardization Administration of the People’s Republic of
China, Beijing, 2010.
 Y. Wang et al. / Construction and Building Materials 152 (2017) 777–793
[37] M.M. Kashani, A.J. Crewe, N.A. Alexander, Use of a 3D optical measurement
technique for stochastic corrosion pattern analysis of reinforcing bars
subjected to accelerated corrosion, Corros. Sci. 73 (2013) 208–221.
[38] C.A. Apostolopoulos, S. Demis, V.G. Papadakis, Chloride-induced corrosion of
steel reinforcement – Mechanical performance and pit depth analysis, Constr.
Build. Mater. 38 (2013) 139–146.
[39] M.M. Kashani, P. Alagheband, R. Khan, S. Davis, Impact of corrosion on low-
cycle fatigue degradation of reinforcing bars with the effect of inelastic
buckling, Int. J. Fatigue 77 (2015) 174–185.
[40] F. Tang, Z. Lin, G. Chen, W. Yi, Three-dimensional corrosion pit measurement
and statistical mechanical degradation analysis of deformed steel bars
subjected to accelerated corrosion, Constr. Build. Mater. 70 (2014) 104–117.
[41] L. Luo, Y. Huang, S. Weng, F.Z. Xuan, Mechanism-related modelling of pit
evaluation in the CrNiMoV steel in simulated environment of low pressure
nuclear steam turbine, Mater. Des. 105 (2016) 240–250.
[42] A.T. Myers, Testing and Probabilistic Simulation of Ductile Fracture Initiation
in Structural Steel Components and Weldments, Stanford University, 2009.
[43] M. Matsumoto, Y. Shirai, I. Nakamura, N. Shiraishi, A proposal of effective
thickness estimation method of corroded steel member, Bridge Found. Eng. 23
(1989) 19–25.
[44] A. Muranaka, O. Minata, K. Fujii, Estimation of residual strength and surface
irregularity of the corroded steel plates, J. Struct. Eng. 44A (1998) 1063–1071.
[45] A. Kariya, K. Tagaya, T. Kaita, K. Fujii, Basic study on effective thickness of
corroded steel plate and material property, IN: Proceedings of the Annual
conference of JSCE 2003.
[46] V.Y. Flaks, Correlation of pitting corrosion of aluminum plates and reduction of
load-bearing capacity under tension, Mater. Sci. 14 (1978) 75–78.
[47] Y. Wang, S. Xu, S. Ren, H. Wang, An experimental-numerical combined method
to determine the true constitutive relation of tensile specimens after necking,
Adv. Mater. Sci. Eng. 2016 (2016) 1–12.
[48] Y. Ling, Uniaxial true stress-strain after necking, AMP J. Technol. 5 (1996) 37–
48.
[49] A.C. Bannister, J.R. Ocejo, F. Gutierrez-Solana, Implications of the yield stress/
tensile stress ratio to the SINTAP failure assessment diagrams for
homogeneous materials, Eng. Fract. Mech. 67 (2000) 547–562.
[50] H. Tao, N. Zhang, W. Tong, An iterative procedure for determining effective
stress–strain curves of sheet metals, Int. J. Mech. Mater. Des. 5 (2009) 13–27.
[51] J.R. Rice, D.M. Tracey, On the ductile enlargement of voids in triaxial stress
fields, J. Mech. Phys. Solids 17 (1969) 201–217.
[52] J.W. Hancock, A.C. Mackenzie, On the mechanisms of ductile failure in high-
strength steels subjected to multi-axial stress-states, J. Mech. Phys. Solids 24
(1976) 147–160.
[53] G. Rousselier, Ductile fracture models and their potential in local approach of
fracture, Nucl. Eng. Des. 105 (1987) 97–111.
793
[54] W.M. Chi, A.M. Kanvinde, G.G. Deierlein, Prediction of ductile fracture in steel
connections using SMCS criterion, J. Struct. Eng. 132 (2006) 171–181.
[55] A.M. Kanvinde, G.G. Deierlein, Void growth model and stress modified critical
strain model to predict ductile fracture in structural steels, J. Struct. Eng. 132
(2006) 1907–1918.
[56] A.M. Kanvinde, Micromechanical Simulation of Earthquake Induced Fracture
in Steel Structures, Stanford University, 2004.
[57] F. Liao, W. Wang, Y. Chen, Parameter calibrations and application of
micromechanical fracture models of structural steels, Struct. Eng. Mech. 42
(2012) 153–174.
[58] D.T. Rusk, W. Hoppe, Fatigue life prediction of corrosion-damage high-strength
steel using an equivalent stress riser (ESR) model Part I: Test development and
results, Int. J. Fatigue 31 (2009) 1454–1463.
[59] D. Ok, Y. Pu, A. Incecik, Artificial neural networks and their application to
assessment of ultimate strength of plates with pitting corrosion, Ocean Eng. 34
(2007) 2222–2230.
[60] D. Ok, Y. Pu, A. Incecik, Computation of ultimate strength of locally corroded
unstiffened plates under uniaxial compression, Mar. Struct. 20 (2007) 100–
114.
[61] Y. Huang, Y. Zhang, G. Liu, Q. Zhang, Ultimate strength assessment of hull
structural plate with pitting corrosion damnification under biaxial
compression, Ocean Eng. 37 (2010) 1503–1512.
[62] X. Jiang, C.G. Soares, A closed form formula to predict the ultimate capacity of
pitted mild steel plate under biaxial compression, Thin-Walled Struct. 59
(2012) 27–34.
[63] T. Nakai, H. Matsushita, N. Yamamoto, Effect of pitting corrosion on the
ultimate strength of steel plates subjected to in-plane compression and
bending, J. Mar. Sci. Technol. 11 (2006) 52–64.
[64] S. Xu, H. Wang, A. Li, Y. Wang, L. Su, Effects of corrosion on surface
characterization and mechanical properties of butt-welded joints, J. Constr.
Steel Res. 126 (2016) 50–62.
[65] T.L. Anderson, Fracture Mechanics: Fundamentals and Applications, 3rd eds.,
CRC Press: Boca Raton, 2005.
[66] Y. Bao, T. Wierzbicki, On fracture locus in the equivalent strain and stress
triaxiality space, Int. J. Mech. Sci. 46 (2004) 81–98.
[67] T. Wierzbicki, Y. Bao, Y.W. Lee, Y. Bai, Calibration and evaluation of seven
fracture models, Int. J. Mech. Sci. 47 (2005) 719–743.
[68] P.E. Barnes, S.B. Lyon, D.L. Engelberg, In-situ surface strain mapping of pitting
corrosion in a high strength carbon Steel, European Corrosion Congress, 2014.
[69] J. Peirs, P. Verleysen, W.V. Paepegem, J. Degrieck, Determining the stress–
strain behaviour at large strains from high strain rate tensile and shear
experiments, Int. J. Impact Eng. 38 (2011) 406–415.