Service Life Prediction of Bridges Subjected To Corrosion Using TimeDependent Reliability Method
Service Life Prediction of Bridges Subjected To Corrosion Using TimeDependent Reliability Method
Service Life Prediction of Bridges Subjected To Corrosion Using TimeDependent Reliability Method
Doctor of Philosophy
Le Li
School of Engineering
RMIT University
July 2018
DECLARATION
I certify that except where due acknowledgement has been made, the work is that of
the author alone; the work has not been submitted previously, in whole or in part, to
qualify for any other academic award; the content of the thesis is the result of work
which has been carried out since the official commencement date of the approved
research program; any editorial work, paid or unpaid, carried out by a third party is
acknowledged; and, ethics procedures and guidelines have been followed.
Name: Le Li
Date: 15/10/2018
i
ACKNOWLEDGEMENTS
Thanks are extended to colleagues and friends within the research unit who provided
laughs and social relief through the years.
ii
CONTENTS
DECLARATION ......................................................................................................... i
ACKNOWLEDGEMENTS ....................................................................................... ii
CONTENTS ............................................................................................................... iii
LIST OF FIGURES .................................................................................................. vi
LIST OF TABLES .................................................................................................... ix
ABSTRACT ............................................................................................................... xi
LIST OF NOTATIONS .......................................................................................... xiii
LIST OF ABBREVIATIONS ................................................................................. xx
LIST OF PUBLICATIONS.................................................................................... xxi
Chapter 1: Introduction ............................................................................................ 1
1.1 Background ........................................................................................................ 1
1.2 Significance of the Research .............................................................................. 5
1.3 Aims and Objectives .......................................................................................... 8
1.4 Outline of the Thesis .......................................................................................... 9
Chapter 2: Literature Review ................................................................................. 10
2.1 Introduction ...................................................................................................... 10
2.2 Bridge Construction ......................................................................................... 10
2.2.1 Construction Materials............................................................................... 11
2.2.2 Bridge Design ............................................................................................ 12
2.2.3 Construction Methods ................................................................................ 13
2.2.4 Structural Analysis Methods...................................................................... 14
2.3 Steel Corrosion ................................................................................................. 14
2.3.1 Basics of Corrosion Science ...................................................................... 15
2.3.2 Uniform Corrosion..................................................................................... 15
2.3.3 Pitting Corrosion ........................................................................................ 19
2.3.4 Stress Effect on Corrosion ......................................................................... 22
2.4 Effect of Corrosion on Steel ............................................................................. 23
2.4.1 Effect of Corrosion on Geometry .............................................................. 24
2.4.2 Effect of Corrosion on Mechanical Properties .......................................... 27
2.4.3 Corrosion-Induced Delamination .............................................................. 35
2.4.4 Experiments on Corrosion Effect .............................................................. 37
2.4.5 Stress Corrosion Cracking ......................................................................... 47
2.5 Assessment of Bridge ....................................................................................... 48
2.5.1 Ultimate Capacity of Bridge Girder .......................................................... 49
2.5.2 Serviceability of Bridge Girder ................................................................. 51
2.5.3 Capacity of Connection ............................................................................. 52
2.5.4 Fatigue of Bridge ....................................................................................... 53
2.6 Reliability Theory and Methods ....................................................................... 58
2.6.1 Basics of Reliability Theory ...................................................................... 58
2.6.2 Time-Dependent Methods ......................................................................... 61
2.6.3 Monte Carlo Simulation ............................................................................ 64
2.6.4 Outcrossing Method................................................................................... 65
2.6.5 System Reliability Analysis ....................................................................... 67
2.7 Summary .......................................................................................................... 69
iii
Chapter 3: Experimental Research on Corrosion................................................. 71
3.1 Introduction ...................................................................................................... 71
3.2 Corrosion Tests................................................................................................. 72
3.2.1 Design of Test Specimens ......................................................................... 72
3.2.2 Immersion Test Procedure ......................................................................... 75
3.3 Corrosion Measurement ................................................................................... 76
3.3.1 Corrosion Loss Measurement .................................................................... 76
3.3.2 Hydrogen Measurement............................................................................. 77
3.3.3 Microstructural Measurements .................................................................. 78
3.4 Results and Discussion ..................................................................................... 83
3.4.1 Visual Inspection ....................................................................................... 83
3.4.2 Corrosion Loss ........................................................................................... 84
3.4.3 Hydrogen Concentration ............................................................................ 88
3.4.4 Microstructural Analysis ........................................................................... 89
3.4.5 Discussion .................................................................................................. 99
3.5 Field Investigations of Corrosion ................................................................... 104
3.5.1 Test Design .............................................................................................. 105
3.5.2 Results, Discussion and Comparison ....................................................... 109
3.6 Summary ........................................................................................................ 117
Chapter 4: Effect of Corrosion on Mechanical Properties................................. 118
4.1 Corrosion Effect on Tensile Properties .......................................................... 118
4.2 Introduction .................................................................................................... 118
4.2.1 Mechanical Tests ..................................................................................... 119
4.2.2 Field Tests ................................................................................................ 121
4.2.3 Results and Discussion ............................................................................ 125
4.3 Corrosion Effect on Fatigue ........................................................................... 144
4.3.1 Fatigue Model Development ................................................................... 144
4.3.2 S-N Curve Measurement .......................................................................... 146
4.4 Summary ........................................................................................................ 151
Chapter 5: Corrosion-Induced Delamination ..................................................... 152
5.1 Introduction .................................................................................................... 152
5.2 Experiment Delamination............................................................................... 153
5.2.1 Test Design .............................................................................................. 153
5.2.2 Observation of Delamination ................................................................... 155
5.2.3 Propagation of Delamination ................................................................... 166
5.3 Microstructural Analysis ................................................................................ 169
5.3.1 Grain Size Measurement.......................................................................... 169
5.3.2 Phase Measurement ................................................................................. 175
5.3.3 Distribution and Composition of Impurities ............................................ 179
5.4 Prevention of Delamination............................................................................ 182
5.4.1 Thermo-Mechanical Treatment ............................................................... 183
5.4.2 Equal-Channel Angular Pressing ............................................................. 183
5.4.3 Adding Alloying Element ........................................................................ 184
5.4.4 Eliminating the Difference in Phase Composition .................................. 184
5.4.5 Eliminating Impurities ............................................................................. 185
5.4.6 Removing Stress Effect ........................................................................... 185
5.5 Summary ........................................................................................................ 185
Chapter 6: Service Life Prediction ....................................................................... 187
6.1 Introduction .................................................................................................... 187
iv
6.2 Formulation of Service Life ........................................................................... 188
6.2.1 Definition of Service Life ........................................................................ 188
6.2.2 First Passage Probability.......................................................................... 189
6.2.3 Stochastic Process .................................................................................... 191
6.3 Limit State Functions for Girders ................................................................... 192
6.3.1 Flexural Failure ........................................................................................ 193
6.3.2 Shear Failure ............................................................................................ 193
6.3.3 Deflection Failure .................................................................................... 194
6.3.4 Fatigue Failure ......................................................................................... 194
6.4 Limit State Function for Connections ............................................................ 195
6.4.1 Tensile Failure ......................................................................................... 196
6.4.2 Fatigue Failure ......................................................................................... 196
6.5 Case Study Bridge .......................................................................................... 197
6.5.1 Summary of Assumptions........................................................................ 197
6.5.2 Structural Configuration .......................................................................... 198
6.5.3 Corrosion Models .................................................................................... 201
6.5.4 Load Information ..................................................................................... 202
6.5.5 Basic Variables ........................................................................................ 203
6.5.6 Time-Dependent Probability of Failure ................................................... 209
6.5.7 Sensitivity Analysis ................................................................................. 219
6.5.8 System Reliability Analysis ..................................................................... 226
6.6 Summary ........................................................................................................ 228
Chapter 7: Conclusions and Future Work .......................................................... 230
7.1 Conclusions .................................................................................................... 230
7.2 Recommendations for Future Work ............................................................... 233
References ............................................................................................................... 237
v
LIST OF FIGURES
vi
Figure 3-15: Hydrogen Concentration vs Corrosion Loss ......................................... 88
Figure 3-16: Element Percentage vs Corrosion Loss ................................................. 91
Figure 3-17: Change in Grain Size after 28 Days Corrosion ..................................... 93
Figure 3-18: Grain Size vs Corrosion Loss ................................................................ 93
Figure 3-19: Phase Composition vs Corrosion Loss .................................................. 95
Figure 3-20: Scanning Electron Microscopy Analysis of Corrosion Pits after 28 Days
Immersion .................................................................................................................. 97
Figure 3-21: Scanning Electron Microscopy Analysis of Micro-Cracks after 28 Days
Immersion .................................................................................................................. 99
Figure 3-22: Corrosion Loss for Stressed and Non-Stressed Steel .......................... 102
Figure 3-23: Test Locations ..................................................................................... 105
Figure 3-24: Site Measurement ................................................................................ 106
Figure 3-25: Girders of Demolished Viaduct with Designated Numbers ................ 107
Figure 3-26: Location of Cutting Samples for Destructive Test .............................. 107
Figure 3-27: Dimension of Specimens Cut from Viaduct ....................................... 108
Figure 3-28: Cutting of the Bridge ........................................................................... 108
Figure 3-29: Specimens before (left) and after (right) Rust Removal..................... 109
Figure 3-31: Corrosion Loss vs Time (Field Specimens) ........................................ 110
Figure 3-32: Hydrogen Concentration vs Corrosion Loss (Field Specimens) ......... 111
Figure 3-33: Element Composition Concentration vs Corrosion Loss (Field
Specimens) ............................................................................................................... 113
Figure 3-34: Grain Size at Various Levels of Corrosion Loss ................................. 115
Figure 3-35: Grain Size vs Corrosion Loss (Field Specimens)................................ 115
Figure 4-1: Set-Up of Tensile Test........................................................................... 120
Figure 4-2: Hardness Measurement on Site ............................................................. 122
Figure 4-3: Specimens before (left) and after (right) Rust Removal....................... 123
Figure 4-4: Relationship between Tensile Properties and Hardness ........................ 124
Figure 4-5: Stress-Strain Curve (1 M HCl, pH =-0.5) ............................................. 126
Figure 4-6: Stress-Strain Curve (3 M HCl, pH=-0.5) .............................................. 127
Figure 4-7: Changes in Ultimate Strength vs Time ................................................. 129
Figure 4-8: Changes in Ultimate Strength vs Time ................................................. 130
Figure 4-9: Yield Strength vs Corrosion Loss ......................................................... 131
Figure 4-10: Ultimate Strength vs Corrosion Loss .................................................. 134
Figure 4-11: Failure Strain vs Corrosion Loss ......................................................... 136
Figure 4-12: Yield Strength vs Corrosion Loss (Field Specimens) ........................ 137
Figure 4-13: Ultimate Strength vs Corrosion Loss (Field Specimens) ................... 138
Figure 4-14: Failure Strain vs Corrosion Loss (Field Specimens) ......................... 139
Figure 4-15: S-N Curve for Steel with and without Corrosion ................................ 146
Figure 4-16: Samples from Demolished Viaduct .................................................... 147
Figure 4-17: Fatigue Test Set-Up ............................................................................. 147
Figure 4-18: Comparison of Predicted S-N Curve with Test Results ...................... 150
Figure 5-1: Measurement of Width and Depth of Delamination ............................. 154
Figure 5-2: Corrosion Loss Measurement................................................................ 156
Figure 5-3: Specimens after 7 Days Immersion ....................................................... 158
vii
Figure 5-4: Specimens after 14 Days Immersion ..................................................... 159
Figure 5-5: Specimens after 28 Days Immersion ..................................................... 160
Figure 5-6: Optical Microscope Images of Samples after 7 Days Immersion (5×
Magnification) .......................................................................................................... 162
Figure 5-7: Optical Microscope Images of Samples after 14 Days (5× Magnification)
.................................................................................................................................. 163
Figure 5-8: Optical Microscope Images of Samples after 28 Days Immersion (5×
Magnification) .......................................................................................................... 165
Figure 5-9: Steel (Completely Delaminated) after 125 Days Immersion in 3 M HCl
.................................................................................................................................. 166
Figure 5-10: Depth and Width of Delamination vs Corrosion Time ....................... 167
Figure 5-11: Changes in Depth and Width of Delamination with Corrosion Loss . 168
Figure 5-12: Grain Size after Each Stage of Immersion in 3 M HCl....................... 172
Figure 5-13: Grain Size vs Corrosion Loss .............................................................. 174
Figure 5-14: Phase Morphology vs Corrosion Loss ................................................ 177
Figure 5-15: Electron Backscatter Diffraction Scan Images of Un-corroded and
Corroded Samples (Edge and Middle of Steel Thickness) ...................................... 179
Figure 5-16: Backscatter Scanning Electron Microscope Images of Un-corroded and
Corroded Samples (Edge and Middle of Thickness) ............................................... 181
Figure 5-17: Comparison of Conventional Continuous Rolling Procedure (left) and
Equal-Channel Angular Pressing Procedures (right) ............................................... 184
Figure 6-1: Geometry of Girders Cross-Section and Connections .......................... 201
Figure 6-2: Axial Load System of Train Carriages on the Case Study Viaduct ...... 203
Figure 6-3: Simulation of S(t) at Different Sample Sizes ........................................ 209
Figure 6-4: Probability of Failure for Various Coefficients of Correlation ............. 213
Figure 6-5: Probability of Fatigue Failure with Corrosion Impact on Mechanical
Properties ................................................................................................................. 217
Figure 6-6: Probability of Failure for Various Coefficients of Correlation ............. 218
Figure 6-7: Relative Contributions of Random Variables to Fatigue Failure of Mid
Girder at Different Times ......................................................................................... 221
Figure 6-8: Probability of Failure of Connection Plates due to Fatigue and Corrosion
for Different k r Values ............................................................................................ 223
Figure 6-9: Probability of Failure of Connection Plates due to Fatigue and Corrosion
for Different Corrosion Rate .................................................................................... 224
Figure 6-10: Probability of Fatigue Failure for Various Critical Damage
Accumulation Index Values ..................................................................................... 225
Figure 6-11: System Reliability Analysis of a Series System (Viaduct) ................. 226
Figure 6-12: Probability of System Failure vs Time................................................ 227
viii
LIST OF TABLES
ix
Table 5-2: Relationship between Grain Size and Corrosion Loss ........................... 169
Table 5-3: Relationship between Grain Size and Corrosion Loss ........................... 175
Table 5-4: Chemical Composition of Impuritiesa .................................................... 182
Table 6-1: Reformat of Limit State Functions ......................................................... 197
Table 6-2: Superimposed Dead Load Components and Values............................... 202
Table 6-3: Basic Variables ....................................................................................... 203
Table 6-4: Probability of Failure for Different ρ values .......................................... 213
x
ABSTRACT
Bridges are key infrastructures that play an important role in transport systems, but
the deterioration of this infrastructure has become a worldwide issue. This
deterioration affects the capacity of a bridge structure and reduces its remaining life.
It is, therefore, vital to study the corrosion effect on bridges and carry out service life
prediction of bridges subjected to corrosion to avoid corrosion-induced failures and
develop cost-efficient methods for the maintenance and rehabilitation of bridges.
The research aimed to develop a new method to predict the service life of a bridge
subjected to corrosion. The first-passage probability theory was used to determine
the time-dependent probability of bridge failure under different failure criterion and,
subsequently, the service life of the bridge. To obtain accurate input information,
simulated corrosion tests were conducted to determine the corrosion behaviour of
steel with and without stress. Mechanical tests were undertaken to determine the
degradation effect of corrosion on mechanical properties (yield strength, ultimate
strength, failure strain and fatigue strength limitation) with and without stress.
Hydrogen concentration measurement and microstructural analysis were conducted
to determine the mechanism of corrosion-induced degradation of mechanical
properties of steel with and without stress. Results of lab tests were correlated to
natural corrosion through site inspection. Based on the results collected from
corrosion tests and site inspection, bridge geometry and mechanical properties of
bridge steel were modelled as time-dependent variables when carrying out time-
dependent reliability analysis.
It was concluded that corrosion can lead to the reduction of ultimate strength and
failure strain of steel. Corrosion can also cause delamination due to intrinsic
differences in the microstructure across the thickness of steel plates. Additionally, the
levels of reduction of mechanical properties and propagation of delamination are
higher for stressed steel than for non-stressed steel at the same degree of corrosion.
Also, due to the effect of corrosion, the stress range (S) magnitude for corroded steel
is smaller than that for un-corroded steel at the same number of load cycles (N) to
failure in the S-N curve. The research also indicated the degradation of mechanical
properties is caused by corrosion- and hydrogen-induced changes in microstructural
xi
features. It was found, through reliability analysis, that fatigue failure is the most
critical failure criterion and that not considering the effect of corrosion on
mechanical properties leads to overestimation of the service life of bridges.
xii
LIST OF NOTATIONS
xiii
bw Web depth C Corrosion loss
xiv
gzm , gze , gzq Grain size at middle, HL Leeb hardness
one-quarter thickness
and edge
[H ] Concentration of I The second moment
hydrogen of inertia
xv
immersion period
xvi
R Resistance of the Ru Calculated ultimate
structure capacity
xvii
VR Shear resistance of girder VSl Shear force due to
load effects
S l
Coefficient of variation s Section slenderness
of Sl
xviii
sp The yield slenderness Mean
limit
xix
LIST OF ABBREVIATIONS
COV covariance
HL Leeb hardness
M Molar
OM optical microscope
S stress range
US United States
xx
LIST OF PUBLICATIONS
The following publications arose from research conducted during this thesis:
Li, L., Mahmoodian, M., Li, C. Q. & Robert, D. (2018). Effect of corrosion and
hydrogen embrittlement on microstructure and mechanical properties of mild
steel. Construction and Building Materials, 170, 78–90.
Li, L., Mahmoodian, M. & Li, C. Q. (Accepted). Effect of stress on corrosion and
mechanical property of mild steel. Journal of Materials in Civil Engineering.
Li, L., Mahmoodian, M. & Li, C. Q. (under review). Prediction of fatigue failure of
corroded steel structures. Structures.
Li, L., Mahmoodian, M. & Li, C. Q. (under review). Preferred corrosion and its
effect on delamination of steel. Construction and Building Materials.
Li, L., Mahmoodian, M., Zhou, S. W. & Li, C. Q. (2016). A comprehensive study on
corrosion and hydrogen embrittlement of low carbon steel. Paper presented at
the 24th Australian Conference on the Mechanics of Structures and Materials,
Perth, Australia.
Li, L., Mahmoodian. M., Zhou, S. W. & Li. C. Q. (2016). Comprehensive study on
hydrogen embrittlement and corrosion propagation in mild steel bridges.
Paper presented at the Sustainable Construction Materials and Technologies
(SCMT4), University of Nevada, Las Vegas, US.
xxi
Chapter 1: Introduction
1.1 Background
Bridges are key infrastructures that play an important role in transport systems, but
the deterioration of this infrastructure has become a worldwide issue (Kayser &
Nowak, 1989; Czarnecki & Nowak, 2006). Deterioration affects the capacity of a
bridge’s structure, reduces its remaining life and leads to bridge collapse (Czarnecki
& Nowak, 2008). For example, Sliver Bridge (see Figure 1-1) collapsed in 1967,
Lowe’s Motor Speedway Bridge (see Figure 1-2) collapsed in 2000 and Minneapolis
Bridge (see Figure 1-3) collapsed in 2007—all collapsed due to corrosion.
1
Figure 1-3: Collapse of Minneapolis Bridge
Therefore, the maintenance, repair and rehabilitation of existing bridge structures are
needed. For example, in the United States (US), 26.6% of the nation’s bridges were
reported to be structurally deteriorated in 2004. Estimation showed that over US$10
billion was used annually for bridge evaluation and maintenance from 1984 to 2004,
which is a significant financial burden (Czarnecki & Nowak, 2008). It is necessary to
predict the service life of steel bridges subjected to corrosion to prevent steel bridges
from unexpected failure and minimise the costs of bridge rehabilitation and
maintenance.
The service life prediction of bridges has been carried out widely in the last four
decades. However, a majority of studies used the deterministic method in the design
codes to determine the service life of the bridge based on different failure modes,
such as flexure, shear, deflection and fatigue (Priestley, Seible, Calvi & Calvi, 1996).
However, codes are primarily used for design (Melchers, 1999). The service life of a
bridge subjected to corrosion as determined by design code is less accurate than that
which can be estimated by time-dependent reliability analysis. This is because the
methods of service life prediction adopted in design codes are the target (i.e.,
constant), while the actual probability of failure changes with time due to the
deterioration of resistance and the increase of service loads. Additionally, design
codes consider the uncertainties associated with loadings and resistance through
‘partial factors’ (Page 4, Melchers, 1999). The probability of failure provided by
partial factors may be ‘lack of invariance’, as the failure modes may change due to
structural deterioration (Page 5, Melchers, 1999).
2
To perform time-dependent reliability analysis for bridges subjected to corrosion,
three aspects of work need to be conducted (Frangopol, Kallen & Van Noortwijk,
2004)—developing corrosion model, estimating corrosion effect and determining
limit state functions and using time-dependent reliability methods to make service
life prediction. There are several corrosion models developed to predict corrosion
loss over time. Kucera and Mattsson (1987) developed a widely accepted power law
model to predict corrosion loss. AS 4312 (Australian Standard, 2008b) assumed
corrosion rate (corrosion loss per year) is constant. Melchers (2015) found that the
power law corrosion model does not provide an accurate prediction of corrosion loss
for steel after exposure to the atmosphere over 100 years and developed a new model
for corrosion loss prediction based on a field study.
The effect of corrosion on the structural integrity of bridges in most cases has been
investigated by considering section loss (Darmawan et al., 2013). However, since the
1940s, some studies suggest corrosion also has a degrading effect on mechanical
properties (i.e., yield strength, ultimate strength, failure strain and fatigue strength
limitation) due to the increase of hydrogen concentration within steel (Revie, 2008;
Eggum, 2013). Revie (2008) and Marcus (2011) indicated that elements composition
changes during corrosion. The changes include an increase of oxygen content and
reduction of iron content which subsequently affects the mechanical properties.
Corrosion may also lead to degradation of mechanical properties by changing three
microstructural features—grain size, phase composition and formation of corrosion
pits (Li & Cheng, 2008; Marcus, 2011). The mechanism of the degradation of
mechanical properties due to corrosion is summarised in Figure 1-4.
3
Figure 1-4: Process of Mechanical Properties Degradation due to Corrosion
Based on the aforementioned studies, both the geometrical and mechanical properties
of steel change with time due to corrosion. Therefore, these two factors should be
treated as time-dependent variables in reliability analysis. However, few studies
considered mechanical properties as time-dependent in reliability analysis, which can
lead to the overestimation of service life. In addition, in most of cases bridge steel is
simultaneously stressed under loading and subjected to corrosion (Wang, Tang,
Wang, Wang & Guo, 2014). To develop an accurate corrosion model and estimate
the corrosion effect on mechanical properties, it is necessary to investigate the stress
effect on corrosion rate and the combined effect of stress and corrosion on
mechanical properties. However, there is limited literature comparing the corrosion
rate and level of degradation of mechanical properties between stressed and non-
stressed steel under the same corrosion degree.
4
performed reliability analysis using first order second moment method. They also
verified their results based on Monte Carlo simulations. Al Badran (2013) also
predicted the remaining life of bridge steel girder through Monte Carlo simulation.
The limit state functions were derived based on flexure, shear and deflection failure
modes, and system reliability analysis was also conducted.
The corrosion effect on a bridge at any point of time depends on previous corrosion-
induced damage (Ni et al., 2010). There is a high auto-correlation among the
corrosion effect at each time (Melchers, 1999). However, for the existing reliability-
based analysis methods on the service life, this auto-correlation has not been
considered (Pipinato & Modena, 2009; Nguyen, Garbatov & Soares, 2013). The
service life of a bridge subjected to corrosion can be overestimated by assuming
there is no auto-correlation (Li & Melchers, 2005). To overcome these limitations,
Mahmoodian and Li (2016) suggested using first-passage probability theory to
predict the fatigue life. This theory treats structural responses as stochastic processes.
The probability of failure can be determined based on the first occurrence of the
structural response up-crossing an acceptable limit (the threshold). The cross-
covariance function can be calculated in the first-passage probability method to
consider the auto-correlation among the fatigue damage at each time (Li & Melchers,
1993).
Therefore, five main aspects of work need to be conducted to make a more accurate
service life prediction of bridges:
This thesis focuses on corrosion and its effect on service life of the bridges, a
worldwide issue for bridge structures. As previously mentioned, 26.6% of bridges in
the US were reported to be structurally deteriorated in 2004. In Australia, the
5
structurally deterioration of bridges is also a serious issue. For example, in New
South Wales approximately 70% of bridges were built before 1985 and are subjected
to serious deterioration (Rashidi & Gibson, 2012). The total cost for maintenance and
rehabilitation of corroded bridges in Australia between 2010 and 2011 was A$1.2
billion, an increase of 67% from 2000 (GHD, 2015). It is imperative to make reliable
service life predictions of bridges subjected to corrosion, which can be achieved
through time-dependent reliability analysis. Based on the predicted service life,
bridge engineers and asset managers can make decision regarding the efficient
maintenance of bridges.
It is also imperative to predict the overestimation of service life, which can cause
catastrophic damage. For example, Biezma and Schanack (2007) pointed out that 5%
of bridge collapses worldwide from 1807 to 2007 were due to the overestimation of
service life and lack of maintenance. Ignoring the corrosion effect on mechanical
properties of steel is one of the main reasons for overestimation. For example,
Lichtenstein (1993) reported the collapse of Sliver Bridge was due to stress corrosion
cracking (SCC). Moriber (2015) reported the failure of anchor rods on the San
Francisco-Oakland Bay Bridge was due to the hydrogen-induced degradation in
mechanical properties. Lu et al. (2016) reported the collapse of a bridge in Port
Talbot in the United Kingdom was due to stress concentration at corrosion pits.
This thesis proposes a new method to make an accurate service life prediction of
bridges subjected to fatigue. Fatigue is a major failure mode of aged steel bridge
structures. Based on studies conducted by the committee of American Society of
Civil Engineers, 80–90% of the failures in steel structures are related to fatigue
(Zhao et al., 1994; Ni et al., 2010). Fatigue failure may occur suddenly even if the
load effect is smaller than the ultimate capacity of the bridge (Ni et al., 2010).
Corrosion leads to cross-sectional loss of bridge girders which affects stress range
under load cycles. It also affects the S-N curve and, subsequently, degrades the
fatigue strength limitation (Adasooriya & Siriwardane, 2014). Therefore, to make an
accurate prediction of fatigue life it is essential to consider the corrosion effects on
both stress range and S-N curve. This thesis investigates the corrosion effect on both
stress range and S-N curve. The damage accumulation index is modelled as a
6
stochastic process to prevent the overestimation of service life of bridges under
fatigue.
There is a commonly held view that stress does not affect corrosion and, thus, does
not affect the mechanical properties of corroded steel. However, based on a thorough
review of corrosion science and mechanics (Gutman, 1989; Revie, 2008; Ren,
Zhang, Pang & Gao, 2012; Xu & Cheng, 2012; Wang et al., 2014) and field survey
of corroded steel (Li, Mahmoodian, Li & Robert, 2018), it is argued in this thesis that
stress would affect corrosion by decreasing the corrosion resistance of steel (Ren et
al., 2012; Xu & Cheng, 2012). It is expected that the reduction in mechanical
properties is more significant for stressed steel than non-stressed steel under the same
degree of corrosion. This is because stressed steel is more susceptible to SCC and
hydrogen-induced degradation than non-stressed steel (Revie, 2008; Marcus, 2011).
However, very few studies have compared the level of degradation of mechanical
properties between stressed and non-stressed steel under the same corrosion degree.
The combined effect of corrosion and stress on mechanical properties has been
carried out mainly for pipeline steel, high tensile stress and stainless steel (e.g.,
Cheng, 2007; Ren et al., 2012). There is almost no research focusing on mild steel
with yield strength less than 350 MPa, which has been widely used for bridge and
building construction (Yakel & Azizinamini, 2005). In this thesis, to better
7
understand the mechanism of degradation of mechanical properties, hydrogen
concentration and microstructural features of steel are monitored and compared
during corrosion between stressed and non-stressed steel. These measurements and
quantitative comparisons have not been comprehensively carried out in previous
research.
The overall aim of this thesis is to develop a new method to predict the service life of
a steel bridge subjected to corrosion. In this method, the first-passage probability
theory is used to determine the time-dependent probability of bridge failure under
different failure criterion and, subsequently, determine the service life of the bridge.
To obtain accurate input data, simulated corrosion tests are conducted to investigate
the corrosion behaviour of steel with stress and without stress. Mechanical tests are
undertaken to determine the degradation effect of corrosion on mechanical properties
of steel with and without stress. Hydrogen concentration measurement and
microstructural analysis are conducted to determine the mechanism of corrosion-
induced degradation of mechanical properties of steel with and without stress. The
corrosion rate and the corrosion effect on yield strength, ultimate strength, ductility
and S-N curve are investigated. Results of lab tests are correlated to natural corrosion
through field studies. In this way, bridge resistance under different failure criterion
can be modelled as time-dependent variables when carrying out reliability analysis.
To meet the aim of this research, the following objectives are set:
8
1.4 Outline of the Thesis
The scope of this thesis is to study the corrosion behaviour and corrosion effect on
the bridge steel and predicts the bridge’s service life using time-dependent reliability
analysis. This thesis comprises seven chapters. Chapter 1 has presented the
background, significance, aims and objectives of the thesis.
Chapter 2 reviews the literature, including theories and methods such as the basics of
bridge construction, theory of steel corrosion, effect of corrosion on steel, effect of
stress corrosion, failure modes of bridges and theory of reliability analysis.
Chapter 7 concludes the thesis and recommends avenues for future research.
9
Chapter 2: Literature Review
2.1 Introduction
10
2.2.1 Construction Materials
Traditionally, the building materials for bridges are stone, timber, steel and concrete.
The first bridge was a supported beam made of wood constructed in the Palaeolithic
era (Pipinato, 2015). Stone later became the dominated bridge material due to its
strength and durability (Esaki & Jiang, 2000; Pipinato, 2015). In the eighteenth
century, steel and steel-reinforced concrete became the dominant materials for bridge
construction due to their strength, durability and workability (Pipinato, 2015;
Collings, 2005). Steel-concrete composite bridges were developed in the late
eighteen century, using the tensile strength of steel to gain girder capacity and the
compressive strength of concrete to gain slab capacity (Doran & Cather, 2013). This
composition provides an efficient solution over a wide range of spans (Collings,
2005).
According to AS 5100.6 (Australian Standard, 2017d), mild steel (low carbon steel)
is commonly used for bridge design and construction of steel-concrete composite
bridges (Collings, 2005) (see Table 2-1).
For the requirements of mechanical properties of the mild steel, yield strength ranges
from 250 MPa to 690 MPa, and the ultimate strength ranges from 300 MPa to 790
MPa (AS 5100.6 2017). The requirements of ductility are given in Clause 2.11 of AS
5100.6 (Australian Standard, 2017d).
11
cementitious material for concrete can be fly ash, ground granulated iron blast-
furnace slag and amorphous silica follow AS/NZS 3582 (Australian Standard, 2016).
The requirements for aggregate follow AS 2758 (Australian Standard, 2009). Other
requirements such as water-cement ratio, chemical admixtures and mixing products
follow AS 1379 (Australian Standard, 2007a).
Both mild steel and stainless steel can be used as reinforcement for concrete. The
yield strength of the reinforcement ranges from 250 to 650 MPa. Detailed
requirements of reinforcement are given in sections 3 and 17 of AS 5100.5
(Australian Standard, 2017c). The connections of the steel elements can be designed
using bolts and rivets. The yield strength of the bolts and rivets should be between
400 MPa and 830 MPa.
Viaduct is the type investigated in this thesis. Viaduct is a common bridge type
(Czarnecki & Nowak, 2008; Revie, 2008; Adasooriya & Siriwardane, 2014). This
thesis focuses on the service life prediction of viaducts subjected to corrosion. The
service life prediction of other bridge types can be conducted using a similar
12
approach. The most common viaduct form is two girder systems coupled by cross
beams (see Figure 2-1). Joints and bearings are installed for viaduct girders over 80
metres in length (Collings, 2005). This research focuses mainly on viaduct girders
and the connection between the girders, as their failure leads to the failure of the
entire system (Collings, 2005).
R Sl (2.1)
where R is the resistance of the structure, Sl is the load effect and is its relevant
capacity reduction factor.
This equation suits both ultimate limit states and serviceability limit states. Sl is
defined based on AS 5100.2 (Australian Standard, 2017b), (summarised in Section
2.2.4 of this thesis). R is determined by calculating the moment, shear, deflection
and fatigue capacity of cross-sections, detailed in AS 5100.6 (Australian Standard,
2017d).
13
Erecting steel girder and slab from the ground using cranes—the simplest
way for the construction of small span bridge
Longitudinal launching—suitable for continuous girders, especially
appropriate for girders high above the ground
Building by the balanced progressive cantilever—entire bridge sections are
elevated and connected to the rest parts of the bridge
Erecting by rotation—suitable when assembly can be done on the sides of the
obstacle to be crossed
Other special types of erection—includes transversal launching or ship
launching of entire girders.
The rating factor (RF) is used for analysis of existing bridge according to AS 5100.7
(Australian Standard, 2017e). Failure of the bridge system occurs if RF < 1. RF can
be determined as follows:
factor for dead load, S g* is load effects due to dead load, rgs is load factor for
superimposed dead load, S gs* is load effects due to superimposed dead load, S *p is
load effects due to parasitic effects or prestress, S s* is load effects due to differential
settlement, St* is load effects due to differential temperature, rL is load factor for live
(Australian Standard, 2017b) and W is multiple track factor for railway bridge and
lane factor for road bridge defined by AS 5100.2 (Australian Standard, 2017b).
14
(Darmawan et al., 2013). Therefore, there is a need to understand the corrosion
behaviour of viaduct steel (mild steel). In this section, the fundamentals of steel
corrosion are critically reviewed. Different forms of steel corrosion are also
introduced and summarised.
Uniform corrosion is the uniform thickness loss of mild steel without any localised
attack (Subramanian, 2008; Revie, 2008; Saha, 2012). The viaduct steel (mild steel)
in the atmosphere is mainly exposed to uniform corrosion (Subramanian, 2008).
In this reaction, mild steel corrodes and leads to the loss of cross-section of the
viaduct elements (Revie, 2008; Saha, 2012). The speed at which any given metal
deteriorates in a specific environment is defined as corrosion rate (Shi, Liu & Atrens,
15
2010). The corrosion rate of steel is controlled by the cathodic reaction. There are
two types of cathodic reactions—hydrogen evolution and oxygen reduction
(Chalaftris, 2003). The availability of oxygen in the environment determines which
reaction plays a dominant role (Chalaftris, 2003). In an environment where there is
limited oxygen (i.e., in deaerated solution), the reaction taking place in the catholic
area is given as follows:
1
H H 2 e (2.4)
2
This reaction proceeds rapidly in acidic solutions. However, it can also occur slowly
in alkaline solutions, neutral solutions and atmosphere. The hydrogen ( H 2 ) released
in this reaction can lead to the reduction of the ductility of mild steel (Chalaftris,
2003), discussed in Section 2.4.
In most of the cases, dissolved oxygen accelerates the catholic reaction by causing
oxygen reduction reaction (Revie, 2008), shown as follows:
1
H 2O O2 2e 2OH (2.5)
2
Further reactions can then happen, which lead to the formation of corrosion products.
By combining Equations 2.3 and 2.5 and using the reaction H 2O H OH , the
reaction is obtained as follows:
1
Fe H 2O O2 Fe(OH )2 (2.6)
2
16
shown in Figure 2-2. The corrosion process and products can also be affected by
trace anions ( , etc.) that exist in atmosphere or rainfall. Graedel and
Frankenthal (1990) summarised corrosion formations and their product (see Figure
2-3).
1 1
Fe(OH )2 H 2O O2 Fe(OH )3 (2.7)
2 2
Figure 2-3: Corrosion Process and Final Products for Steel Exposed to Different
Trace Anions
17
The corrosion rate of mild steel can be influenced by several factors (Revie, 2008;
Saha, 2012)—environmental conditions, microstructural features (element
composition, grain size, steel phase composition, morphology and impurities) of steel
and the presence of stress.
The microstructural features of mild steel (e.g., grain size, phase composition and
distribution of impurities) affect its resistance to corrosion (Syugaev et al., 2008;
Ralston & Birbilis, 2010; Marcus, 2011; Ferhat, Benchettara, Amara & Najjar,
2014). As mentioned above, passive oxide films are formed before and during
corrosion which provides a protective layer for steel against corrosion (Marcus,
2011). Smaller grain size and a larger proportion of cementite within steel help to
maintain the stability and adherence of these passive oxide films (Ralston & Birbilis,
2010). Impurities in steel accelerate the corrosion process by creating stress
concentration and galvanic reactions (Syugaev et al., 2008). Corrosion rate of mild
steel can be increased with the presence of stress (see Section 2.4.5).
From the literature, it is clear that uniform corrosion is the most common form of
corrosion the bridge is subjected to. The literature also gives a clear summary of the
18
mechanism of uniform corrosion, possible environmental factors that affect corrosion
rate and factors that affect the corrosion resistance of steel. However, the review
suggests there are three aspects of uniform corrosion that need further investigation.
Firstly, corrosion may change the element composition of steel based on Equations
2.2–2.8. However, there is a lack of research on monitoring the changes in element
composition during corrosion. In addition, as stated earlier, microstructural features
of mild steel affect its corrosion resistance. Corrosion can, in turn, change the
microstructural features by reducing grain size and number of impurities. However,
there are few studies that monitor the changes in microstructural features during
corrosion and its subsequent effect on the corrosion resistance of steel.
19
Figure 2-4: Pitting Corrosion
Corrosion pits are initiated due to the localised breakdown of the passive oxide film
(Revie, 2008). The passive oxide film can breakdown due to:
Corrosion pit can propagate when the pitting potential (the least positive current and
voltage at which pits develop or grow on a metallic surface) exceeds the re-
passivating potential (the critical current and voltage below which pitting corrosion
do not occur) (Soltis, 2015). Five main parameters can affect the growth rate of pits
(Revie, 2008):
20
1) Chloride concentration in the solution. The increasing concentration of
chloride prohibits the reformation of the passive oxide films and, consequently,
increases the growth rate of pits.
2) The proportion of alloying elements of steel. For example, high chromium
concentration within steel attributes to the reformation of the passive oxide film
at corrosions and, consequently, reduces the growth rate of the corrosion pits.
The growth rate of pitting can also be reduced for steel with the high
concentration of Molybdenum, as this can decrease the pit dissolution kinetics
and reduce the growth rate of pits.
3) The microstructural features of steel. Pits can grow faster at the locations
where there is a large number of impurities, a high proportion of pearlite phases
and high dislocations density at grain boundaries.
4) Temperatures. At low temperatures, the electrochemical potential to initiate
corrosion pits is extremely high and, therefore, pitting can only be initiated when
the temperature is above a critical pitting temperature (CPT). The CPT for steel
ranges from 10°C to 100°C. The growth rate of corrosion pits increases with the
increase of temperature when the temperature is above the CPT.
5) The presence of bacteria and pollutes at some location. Bacteria and pollutes
can contribute to localised corrosion and increase the growth rate of pits.
Pitting corrosion mainly initiates due to the rupture of the passive oxide film.
Therefore, it is reasonable to assume that pitting corrosion can be more serious under
combined stress and corrosion environment, as stress can rupture the passive oxide
film and making pitting corrosion more critical (Revie, 2008). Additionally, as stated
above, the growth rate of corrosion pits can be affected by the microstructural
features of steel. Therefore, it is vital to investigate the changes in microstructural
features of steel during corrosion to understand how corrosion pits can be initiated
and propagated. Pits can also transfer to cracks under combined stress and corrosion
environment when two criteria are stratified; the pit depth exceeds a threshold value
and mean growth rate of cracks is greater than growth rate of the pit. The existence
of corrosion pits can lead to degradation of mechanical properties of mild steel due to
stress concentration and crack initiation.
21
The literature states the initiation and propagation mechanism of pitting corrosion
and the factors that affect the propagation of pits. However, there are gaps. The
condition of passive oxide films affects the initiation and propagation of corrosion
pits. Stress can break the passive oxide film and contribute to the initiation and
propagation of corrosion pits. However, the growth rate of corrosion pits has not
been compared between stressed and non-stressed steel during corrosion. Moreover,
pits can transfer to cracks when the pit depth exceeds a threshold value and mean
growth rate of cracks is greater than growth rate of the pit. However, the mechanism
of pits to cracks transformation is still not clear.
In most of the cases, mild steel in the structure is simultaneously stressed under
loading and subjected to corrosion (Wang et al., 2014). Therefore, it is vital to
monitor and quantify the corrosion behaviour and corrosion impact of mild steel
under combined stress and corrosion environment.
Applied stress can lead to the decrease of corrosion resistance of steel (Javaherdashti
et al., 2006; Ren et al., 2012; Xu & Cheng, 2012). Corrosion is an electrochemical
process; corrosion rate increases with the decrease of electrochemical potential ( )
(see Section 2.3.2). The effect of stress on electrochemical potential can be
quantified by Gutman’s assumption, with equations showing as follows
PV
(2.9)
zF
where P is the applied stress, z is the ionic valency, V is the gram molecular
volume and F is Faraday’s constant. For the mild steel, z 2 , F 96, 485C / mol
and V 7cm3 .
From Equation 2.9, stress can decrease the electrochemical potential and,
consequently, increase the corrosion rate of steel. Applied stress can reduce the
electrochemical potential due to the following reasons (Gutman, 1989):
1) Passive oxide films are formed before and during corrosion (see Section
2.3.1). Stress ruptures the passive oxide film and causes the galvanic reaction
22
between the exposed steel and the surrounding oxide film at the film rupture site.
The extra galvanic reaction occurring during corrosion can reduce the
electrochemical potential.
2) Stress causes deformation of the steel surface, which increases the surface
energy and makes the corrosive solution easier to penetrate the steel surface.
3) With the presence of stress, local stress concentration can lead to the plastic
deformation at grain boundaries. The plastic deformation decreases the activation
energy of anodic dissolution and, subsequently, decreases the electrochemical
potential. Also, plastic deformation at grain boundaries can further rupture the
passive oxide films and reduce the corrosion resistance (Revie, 2008; Marcus,
2011).
In addition, corrosion under combined stress and corrosion environment can, in turn,
create more defects and stress concentration. Corrosion can also release hydrogen
and initiate hydrogen-induced cracks within steel. These impacts can lead to more
rupture of passive oxide film and dislocation among grain boundaries, further
reducing the electrochemical potential.
There are several studies that estimate the stress effect on corrosion rate of steel.
However, there are conflicting views on the effect of elastic stress on corrosion (see
Section 2.4.4). In addition, the changes in microstructural features, including the
morphology of steel microstructural, the reduction of grain size and the changes in
steel phase composition, need to be compared between stressed and non-stressed
steel to understand how combined stress and corrosion environment damage the
passive oxide films and, subsequently, the electromechanical potential of steel.
However, few studies have investigated these.
This section summarises the corrosion effect on mechanical properties of mild steel
and structural integrity of the viaduct. The models to predict corrosion-induced loss
23
of cross-section of the viaduct elements are reviewed. The mechanical properties of
mild steel can be affected by five main factors—hydrogen concentration, element
composition, grain size, morphology and phase composition (Revie, 2008; Ralston &
Birbilis, 2010; Eggum, 2013). The effect of corrosion on these factors is reviewed for
steel with and without stress. Corrosion due to intrinsic differences in the
microstructure is known as preferred corrosion. Localised stress concentration
caused by preferred corrosion can lead to the most severe form of corrosion-induced
degradation of steel—delamination, the mechanism of which is also reviewed in this
section.
Both uniform corrosion and pitting corrosion can lead to loss of cross-section of steel
elements (Kayser & Nowak, 1989; Czarnecki & Nowak, 2008). For uniform
corrosion of mild steel, there are several models to predict the loss of cross-section.
To begin with, corrosion rate can be determined by Australian Standard to predict
cross-section loss. AS 4312 (Australian Standard, 2008b) categorised corrosive
environments into five categories (see Table 2-2). The corrosion rate of steel in
different environmental categories can be determined by measuring the thickness
loss of the specimens for the first year (Australian Standard, 2008b; Darmawan et al.,
2013).
Kayser and Nowak (1989) developed the power law function, which is widely used
for corrosion rate prediction. The power law function is shown as follows
C kt m (2.10)
24
where C is the corrosion loss (thickness loss) after the exposure time of t (year), k
is corrosion loss when t = 1 and m is the regression exponent.
Landolfo, Cascini and Portioli (2010) pointed out that the power law function can
only predict the corrosion rate for steel exposed to atmosphere in a very short period
(within 10 years). For long-term exposure, Landolfo et al. (2010) developed a bi-
linear law function, shown as follows:
C cr t t 10 years (2.11)
where C is the corrosion loss, cr is the average corrosion rate (micrometres per
year) in 10 years, crlin steady state corrosion rate (micrometres per year) after time
years and t is the time. The values of cr and crlin are determined according to ISO
9224 (International Organization for Standardization, 2012) (see Table 2-4).
25
Table 2-4: Model Parameters for Bi-Linear Law Function
ISO 9224 category Corrosivity Carbon steel
cr crlin
C1 Very low cr 0.4 crlin 0.3
C2 Low 0.4 cr 8.3 0.3 crlin 4.9
C3 Medium 8.3 cr 17 4.9 crlin 10
C4 High 17 cr 27 0 crlin 16
C5 Very high 27 cr 67 16 crlin 138
Melchers and Jeffrey (2008) also developed a corrosion model suitable for uniform
corrosion of mild steel over an extended period (100 years) based on field studies.
The model is summarized as follows:
where c(t , E ) is the corrosion loss of material, fn(t , E ) is a mean valued function,
b(t , E ) is a bias function, (t , E ) is a zero mean error function (a probability
For pitting corrosion, the model to predict pit depth can also follow the power law
function in Equation 2.10, bi-linear law function in Equations 2.11 and 2.12, or the
model developed by Melchers and Jeffrey (2008) in Equation 2.13. The related
parameters k, m, cr , crlin , b(t , E ) , fn(t , E ) and (t , E ) can be determined by
experiments.
Each model has its limitations. The models suggested by AS 4312 (Australian
Standard, 2008b) have been widely used to predict section loss for bridges in
Australia. However, the corrosion rate suggested in Table 2-2 is conservative and can
lead to underestimation of the service life of a structure. The power law function in
Equation 2.10 and the bi-linear law function in Equation 2.11 provide a more
accurate estimation of corrosion rate compared with the standard’s method.
However, the parameters summarised in Tables 2-3 and 2-4 were derived from field
studies in the US and may not be suitable for the prediction of infrastructure
26
subjected to corrosion in Australia. Also, more field studies and experiments should
be conducted to determine the reliable value of b(t , E ) , fn(t , E ) , (t , E ) and E in
Melchers’ model. There is lack of field studies to develop models to predict the
corrosion rate of steel structures and infrastructure in Australia. In this thesis, a field
study is conducted to measure the thickness loss of steel elements at different times.
The design of the site inspection is summarised in Section 3.5.
Also, for a structure (such as a steel bridge) exposed to atmosphere, corrosion loss of
the cross-section is caused mainly by uniform corrosion. Therefore, this thesis
mainly focuses on the effect of uniform corrosion on the service life of the viaduct.
The effects of pitting corrosion on service life can be conducted using a similar
method as that of uniform corrosion (Czarnecki & Nowak, 2008).
Corrosion can affect the mechanical properties of mild steel by causing changes in
microstructural features. Also, corrosion-induced hydrogen damage (hydrogen
embrittlement) is a key mechanism leading to the changes in microstructural
features.
Corrosion can increase the hydrogen concentration within mild steel, which
subsequently affects the microstructural features and reduces the mechanical
properties (Chalaftris, 2003; Revie, 2008; Marcus, 2011; Eggum, 2013). In theory,
by rearranging the reaction from Equations 2.3–2.8, the cathode action of steel
corrosion can be rewritten (Eggum, 2013) as follows:
The hydrogen atoms can then ingress to the surface of steel due to its surface energy.
Afterwards, the hydrogen atoms can either form hydrogen molecules or diffuse into
the steel driving by the concentration gradient between the surface and interior
(Eggum, 2013). The ingress of hydrogen atom (H+) into steel can occur in seven
steps (see Figure 2-5).
27
Figure 2-5: Hydrogen Ingress into Steel
The hydrogen atoms, dissolved hydrogen gas ( ) released by corrosion can also be
diffused into steel (Eggum, 2013). The diffusion of dissolved hydrogen into steel
follows the Fick’s first law, which is described as follows:
[ H ]
J Df (2.15)
x
[ H ]
where J is the diffusion flux (moles cm-2 s-1), is concentration gradient
x
(moles cm-3 cm-1) and D f is the diffusion coefficient (cm-2 s-1).
For mild steel exposed to corrosion released hydrogen, more attention should be paid
to hydrogen concentration at any given time than the diffusion flux (Eggum, 2013).
Fick’s second law gives a relation that describes the change in the concentration
gradient with time, which is shown as follows:
[ H ] 2[H ]
Df (2.16)
x 2 x
28
Hydrogen, both in atoms and gas phase, can be trapped within the steel (Chalaftris,
2003; Revie, 2008; Eggum, 2013). Many locations within the steel can trap
hydrogen. The trapping of hydrogen can be considered as reversible or irreversible
depending on the relative activation energies required for hydrogen to escape
(Eggum, 2013). The trap locations and trap energies are shown in Table 2-5.
Once hydrogen enters and is trapped within the steel, the mechanical properties of
steel can be changed. The process by which the mechanical properties of steel are
changed due to the introduction and subsequent diffusion of hydrogen into the metal
is defined as hydrogen embrittlement (Chalaftris, 2003). The mechanism of hydrogen
embrittlement can be explained by internal pressure theory (Chalaftris, 2003).
In hydride-induced cracking, hydrogen atoms can be trapped in the regions with high
residual stress (e.g., voids, dislocations and holes) (Varias & Feng, 2004). A metal
hydride might form in these regions. Crack propagation could occur because of the
cracking of the brittle hydride phase.
29
2.4.2.1.3 Surface Energy (Adsorption Mode)
As mentioned earlier in this section, hydrogen atoms can be adsorbed onto the
surface of the steel and, subsequently, reduce the surface energy of steel (Tromans,
1994; Chalaftris, 2003). The reduction of surface energy leads to the initiation of
cracks and, subsequently, reduced ductility.
Once cracks are initiated on steel due to the combined effect of stress (residual stress
and applied stress) and corrosion, the plastic zone formed at the crack tip can make
the crack blunt and, subsequently, prevent the crack from propagating (Chalaftris,
2003). Hydrogen atoms can accumulate at the crack tip, deter the blunting of a sharp
crack and contribute to the crack propagation (Chalaftris, 2003).
Hydrogen (H+ and H2) can be absorbed or diffused into steel and reside in the regions
of high tri-axial stress (Oriani, 1972; Chalaftris, 2003). When the local concentration
of hydrogen reaches a critical level, hydrogen can weaken the cohesive force
between iron atoms (Oriani, 1972; Chalaftris, 2003). The reduction of cohesive force
between iron atoms can facilitate the nucleation of micro-cracks and reduce the
ductility of steel. Also, the local stress concentration at the crack tips traps more
hydrogen and further reduction the cohesive force between iron atoms at the crack
tips, which contributes to the propagation of cracks (Oriani, 1972; Chalaftris, 2003).
Corrosion can degrade the mechanical properties of steel, especially the ductility of
steel, by causing hydrogen embrittlement. However, there are several limitations in
existing literature. Hydrogen embrittlement phenomenon has been mainly carried out
for high-strength low-alloy steel and stainless steel (Revie, 2008; Olasolo, Uranga,
Rodriguez-Ibabe & López, 2011). For example, Zucchi, Grassi, Monticelli and
30
Trabanelli (2006) found hydrogen embrittlement can lead to degradation of
mechanical properties of stainless steel by charging hydrogen into steel specimen.
Hardie, Charles and Lopez (2006) found there is a drop in mechanical properties
after high-strength low-alloy steel is charged with hydrogen. For mild steel,
Chalaftris (2003), Eggum (2013) and Hardie et al. (2006) stated that steel with yield
strength less than 350 MPa is immune to hydrogen embrittlement. However, as
discussed earlier, hydrogen embrittlement occurs due to the accumulation of
hydrogen at voids or defects, which subsequently leads to inner pressure increment.
There is no indication that mild steel, especially that with yield strength less than 350
MPa, is safe from hydrogen embrittlement.
Considering these research gaps, the G250 mild steel, a common type of mild steel
used for bridge construction with yield strength around 300 MPa, is used in this
thesis to check its vulnerability to hydrogen embrittlement. The hydrogen
concentration can be monitored for stressed and non-stressed G250 mild steel during
corrosion. Detailed design of this test is in Section 3.2.1.
Corrosion can change the element composition, grain size, steel phase composition
and morphology of microstructure (Horner et al., 2011; Gonzaga, 2013; Zhou &
Yan, 2016).
The composition of chemical element of mild steel can affect its mechanical
properties (Bain & Paxton, 1966; Aravinthan & Nachimani, 2011; Lino et al., 2017).
The major chemical elements of mild steel include Iron, Carbon, Manganese,
31
Phosphorus, Silicon, Aluminium and Chromium. The effect of each element on the
mechanical properties of mild steel are summarised in Table 2-6.
Manganese (Mn) Can increase yield strength and ultimate strength, but reduce
ductility.
During corrosion, the corrosion reaction (Equations 2.3–2.8) can reduce the
proportion of iron of steel (De la Fuente, Díaz, Simancas, Chico & Morcillo, 2011).
Carbon proportion may be subsequently increased since they are not reacting with
acid. The portion of other alloying elements, such as Manganese, Phosphorus,
Silicon, Aluminium and Chromium can reduce during corrosion since they can either
be washed away or reacted with the corrosive solution (Krupp et al., 2004; Revie,
2008; Zhou, 2010; Eggum, 2013).
The proportion of two other elements may increase during corrosion, namely oxygen
and chloride (Revie, 2008; Saremi & Mahallati, 2002; Román, Vera, Bagnara,
Carvajal & Aperador, 2014). Corrosion may increase the oxygen composition due to
the formation of brittle rust layers during corrosion process (Revie, 2008). Exposing
steel to chloride enriched environment also results in chloride penetrations, in which
chloride ion breaks the passive oxide film formed during corrosion and makes the
steel vulnerable to pitting corrosion and SCC (Saremi & Mahallati, 2002; Román et
al., 2014).
32
As mentioned in Section 2.3.3, corrosion affects the morphology of microstructure
by initiating pits and cracks and, subsequently, changes the mechanical properties of
mild steel. The grain size of mild steel can be reduced due to intergranular corrosion.
Intergranular corrosion, by definition, is the preferred corrosion at grain boundaries
(Sinyavskij, Ulanova & Kalinin, 2004; Zhou & Yan, 2016). Sinyavskij et al. (2004)
indicated that this type of corrosion occurs because the boundaries of grains are
more susceptible to corrosion than their centres, as the alloying elements (see
Table 2-6) are likely to be depleted at the grain boundaries. The mechanism of
intergranular corrosion is shown in Figure 2-6. Intergranular corrosion weakens the
bonding force between grains and makes grain boundaries vulnerable to cracking
(Sinyavskij et al., 2004). The interaction of stress and corrosion can further reduce
the grain size by causing SCC along grain boundaries (IGSCC) (Revie, 2008;
Marcus, 2011). IGSCC is initiation and growth of cracks by localised corrosion
along the grain boundaries in steel with the presence of stress (Revie, 2008; Marcus,
2011).
Surface
Corrosion attack
at grain
boundaries Alloying element
Grain
For the corrosion effect on phase composition, mild steel contains two main phases
judging from its crystal structure—ferrite and pearlite. Ferrite is known as α-iron (α-
Fe), which provides ductility of steel, and pearlite is composed of ferrite (α-Fe) and
cementite (Fe3C) which makes steel brittle (Gonzaga, 2013). Sun, Li and Cheng
(2014) revealed ferrite is corrosion prone and cementite is corrosion resistant.
33
Therefore, it is expected that corrosion can change the proportion of ferrite and
pearlite within steel and, subsequently, affect the mechanical properties of steel.
Also, as pearlite is more brittle than ferrite, the composition of pearlite can be
reduced in combined stress and corrosion environment due to the pearlite fracture
(Gonzaga, 2013).
There are gaps in the current literature on the effect of corrosion on microstructural
features of mild steel. It is reasonable to believe corrosion changes element
compositions of mild steel and, consequently, affects the mechanical properties.
However, there is a lack of research that monitors changes in element composition of
mild steel during corrosion. According to the literature, the element composition of
steel is directly related to its mechanical properties. Therefore, the changes in
element composition can be a critical mechanism of the degradation of mechanical
properties during corrosion.
Also, cracks, grain boundaries dislocation and rupture of passive oxide film caused
by stress can contribute to the diffusion of oxygen (Revie, 2008). It is expected that
the oxygen proportion is higher for stressed steel than non-stressed steel under the
same degree of corrosion. Nonetheless, the element compositions for stressed and
non-stressed steel have not been compared during corrosion. The changes in element
composition can be measured for stressed and non-stressed steel through X-ray
fluorescence (XRF) (see Section 3.3.3).
The grain size analysis during corrosion has mainly been carried out for stainless
steel instead of mild steel and there are no studies that monitor the changes in steel
phase composition during corrosion. Additionally, the interaction of stress and
corrosion can further reduce grain size and affect steel phase composition (Revie,
2008; Marcus, 2011) which has not been quantified. There are few studies comparing
the reduction of grain size between stressed and non-stressed steel during corrosion.
The changes in phase composition have also not been monitored in previous studies
for stressed and non-stressed steel after corrosion. Therefore, a comprehensive study
on the grain size and steel phase of stressed and non-stressed steel after corrosion
needs to be conducted. The grain size measurement can be conducted through linear
intercept procedure and the steel phase composition can be measured by Electron
Backscatter Diffraction (EBSD).
34
2.4.3 Corrosion-Induced Delamination
In continuous casting, molten steel in the edges solidifies faster than that in the
middle of the thickness of steel (Thomas, 2002). This difference in solidification
velocity affects the microstructure of steel in three aspects (Zhang & Thomas, 2003;
Shanmugam et al., 2007; Olasolo et al., 2011)—distribution of impurities, phase
composition and grain size. Molten steel contains other chemical compounds which
are categorised as impurities. Zhang and Thomas (2003) suggested that dissolved
impurities (mainly Oxygen, Aluminium, Chromium) in molten steel precipitate when
their concentration increases. This causes impurities to accumulate and reside in the
middle of the steel where final solidification of molten steel takes place. Shanmugam
35
et al. (2007) reported that the average grain size of steel products increases with
decreasing solidification velocity. Therefore, it is expected that the grain size in the
middle is larger than that in the edge after continuously casting. As mentioned in
Section 2.4.2, steel contains two main phases regarding its crystal structure, ferrite
and pearlite. Shanmugam et al. (2007) and Olasolo et al. (2011) indicated that an
increase in solidification velocity disperses pearlite into cementite and ferrite
particles, which changes the phase composition of the steel. The middle of the steel
thickness can have a higher proportion of ferrite and a lower proportion of cementite
than the edge.
The microstructural characteristics of steel (e.g., grain size, phase composition and
distribution of impurities) affect its resistance to corrosion (Syugaev et al., 2008;
Marcus, 2011; Ralston & Birbilis, 2010; Ferhat et al., 2014). The passive oxide film
is formed before and during corrosion which provides a protective layer for steel
against corrosion (Marcus, 2011). Smaller grain size and a larger proportion of
cementite within the steel help to maintain the stability and adherence of the passive
oxide film (Ralston & Birbilis, 2010; Ferhat et al., 2014). Impurities in steel
accelerate corrosion process by creating a concentration of local stress and galvanic
reactions (Syugaev et al., 2008). For continuously cast steel, variation in
solidification velocity affects the microstructure and, thus, the corrosion resistance is
lower in the inner region (i.e., middle) compared to the outer regions (i.e., edges).
Corrosion due to intrinsic differences in the microstructure is known as preferred
corrosion by steel manufacturing industries (Chilingar, Mourhatch & Al-Qahtani,
2013). Preferred corrosion is localised, non-uniform corrosion which causes stress
concentration (Zhao et al., 2011). The localised stress concentration initiates cracking
in the steel, the scale of which can be such that the steel fractures completely. This
phenomenon is referred to as preferred corrosion-induced delamination in this thesis.
Delamination of steel completely destroys the integrity of steel as a building material
(e.g., Beidokhti, Dolati & Koukabi, 2009; Pantazopoulos & Vazdirvanidis, 2013).
There is little quantitative knowledge about the causes of preferred corrosion. There
is no knowledge as to how preferred corrosion leads to steel delamination. The latter
would directly destroy the integrity of steel and, subsequently, the safety of steel
structures. The understanding of the cause and effect of preferred corrosion can be
36
gained through simulated corrosion tests and detailed microstructural analysis of
corroded steel. Previous corrosion tests for continuously cast steel only focused on
the impact of manufacturing defects (e.g., central segregation, voids and cracks) on
the corrosion process (Kajatani, Drezet & Rappaz, 2001; Thomas, 2001; Nieto et al.,
2015). This represents only a small proportion of steel with poor manufacturing
quality (Thomas, 2001), which is diminishing due to increased quality control and
advances in steel manufacturing technology. There is no published study to date on
corrosion-induced steel delamination. Therefore, there is a clear need to conduct
corrosion tests on continuously cast steel with new testing methodology using
specimens with no manufacturing defects to acquire quantitative knowledge as to
how the microstructure of steel affects preferred corrosion and how corrosion leads
to delamination.
37
From the preliminary measurement of the changes in tensile properties during
corrosion, a model to predict tensile properties of mild steel based on corrosion
degree was then developed. In Garbatov, Soares, Parunov and Kodvanj (2014), steel
specimens were cut from a corroded steel box girder. The thickness losses of the
specimens were measured as an indication of corrosion degree. The models to predict
the tensile properties (modulus of elasticity, yield strength, ultimate strength and
elongation) based on corrosion loss were determined based on regression (see Table
2-7).
Hou, Lei, Li, Yang and Li (2016) also developed models to predict the corrosion
effect on mechanical properties of steel through a comprehensive test program. In
their test, two types of steel specimens were designed—the specimens designed for
tensile test followed ASTM E8/E8M-16a (ASTM, 2016) to test its mechanical
properties, and the specimens for fracture toughness test followed ASTM E1820-13
(ASTM, 2013b) to test its fracture toughness. The fracture toughness gave an
indication of the vulnerability of steel to cracking. The simulated corrosion test was
then conducted by immersing the steel in soil solutions with various pHs (3.0, 5.5
and 8.0) for a duration of three periods of 90, 180 and 270 days. The models are
summarised in Table 2-8.
38
Properties pH Function R2
Yield strength 3.00 y 6 105 C 0.0024 0.99
reduction
5.50 y 7 105 C 0.014 0.99
These studies indicate that there is a reduction in yield strength, ultimate strength and
ductility for steel after corrosion. The model developed to predict the level of
degradation of mechanical properties can vary according to the environmental
conditions in the corrosive simulation test (Tables 2-7 and 2-8). There is a need to
correlate the model developed in the lab to the natural corrosion through site
inspections as a way to develop more reliable models to predict the corrosion-
induced degradation in mechanical properties. Also, there is a lack of research
determining the mechanism of this reduction, in both macro and micro levels. Per
Section 2.4.2, hydrogen embrittlement and changes in microstructural features
(element composition, grain size, steel phase composition, morphology and
impurities) are all possible mechanisms leading to the degradation of mechanical
properties of steel during corrosion (Revie, 2008; Marcus, 2011). Therefore, it is
vital to carry out a comprehensive study to investigate the variation in all potential
influencing factors (hydrogen concentration, element composition, grain size, steel
phase composition, morphology and impurities) during corrosion.
The effect of hydrogen on mechanical properties and microstructure of steel has been
widely investigated by hydrogen charging test (e.g., Luo, Dong, Liu, Maha & Li,
2013). Hydrogen charging is carried out through Devanathan and Stachurski
39
apparatus (see Figure 2-8). The apparatus consists of two electrochemical cells
joined together by the steel specimens (Adams & Mickalonis, 2007). The solutions in
the electro-technical cell are shown in Figure 2-8. The charging side applies a
cathodic potential to encourage the hydrogen evolution reaction on the steel surface.
Hydrogen (H2) generated electrochemically at the cathode is then diffused through
the steel and oxidised at a constant potential at the anode (ASTM, 2018). The
oxidation current can be recorded, which indicates the amount of hydrogen charged
into the steel. Also, hydrogen charging can be achieved by immersing steel
specimens in de-aerated acidic solution (e.g., hydrochloride acid and sulphuric acid).
The immersion container is sealed and, therefore, the hydrogen (H2) released during
the hydrogen evolution reaction can be diffused directly in the steel driven by the
concentration ingredient.
After hydrogen charging, tensile test can be conducted to detect the changes in
mechanical properties of steel. According to Luo et al. (2013), when the hydrogen
within steel increased by 2 ppm, there were no significant changes in the yield
strength and ultimate strength, but there was 13.5% reduction in elongation
(ductility) of the specimens. While according to Eggum (2013), there was a 17.8%
decrease in yield strength, 16.1% decrease in ultimate strength and 34.5% decrease in
ductility when the hydrogen concentration increased by 2.0 ppm. The comparison
shows that the vulnerability of steel to hydrogen embrittlement varies according to
the steel grade, as Luo et al. (2013) used duplex 2205 steel and Eggum (2013) used
G550 steel.
40
The changes in the morphology of microstructure have been analysed for steel
through SEM after hydrogen charging. According to Revie (2008), hydrogen-
induced cracks have been found for charged specimens next to corrosion pits, voids
and impurities where there is stress concentration. Blisters have been found next to
cracks, which are signs of inner pressure increment due to the accumulation of
molecular hydrogen (H2) within the steel. The observation of blisters proves the
internal pressure theory (Section 2.4.2.1.1).
The combined stress and corrosion environment need to be created to investigate the
stress effect on corrosion of mild steel. Three methods can achieve the combined
stress and corrosion environment—constant load, bending test and slow strain rate
test (Jia, Wang, Han & Ke, 2011; Wang et al., 2014; Delaunois, Tshimombo, Stanciu
& Vitry, 2016). In the constant load test, the specimen is axially loaded in tension
41
and simultaneously subjected to corrosion. This stress and constant load combined
environment can be achieved commonly by three methods—the dead weigh constant
load machine, stress ring and clamp and rigs system (see Figure 2-9).
For example, Jia et al. (2011) used a dead weigh constant load machine to apply
series of axle load. SEM analysis was then conducted to observe the morphology of
cracks initiated under combined stress and corrosion environment. Wang et al.
(2014) designed rigs and bolts system to apply series of axle load on the steel. By
monitoring the corrosion current and electrochemical potential, the literature
concludes that stress can reduce the electrochemical potential and increase the
corrosion rate of steel. Also, Delaunois et al. (2016) used stress ring to apply 80% of
the ultimate strength to the specimens. After immersion test, the initiation and
propagation of SCC were monitored using an acoustic emission machine.
42
(c) Clamp and rigs system (Wang et al., 2014)
In bending test, the combined stress and corrosion environment can be prepared by
bending strip 180 degrees around a mandrel with a predetermined radius (see Figure
2-10a), which is defined as U-bending test according to Gerhardus (2001). In
addition to the U-bending test, the stress can also be applied to the specimens using
the bent-beam system where specimens are stressed by bending in a stressing device
while restraining the ends (see Figure 2-10b). The stressed system is then immersed
in corrosive solutions. The maximum stress on the test specimens (stress at the apex
of the U-bend or bent-beam) can be determined by attaching a strain gauge on the
apex (Gerhardus, 2001). During immersion test, corrosion rate can be recorded and
the initiation and propagation of cracks (SCC) can be observed.
43
(a) U-bending test (b) bent-beam system
Slow strain rate testing (SSRT), also called constant extension rate tensile testing, is
a widely-used test to study the combined stress and corrosion effect on the
mechanical properties of steel. It involves a slow (compared to conventional tensile
tests) dynamic strain applied at a constant extension rate in the environment of
interest (corrosive environment) (see Figure 2-11). The procedure of SSRT follows
ASTM G129-00 (ASTM, 2013d). The stress-strain curve for steel subjected to
combined stress and corrosion environment can be derived from SSRT. The changes
in tensile properties (elastic modulus, yield strength, ultimate strength and failure
strain) due to combined stress and corrosion environment can be estimated. For
example, through SSRT, Cheng (2007) indicated there was a 16.7% reduction in
yield strength, 15.3% reduction in ultimate strength and 19.0% reduction in ductility
for steel by exposing steel specimens to the simulated soil solution.
44
Figure 2-11: Slow Strain Rate Test
Among the three ways to create combined stress and corrosion environment, the
constant load test is preferred. Compared with the bending test, stress is uniformly
distributed cross the specimen thickness in the constant load test. In this way, the
stress applied on the specimen can be easily quantified and the effect of stress on
corrosion rate and mechanical properties of the specimen can then be estimated. In
SSRT, the stress applied to the specimen is not constant during corrosion and the
specimens are damaged by the test. In this way, it is difficult to estimate the effect of
different level of stress on corrosion rate through SSRT.
The review of the literature highlights the limitations of and gaps in the current
knowledge (Ren et al., 2012; Xu & Cheng, 2012). Firstly, existing literature focuses
on the acceleration effect of plastic stress on corrosion rate (Xu & Cheng, 2012).
However, in most of the cases, steel is subjected to elastic stress (70% to 80% of the
yield strength) under serviceability load. There is a lack of knowledge on the effect
of elastic stress on corrosion. Ren et al. (2012) and Wang et al. (2014) indicated that
45
elastic stress attributes to the surface energy increment and, therefore, increases the
corrosion rate of steel. However, Czarnecki and Nowak (2008) indicated that while
stress can accelerate the corrosion rate of steel, this acceleration effect is very small
and it is not a critical issue for steel elements. Lu et al. (2009) and Xu and Cheng
(2012) pointed out that only plastic stress can accelerate corrosion and an application
of elastic stress has no effect on the corrosion rate of steel. To examine these
conflicting views, a comprehensive experimental research needs to be conducted to
investigate the effect of elastic stress on corrosion rate.
46
Despite the previous experiments on corrosion-induced changes in steel
microstructure, the relationship among element composition, grain size, steel phase
and corrosion degree has not been developed. In addition, limitations have been
found in the current studies on the combined effect of stress and corrosion on
microstructural features of steel. It is expected the changes in element composition,
grain size and steel phase are larger for stressed steel than non-stressed steel at the
same degree of corrosion (Arioka, Yamada, Terachi & Staehle, 2006; Arafin &
Szpunar, 2009). However, the literature comparing these changes between stressed
and non-stressed steel during corrosion is insufficient.
SCC is the type of crack that takes place when steel is subjected to stress and
corrosion at the same time (see Figure 2-12) (Revie, 2008; Eggum, 2013). For mild
steel, SCC mechanism is usually explained by the slip dissolution model (Revie,
2008). The passive oxide film mentioned in Section 2.3.1 can be ruptured by stress.
Consequently, exposed steel and the non-rupture film form electrochemical system
and the galvanic reaction occur at the film rupture site. Steel acts as the anode and
the passive oxide film acts as the cathode in the galvanic reaction. As a result, the
mild steel begins to dissolve by anode reactions, which causes the crack to initiate.
This entire process repeats itself and leads to crack propagation (Revie, 2008;
Eggum, 2013). SCC can significantly reduce the yield strength, ultimate strength and
ductility of steel. In addition, during the crack propagation, the stress intensity factors
at the crack tips increase and the steel specimen can be fractured when the stress
intensity factor is larger than the fracture toughness of steel (Revie, 2008; Eggum,
2013).
47
Figure 2-12: Stress Corrosion Cracking
Judging from the mechanism of SCC, it can be assumed that SCC can occur on any
steel subjected to combined stress and corrosion environment. However, the
vulnerability of mild steel to SCC in natural corrosion environment is not clear.
According to Revie (2008), SCC can only occur for particular alloys exposed to a
few very specific environments (e.g., ferritic steels exposed to hydroxides or nitrates
and brasses exposed to ammoniacal environments). However, according to Uhlig
(2011), SCC can also occur for alloys beyond these specific environments. The
corrosion product, the element composition of steel and the loading conditions can
also affect the threshold stress initiation for SCC. For mild steel subjected to natural
corrosion, there is a need to collect the steel specimens from in-service structures and
use the SEM to observe whether SCC has been initiated.
As mentioned in Section 2.1, design and assessment of a steel viaduct are carried out
based on the relationship between resistance and viaduct load effect. The girders and
connections between girders are essential elements of viaducts, as their failure can
lead to the collapse of the entire viaduct system (Ni et al., 2010). Therefore, the
viaduct assessment carried out in this research focused on viaduct girders and
connections. For girders, resistance is defined by calculating the moment, shear,
deflection and fatigue capacity of cross-sections, with the details given in AS 5100.6
(Australian Standard, 2017d). For connections subjected to tension, shear, tension-
48
shear combined environment and fatigue, its capacity can also be determined by AS
5100.6 (Australian Standard, 2017d). This section summarises this method to
determine the moment, shear, deflection and fatigue capacity of a viaduct girder and
the capacity of a connection.
M * Mb (2.17)
where M * is the largest design bending moment the girder is subjected to, is the
capacity reduction factor, which is 0.9 for members subjected to bending, and M b is
For a section with full lateral support, the moment capacity can be determined as
follows:
M b y Ze (2.18)
where y is the yield strength and Ze is the effective section modulus of the cross-
b y
s ( ) (2.19)
t 250
49
where b is the clear width of the element and t is the element thickness. For compact
section ( s sp ), where sp is the yield slenderness limit determined based on AS
5100.6 (Australian Standard, 2017d), the effective section modulus ( Ze ) is the lesser
of the plastic section modulus and 1.5 times the elastic section modulus. For non-
compact section ( s sp ), the effective section modulus ( Ze ) is determined based
on the effective cross-section. The details to determine the effective cross-section are
given in Clause 5.1.4 in AS 5100.7 (Australian Standard, 2017e)
For a section partially restrained at one edge or both edges, the moment capacity can
be determined as follows:
M b m s M sl (2.20)
where m is the moment modification factor for bending and s is the slenderness
reduction factor (Australian Standard, 2017d). These two factors are determined by
Clause 5.6.1 in AS 5100.7 (Australian Standard, 2017e). M sl is the nominal section
V * Vb (2.21)
where V * is the largest shear force the girder subjected to, is the capacity
reduction factor, which is 0.9 for members subjected to shear and Vb is the nominal
shear capacity of the girder (Australian Standard, 2017d). Most of the viaduct girders
are designed as I shape or hollow sections. For I shape girder, the shear capacity Vb
can be determined as follows:
Vb 0.6 y Ae (2.22)
where Aw is the gross cross-sectional area of the girder web (Australian Standard,
2017d). For a girder with the hollow section, the shear capacity can be determined
as:
Vb 0.36 y Ae (2.23)
50
where Ae is the effective area of the cross-section. When the net area of hollow
section is greater than 0.9 times the gross area, Ae shall be taken as the gross area of
hollow section. Otherwise, it shall be taken as the net area of the hollow section
(Australian Standard, 2017d).
R Sl (2.24)
where R is the allowable deflection under live load and Sl is the deflection at the
middle of the span due to live load (Australian Standard, 2017d). The maximum mid-
span deflection under live load should not exceed 1/640 of the span length. Sl can
be determined by the elastic theory in structural analysis. For example, for a simply
supported girder subjected to the load condition (see Figure 2-13), Sl can be
determined as follows:
l l
P( x)[3l 2 ( x) 2 ]
Pl 3 2 2
Sl (2.25)
48EI 48EI
where P is the axle load of the train on the girders, l is the length of the span, x is
the distance between axial load, E is the elastic modulus of steel and I is the second
moment of inertia.
51
2.5.3 Capacity of Connection
The connections between girders are designed so the structure can resist all design
actions. Girder connections can be bolts or rivets. For bolts and rivets subjected to
shear, the requirement is shown as follows:
V f* V f (2.26)
where V f* is the shear force the connection subjected to, is the capacity reduction
factor, which is 0.8 for bolts and power-driven rivets and 0.6 for hand-driven rivets
(Australian Standard, 2017d). V f is the nominal shear capacity of the bolts or rivets,
where uf is the tensile strength of the bolts or rivets, kre is the reduction factor
determined based on the length of the connection, nn is the number of shear planes
with threads intercepting the shear plane in a bolted or riveted connection, Ac is the
minor diameter area of the bolt or rivet, nx is the number of shear planes without the
threads intercepting the shear plane and Ao is the nominal plain shank area of the
bolt or rivets (Australian Standard, 2017d).
where N tf * is the tensile force of the connection, is 0.8 for bolts and power-driven
rivets and is 0.6 for hand-driven rivets (Australian Standard, 2017d). N tf is the
52
For bolts and rivets subjected to the combined shear and tension, the requirement is
shown as follows:
2 2
V f* Ntf *
1.0 (2.30)
V f Ntf
where is 0.8 for bolts and power-driven rivets and is 0.6 for hand-driven rivets, V f
The fatigue capacity of a structural component has been estimated widely through
the S-N curve and damage accumulation role (Adasooriya & Siriwardane, 2014). The
S-N curve is a plot of the magnitude of stress range ( ) versus fatigue and the number
of load cycles (N) to failure, which can be expressed as follows:
N AS B (2.31)
where A is the fatigue strength coefficient and B is the fatigue strength exponent.
Equation 2.31 assumes the stress ranges of the structural component are constant.
Equation 2.31 can be expressed in the log scale and it leads to a straight line as
follows (Zhao et al., 1994; Ni et al., 2010):
53
For mild steel subjected to normal stress range, the typical S-N curve in log-log plot
is shown in Figure 2-14. N 0 is the largest load cycles in the low load cycle region
and N1 is the minimum load cycle in the high load cycle region. S-N curve for mild
steel, based on its log-log plot, can be simplified and represented by three straight
lines (Adasooriya & Siriwardane, 2014; Bandara, Dissanayake & Dissanayake,
2015). Line 1 is the horizontal line across ultimate strength, which indicates that the
stress range is equal to ultimate strength ( u ) in the low load cycle region (
0 N N0 ).Line 2 is the horizontal line across endurance limit, which indicates that
the stress range is equal to fatigue strength limitation ( ) in the high load cycle
(British Standards Institution, 1980) as the stress range when load cycle N N1 .
Line 3 can be described by Equation 2.32 (Adasooriya & Siriwardane, 2014). Based
on Bandara et al. (2015), fatigue strength limitation ( ) is 0.4–0.6 of ultimate
strength.
𝜎𝑢
𝜎∞
Figure 2-14: S-N Curve of Mild Steel Subjected to Normal Stress Range
For any specific mild steel subjected to normal stress range, the S-N curve can be
determined by knowing the classification (Kerekes & Petrovics, 2000; Zhiyuan et al.,
2016) (see Figure 2-15 and Table 2-9).
54
Figure 2-15: Example of Determination of S-N Curve
For typical mild steel subjected to shear stress, the S-N curve in the log-log plot is
described by Equation 2.32 and shown in Figure 2-16 (Nussbaumer, Borges &
Davaine, 2012). A and B can be determined by knowing the strength value at N =
105 and N = 108 ( S5 and S8 ). Since BS 5400.10 (British Standard Institution, 1980)
does not provide S-N curve for rivets or girder under shear stress, S5 and S8 are
determined based on AS 5100.6 (Australian Standard, 2017d).
55
Figure 2-16: S-N Curve of Mild Steel Subjected to Shear Stress Range
K K
ni
D D (2.33)
i 1 i 1 Ni
where ni is the number of stress cycles, N i is the total number of cycles to failure
under constant stress range Si and D is the Miner’s damage accumulation index
which indicates the fatigue damage.
Combining S-N curve, the variable amplitude of S and Miner’s rule, the damage
accumulation index ( D ) can be expressed as follows (Zhao et al., 1994):
n
D E (S B ) (2.34)
A
where n is the total number of stress cycles under variable stress ranges, S is the
stress range parameter and E ( S B ) is the expected value of S B . E ( S B ) can be
calculated as follows (Zhao et al., 1994):
B
E ( S B ) 2 S 0 ( 1) (2.35)
2
56
where S 0 is a statistical parameter determined as follows:
2
S0 E (S ) (2.36)
where E ( S ) is the mean stress effect. It can either be normal stress effect or shear
stress effect.
According to Miner’s rule, fatigue failure occurs when the damage accumulation
index ( D ) exceeds the critical damage accumulation index ( , which can be
expressed as follows (Zhao et al., 1994):
D 0 (2.37)
where is the critical damage accumulation index, the limit of which is equal to 1.0.
The standard summarises the equations to determine the capacity of a girder under
different failure criteria (i.e., Equations 2.18, 2.22, 2.23, 2.25, 2.27, 2.29 and 2.34)
and this can be used to develop the limit state functions for reliability analysis (see
Section 2.6). However, the design codes add conservativeness to the assessment of
the probability of viaduct failure and prediction of service life. Also, design codes
consider the uncertainties associated with loadings and resistance through ‘partial
factors’ (Melchers, 1999) (e.g., m and s in Equation 2.20 and kre in Equation
2.27). The probability of failure provided by partial factors may be ‘lack of
invariance’ for different failure criteria (i.e., limit states), since the failure modes and
their uncertainties may change with time. Therefore, time-dependent reliability
analysis needs to be conducted to make an accurate assessment of the failure
probability of existing viaducts and their service life under different failure modes.
conflicting views on which failure modes are the most critical for the viaduct
57
subjected to corrosion. For example, Czarnecki and Nowak (2008) indicated the
deflection is the most critical failure modes for viaduct subjected to long-term
corrosion, but Siriwardane et al. (2008) stated that fatigue is the most critical.
Therefore, for any selected viaduct subjected to corrosion, the assessment needs to be
conducted for multiple failure modes to define which one is the most critical.
Additionally, very few studies consider the corrosion effect on stress range and S-N
curve in fatigue analysis. For example, Siriwardane et al. (2008) and Imam,
Righiniotis and Chryssanthopoulos (2010) investigated the corrosion effect on stress
range but did not consider corrosion effect on S-N curve. Similarly, Adasooriya and
Siriwardane (2014) used finite element model to estimate the changes in stress range
of viaduct due to corrosion-induced section loss and used the S-N curve without cut-
off limit (i.e., fatigue strength limitation equal to zero) to estimate the impact of
corrosion on S-N curve. However, there is an absence of experimental and field data
to estimate the changes in S-N curves as a function of time and corrosion loss.
58
can be geometry of the cross-section, mechanical properties of steel and the load
effect (Bazovsky, 2004).
0
P( X x) FX ( x) f X ( )d (2.38)
59
where FX ( x) is defined as the cumulative distribution function of X and f X ( ) is the
probability density function.
Reliability is defined as the probability that a structure performs its intended function
during a specified period under stated conditions. For a structural component (or
system) such as a steel viaduct, the basic reliability problem considers only one load
effect Sl resisted by one resistance R. R and Sl can be described by a probability
As mentioned in Section 2.2.2, structural component (or system) fails if its resistance
R is less than the load effect Sl . The probability of failure p f can be stated as
p f P( R Sl ) (2.39)
Or
p f P[G( R, Sl ) 0] (2.40)
where G () is defined as ‘limit state function’ and the probability of failure is the
same probability when limit state function is less than zero. The density function of R
and Sl can be combined and described by the joint density function f RSl (r , sl ) .
Therefore, based on Equations 2.39 and 2.40, the probability of failure can be
expressed as (Melchers, 1999):
where D0 is the failure domain. R and Sl, in most cases, are independent. Therefore,
Equation 2.41 can be expressed as follows (Melchers, 1999):
sl r
p f P( R Sl 0) f RSl (r , sl )drdsl (2.42)
60
p f P( R Sl 0) FR ( x) f Sl ( x)dx (2.43)
is less than some value , which presents the failure. f Sl ( x) represents the
probability that the load effect on the structural component (or system) has a value
between and x x in the limit as x 0 . The probability of failure can be
geometrically seen as the shaded margin in Figure 2-17 (Melchers, 1999).
reliability 1 p f (2.44)
In general, the structural resistance R and the load effect Sl vary with time due to
corrosion and the variation in loading conditions. Therefore, the time-dependent
reliability theory needs to be involved in determining changes in the probability of
failure versus time and, subsequently, to determine the service life of a structural
component (or system). In time-dependent reliability theory, all or some of the basic
variables are considered as a random function of time (the stochastic process). The
probability of failure that determined based on Equation 2.40 also changes with time.
61
Time-dependent reliability theory geometrically is shown in Figure 2-18 and
expressed as follows (Li & Melchers, 1993; Melchers, 1999):
In the time-integrated method, the whole lifetime [0, tL ] of the structural component
(or system) is considered as a unit. The resistance R is assumed to be constant
through the whole lifetime. The probability of failure during the whole lifetime (
p f (tL ) ) can be determined as follows (Melchers, 1999):
where Sl max (tL ) denotes the maximum load effect in the period [0, tL ] .
62
2.6.2.2 Discretised Method
In this method, the lifetime is discretised into a number of units nL . Each unit nL is
time independent and the probability of failure for each unit is determined similarly
as that in time-integration method through Equation 2.46 (Melchers, 1999).
In this method, the lifetime is discretised into a number of units, nL . Each unit is
time independent and the probability of failure for each unit is determined based on
Monte Carlo simulations (Melchers, 1999; Mahmoodian & Alani, 2013). Details are
summarised in Section 2.6.3.
In this method, the limit state function is modelled as a random function of time. The
safety index (t ) is at time t is determined as follows (Melchers, 1999):
G (t )
(t ) (2.47)
G (t )
where G (t ) and G (t ) is the mean and standard deviation of limit state function
p f (t ) [ (t )] (2.48)
In this method, the probability of failure is determined based on the first passage
probability when the load effect, Sl (t ) , up-crossing the resistance, R(t ) (Li &
Melchers, 1993; Melchers, 1999). Details of the outcrossing method are summarised
in Section 2.6.4.
Among these methods, the outcrossing method considers the auto-correlation among
resistance variations in time through cross-covariance function (Li & Melchers,
63
1993). The method gains its advantages among others since there is a high auto-
correlation among the structural resistance at each time due to the effect of corrosion
(Ni et al., 2010). The probability of failure and service life of the viaduct can be
overestimated without considering this auto-correlation (Li & Melchers, 2005). In
the out-crossing method, the statistics of R(t ) and Sl (t ) are obtained using Monte
Carlo simulation. The basics of Monte Carlo simulation and out-crossing method are
summarised below.
In this method, the lifetime is discretised into a number of units, nL . Each unit is
time independent and the probability of failure for each unit is determined based on
Monte Carlo simulations (Mahadevan & Dey, 1997; Melchers, 1999).
num[G ( xi ) 0]
pf (2.49)
Num
where Num is the number of trials and num the number of trials leading to G( xi ) 0 .
Sample generation and p f calculation could be achieved by Matlab code (Rubinstein
Random variables are generated by using ‘inverse transform’ method. For the basic
variable X i with accumulative distribution function FX i ( xi ) , a uniformly distributed
xi xi FX i 1 (ri ) (2.50)
64
The number of trials Num needed for a given confidence level Con in the failure
probability must satisfy the criteria as follows (Melchers, 1999):
ln(1 Con )
Num (2.51)
pf
Since the failure events for each time interval are independent, the probability of
failure over a given period p f (0, tn ) can be predicted as follows (Stewart, 2009):
K
p f (0, tn ) 1 [1 p f (nti )] (2.52)
i 1
where p f (0, tn ) is the probability of failure for time interval nti and K is the total
In the outcrossing method, the failure of the structural component (or system)
depends on the time that is expected to elapse before the first occurrence of the
action process Sl (t ) up-crossing an acceptable limit R(t ) sometime during its
service life. Equivalently, the probability of the first occurrence of such an excursion
is the probability of the structural component (or system) failure ( p f (t ) ) during that
period. This is known as ‘first passage probability’ and under the assumption of
Poisson processes it can be expressed as follows (Li & Melchers, 1993):
65
t
vdt
p f (t ) 1 [1 p f (0)]e 0
(2.53)
where p f (0) is the probability of pipe failure at time t = 0 and υ is the mean rate for
the process Sl (t ) to up-cross the threshold R(t ) . Based on Li and Melchers (2005),
the mean up-crossing rate is very close to zero in practical problems. Thus, Equation
2.53 can be rewritten as follows:
t
p f (t ) p f (0) vdt (2.54)
0
The up-crossing rate in Equation 2.54 can be determined by Rice formula (Rice,
1944) as follows:
. . .
v vR ( Sl R) f . ( R, Sl )dR (2.55)
Sl Sl
Sl
where vR is the up-crossing rate of the stochastic process Sl (t ) relative to the
. .
threshold R , R is the slope of R with respect to time, Sl is the time derivative
.
process of Sl and f . is the joint probability density function for Sl and Sl . To
Sl Sl
the structural component (or system) fails. This can be determined as:
p f (Tc ) Pa (2.56)
where Tc denotes the service life based on the criterion employed and Pa is the
acceptable risk and can be determined from AS 5100.1 (Australian Standard, 2007a)
for any structure in Australia.
66
2.6.5 System Reliability Analysis
For the series system, the failure of any elements (or failure modes) can lead to the
failure of the whole system. A statically determinate structure (i.e., viaduct girder
simply supported at the two ends by piers) is a typical example of the series system.
The block diagrams for series system are shown in Figure 2-19 (Melchers, 1999;
Lisnianski, Frenkel & Ding, 2010).
p f P( F1 F2 F3 ... Fm ) (2.57)
where Fi is the defined as the failure of element or failure mode i and P() is the
between the failure of each element or failure mode, namely Fi and F j . Considering
the reliability bonds for the probability of failure in the series system can be
determined as follow (Melchers, 1999):
67
m m
max{P( Fi )} p f 1 [1 P( Fi )] (2.58)
i 1
i 1
For the parallel system, the failure of one or more element (or failure mode) does not
necessarily indicate the failure of the whole system. A statically in-determinate
structure (i.e., viaduct girder simply fixed at the two ends by piers) is a typical
example of the parallel system (Melchers, 1999). The block diagram for the parallel
system is shown in Figure 2-20.
The probability of failure of the parallel system composed of m members (or failure
modes) is shown as follows (Melchers, 1999; Lisnianski et al., 2010):
p f P( F1 F2 F3 ... Fm ) (2.59)
The reliability bonds for the probability of failure in the parallel system can be
determined as follows:
m m
P( F ) p
i 1
i f min{P( Fi )}
i 1
(2.60)
The series and parallel system can be combined to form a mixed system. The
elements (or failure modes) are connected in series or parallel arrangement to
perform the operation of the system. The block diagram for a series system is shown
68
in Figure 2-21. The mixed system can be sequentially reduced into series or parallel
systems to determine the upper and lower bonds of the probability of failure (see
Melchers, 1999).
2.7 Summary
This chapter, through a review of the existing literature, summarised the basics of
structural analysis of a steel viaduct, corrosion rate and corrosion effect on mild steel
properties and time-dependent reliability analysis to estimate the service life of a
viaduct. Different aspects of designing experiments to detect the corrosion effect on
mechanical properties, microstructural features of mild steel, failure criteria for
viaduct assessment and methods to predict the service life of steel viaducts through
reliability theory were also reviewed. Several limitations and/or gaps in the existing
literature were found:
69
Lack of research to determine the degradation mechanism of mechanical
properties of mild steel due to corrosion, in both macro and micro levels
Lack of knowledge on the effect of elastic stress on corrosion rate and the
combined effect of stress and corrosion on the mechanical properties of mild
steel
Lack of research monitoring the level of increase of hydrogen concentration,
element composition and the changes in microstructural features of steel (i.e.,
grain size and steel phase) with or without stress during corrosion
Lack of knowledge on the mechanism of corrosion-induced delamination for
continuously casted steel and the effect of stress on delamination
In time-dependent reliability analysis of steel viaducts subjected to combined
corrosion and fatigue environment, very few studies consider stress range and S-
N curve as time variant, and none treat the damage accumulation index as a
stochastic process
Lack of research using the out-crossing method to predict the service life of a
steel viaduct based on multiple failure criteria and system reliability analysis.
This chapter highlighted these shortcomings in the existing literature and positioned
this thesis to address them.
70
Chapter 3: Experimental Research on Corrosion
3.1 Introduction
Per the literature review in Chapter 2, mild steel—steel with carbon content of 0.15–
0.20%—is widely used in the construction of bridges. Steel corrosion is the biggest
environmental threat faced by structures. Corrosion can lead to losses of cross-
sections and reduction in yield strength, ultimate strength and ductility. In most
cases, structural steel is simultaneously stressed under loading and subjected to
corrosion (Wang et al., 2014). Therefore, it is vital to monitor and quantify the
behaviour of mild steel under combined stress and corrosion environment to optimise
maintenance, repair and rehabilitation of steel structures.
Chapter 2 showed there are conflicting views on the effect of stress on corrosion rate.
Therefore, this needs further investigation, which is presented in this chapter through
a comprehensive lab experiment and field study. Chapter 2 also showed corrosion
can change the microstructural features (element composition, grain size, steel phase
and morphology) of steel. These changes can be more dramatic for stressed steel than
non-stressed steel due to more severe hydrogen embrittlement. Therefore, it is vital
to investigate the changes in hydrogen concentration and microstructure for stressed
and non-stressed steel during corrosion. This chapter presents a comprehensive lab
experiment and field study to investigate corrosion behaviour of steel in stressed and
non-stressed cases. Simulated corrosion tests were conducted by immersing stressed
and non-stressed steel in acidic solutions in the lab. Corrosion losses of specimens
after each stage of immersion were measured. The changes in hydrogen
concentration and microstructural features were measured and compared for stressed
and non-stressed steel during corrosion. From this, the relationship between
corrosion loss and hydrogen concentration, grain size, element composition and
phase composition were determined. To correlate the relationship developed in the
lab to that of natural corrosion (i.e., field corrosion), a comprehensive field
investigation was conducted to measure the corrosion loss of steel at different
exposure times to atmosphere. Also, samples at different levels of corrosion were cut
from the corroded steel to investigate the changes in microstructural features during
corrosion.
71
3.2 Corrosion Tests
Simulated corrosion tests for non-stressed steel are designed to achieve two aims—1)
estimating changes in hydrogen concentration, element composition and
microstructure during corrosion; and 2) quantifying changes in tensile properties
during corrosion. The first aim is discussed in this chapter, while the second is
discussed in Chapter 4.
Simulated corrosion tests for stressed steel are designed to achieve three aims—3)
estimating the effect of elastic stress on corrosion loss, 4) investigating the combined
effect of stress and corrosion on the changes in hydrogen concentration and
microstructure, and 5) investigating the combined effect of stress and corrosion on
tensile properties. Aims 3 and 4 are discussed in this chapter, while Aim 5 is
discussed in Chapter 4.
G250 mild steel—a common material used in building and bridge construction—was
used as the test material. The chemical composition of steel specimens used in this
research is shown in Table 3-1.
Two series of steel specimens were designed for non-stressed steel—specimens for
corrosion measurement and tensile test (see Figure 3-1) following ASTM E8/E8M-
16a (ASTM, 2016), and specimens for hydrogen concentration test (see Figure 3-2)
following ASTM F1113-87 (ASTM, 2017a). The tensile test specimens were
wrapped with acid-resistant tape at both ends so that corrosion only took place in the
tested area (i.e., 50 mm gauge length in the middle). The specimen shown in Figure
3-2 (90×14×6 mm) was fully immersed in acid solution for testing at a designated
time after corrosion. It was then cut into sample 1 (36×14×6 mm) for element
72
composition analysis and samples 2, 3 and 4 (14×6×4 mm) for grain size and phase
analysis (see Figure 3-2).
For stressed specimens, the sets were identical to the non-stressed specimens (in
Figure 3-1). However, to fit the stressed system, two holes were drilled on the
specimens at both ends (see Figure 3-3). The specimens were also wrapped with
acid-resistant tape at both ends. There were four identical specimens for each
measurement. Of these four specimens, three were used for corrosion measurement
and tensile test and one for hydrogen concentration and microstructural analysis in
which four samples (samples 1, 2, 3 and 4) were cut and used (see Figure 3-3).
73
Notes. Units are in millimetres.
The steel specimens were stressed up to 70% of their yield strength (220 MPa),
which was determined based on the ratio between serviceability load and ultimate
load as per AS 5100.2 (Australian Standard, 2017b). Specimens were stressed
through a tailor-made clamp and jig system (see Figure 3-4a). The system was
composed of two 1800×30×26 mm Grade 904 stainless steel plates and four M16
high-strength steel bolts. Specimens were pulled on tensile testing machine and then
nuts were fixed to maintain the stress (see Figure 3-4b). The clamps and jigs were
wrapped with acid-resistant tapes and then coated with vaseline (see Figure 3-5c) to
prevent their corrosion. The distance between the stainless steel clamps was
constantly measured during immersion test to prevent stress loss.
Overall, the total number of specimens and samples (stressed and non-stressed) were
75 and 100 respectively. The test plan for the research including the number of
specimens and immersion time is presented in Table 3-2.
74
(days) Non-stressed Stressed Non-stressed Stressed
Before immersion 0 3 4
(control specimens)
0.00001 M HCl 7 3 3 4 4
(pH=5) 14 3 3 4 4
28 3 3 4 4
0.003 M HCl 7 3 3 4 4
(pH=2.5) 14 3 3 4 4
28 3 3 4 4
1 M HCl 7 3 3 4 4
(pH=0) 14 3 3 4 4
28 3 3 4 4
3 M HCl 7 3 3 4 4
(pH=–0.5) 14 3 3 4 4
28 3 3 4 4
Total 75 100
Notes. HCl = Hydrochloride acid, M = Molar.
Corrosion simulation was conducted through acid immersion following ASTM G31-
72 (ASTM, 2004). Test set-up is shown in Figure 3-5. Specimens were removed
from the solutions after 7, 14 and 28 days for the tests and measurements (per the
plan in Table 3-2). Hydrochloride acid (HCl) was selected as the immersion solution,
consistent with previous studies (Noor & Al-Moubaraki, 2008). Four HCl solutions
at different pHs were chosen for the corrosion simulation (see Table 3-2). The reason
for using an acidic solution was accelerating corrosion, thus, enabling collection of
adequate data to develop the relationship between corrosion loss, tensile properties
and microstructural features of steel within a given timeframe. The range of acidity
selected in immersion tests, even for a high concentration of acid with very low pH
(i.e., 1 Molar [M] HCl and 3 M HCl) are found in the real world. For example, the
solution with 0.00001 M HCl (pH=5) can simulate the natural corrosion of steel
exposed to an environment containing a large amount of organic matter (e.g., steel
buried in soil) (Liu et al., 2014). The solution with 0.003 M HCl (pH=2.5) can
simulate steel under long-term exposure to acid rain (Wang et al., 2006) .The
75
solution with 1 M HCl (pH=0) and 3 M HCl (pH=-0.5) can simulate steel wells in oil
fields subjected to chemical cleaning (Finšgar & Jackson, 2014).
(a) Non-stressed specimens (Figure 3-1) (b) Non-stressed specimens (Figure 3-2)
76
m m0 m1 (3.1)
where m0 is the average mass of the tested area before corrosion (mg) and m1 is the
average mass of the tested area measured after each immersion period.
To make sure the process of sample cleaning did not affect the test specimens and
only removed rust, hydrogen concentration and microstructural features were
measured on three trial samples before and after cleaning. The details of
measurements are shown in Sections 3.3.2 and 3.3.3—there were no noticeable
changes in hydrogen concentration and microstructural features before and after the
cleaning process (i.e., the process did not affect the specimens).
Based on Equation 3.2, corrosion loss for each immersion period was determined as:
where C is the corrosion loss (mm), m is the weight loss of the specimens in
milligrams (mg), st is the metal density in g/cm3, which is 7.85 g/cm3 for mild
steel, Aarea is area of the specimens exposed in acid in cm2 (i.e., the tested area of the
specimens).
For the measurements, steel specimens were exposed to 0.2 M sodium hydroxide
(NaOH) solutions restored in a Teflon cell. Hydrogen atoms within steel react
with hydroxide solutions. The current of reaction was recorded after 30 minutes
77
to calculate hydrogen concentration. The hydrogen concentration can be
computed as follows:
Df
I P F[ H ] (3.3)
t
78
Figure 3-7: Sample after Polishing
XRF measurement was conducted to detect the element compositions of each sample
with and without stress. XRF measurement was conducted on sample 1 after each
stage of immersion. Samples were further cleaned with acetone and dried with air.
They were then placed in the sample holder (see Figure 3-8a) and loaded in Bruker
Axs S4 Pioneer XRF equipment (see Figure 3-8b). The x-ray tube of the
equipment operated at a potential of between 10 and 100 kV. The element
compositions of the samples are determined and shown on the screen of the XRF
equipment.
79
3.3.3.2 Grain Size Measurement
Grain size was quantified using OM (see Figure 3-9) at 100× magnification.
Measurement was conducted on samples 2, 3 and 4 to determine average grain size
at each stage of immersion. All samples were polished and etched. Measurement
was conducted at the RMIT Microscopy and Microanalysis Facility through linear
intercept procedure (ASTM, 2013a) using ImageJ (a software to edit and analyse
images).
80
intersections
In Figure 3-10, a point where a straight line is cut by a grain boundary is called an
intersection. The numbers of intersections were counted and the grain size ( LL ) for
each sample was determined as follows:
LT
LL (3.4)
Ptn M A
where LL is the grain size, LT is the total length of straight lines drawn on the
images (total length of the six lines in Figure 3-10), M A is the magnification
Steel phase analysis was performed using EBSD scanning. The measurement was
undertaken with FEI Nova NanoSEM and Oxford Instruments Aztec software suite
at the RMIT Microscopy and Microanalysis Facility (see Figure 3-11). Scanning
was conducted on samples 1, 2 and 3 to determine average phase composition. Each
polished sample was loaded into FEI Nova NanoSEM to acquire its clear image at
1000× magnification. Afterwards, the EBSD detector was inserted to carry out
phase analysis.
81
Figure 3-11: FEI Nova NanoSEM
Mild steel contains two major phases—ferrite (α-Fe) and pearlite (Revie, 2008).
However, since pearlite has a two-phase structure composed of ferrite (α- ) and
cementite ( ), its proportion cannot be measured by EBSD directly. Therefore,
the two phases selected for EBSD analysis were ferrite (α- ) and cementite ( ).
The EBSD detector was operated at 20 kV voltages and the step size was 0.2 µm to
obtain accurate phase composition results (Man, Pantělejev & Pešina, 2009). The
scanned area was approximately 50×25 µm next to steel/solution interface (Man et
al., 2009).
SEM analysis was conducted on the corroded surface of sample 1 for stressed and
non-stressed steel to study the formation of corrosion pits and micro-cracks during
corrosion. At the end of element composition measurement, sample 1 was prepared
according to ASTM E2809-13 (ASTM, 2013c) and scanned under a Philips XL30
Scanning Electron Microscope (see Figure 3-12) at 80× magnification for pits
observation and 1000× magnification for cracks observation.
82
Figure 3-12: Philips XL30 Scanning Electron Microscope
During measurement, the Philips XL30 Scanning Electron Microscope was operated
at the voltage of 30 kV and spot size of 3.0. The images were exported and the width
and depth of corrosion pits measured through ImageJ.
The corrosion of the tested areas of stressed and non-stressed specimens after 28
days immersion are shown in Figure 3-13. More rusts, spalls and pits can be seen in
the stressed specimens. It is clear that elastic stress can accelerate corrosion and,
subsequently, there are more corrosion products (rusts) formed for stressed steel than
non-stressed steel at the same immersion time. The acceleration effect of stress on
corrosion has also been reported by Ren et al. (2012) and Xu and Cheng (2012).
However, the morphology of corroded sample surface has not been shown and
compared between stressed and non-stressed samples. Stress can cause the rupture of
the passive oxide films. Localised corrosion occurs at the film rupture site, which
leads to the formation of rust and the initiation of spalls and pits for stressed steel.
However, further research is needed to monitor the condition of passive oxide films
for stressed and non-stressed steel during corrosion.
83
(a) Un-corroded steel
Corrosion progress was measured physically by corrosion loss over time (see Figure
3-14) for stressed and non-stressed specimens. It can be seen that corrosion loss
increases more rapidly with the reduction of pH. Also, corrosion loss is larger for
84
stressed steel than non-stressed steel at the same immersion time for all four
solutions in Figure 3-14. The difference in corrosion loss between stressed and non-
stressed steel also increases with the increase of the immersion time. After 7 days,
the corrosion loss for stressed steel is 1.10, 1.70, 1.77 and 1.25 times that of non-
stressed steel in 0.00001 M HCl (pH=5), 0.003 M HCl (pH=2.5), 1 M HCl (pH=0)
and 3 M HCl (pH=-0.5) respectively. However, after 28 days, the corrosion loss for
stressed steel is 1.50, 1.85, 1.87 and 1.34 times that of non-stressed steel in these four
solutions.
Clearly, stress has increased the corrosion activity significantly and consistently for a
range of corrosive environments. Corrosion is an electrochemical process and
corrosion rate increases with the decreases of electrochemical potential (Gutman,
1989; Revie, 2008; Ren et al., 2012). Stress can reduce the electrochemical potential
by causing the rupture of passive oxide film, increase of surface energy and
dislocation among grain boundaries (Ren et al., 2012).
0.2
0.00001M HCl stessed
0.00001M HCl non-stressed
0.16
Corrosion loss (mm)
0.08
C = 0.0021t + 0.0034
R² = 0.9848
0.04
C = 0.0013t + 0.0016
0 R² = 0.9834
0 5 10 15 20 25 30
Time (Days)
85
0.2
0.03M HCl stessed
0.03M HCl non-stressed
0.16 Linear (0.03M HCl stessed)
0.12
C= 0.0045t + 0.0082
R² = 0.9739
0.08
0
0 5 10 15 20 25 30
Time (Days)
2 1M HCl stessed
1M HCl non-stressed
Linear (1M HCl stessed)
1.6 Linear (1M HCl non-stressed)
Corrosion loss (mm)
1.2
0.4
C = 0.0163t - 0.0314
R² = 0.9538
0
0 5 10 15 20 25 30
Time (Days)
(c) 1 M HCl
86
2
3M HCl stessed
3M HCl non-stressed
1.6 Linear (3M HCl stessed)
1.2
C = 0.065t - 0.0265
R² = 0.9977
0.8
C = 0.05t - 0.1306
R² = 0.9528
0.4
0
0 10 20 30
Time (Days)
(d) 3 M HCl
Table 3-3: Relationship between Corrosion Loss and Time (Lab Specimens)
Corrosive Stress Equation R² p
solution condition
0.00001 M HCl stressed C 0.0021t 0.0034 0.9848 1×10–9
(pH=5)
87
3 M HCl (pH=- stressed C 0.0650t 0.0265 0.9977 1×10–10
0.5)
Figure 3-15 shows the relationship between corrosion loss and hydrogen
concentration for stressed and non-stressed specimens in all four HCl solutions.
Higher hydrogen concentrations were observed in stressed samples compared to non-
stressed samples. For non-stressed specimens, hydrogen concentration increases
from 0 ppm to 5.97 ppm when corrosion loss reaches 1.36 mm. For stressed
specimens, hydrogen concentration increases to 12.10 ppm when corrosion loss
reaches 1.82 mm. The difference in hydrogen concentration between stressed and
non-stressed steel increases with the increase of corrosion loss. For example, at
corrosion loss of 0.5 mm, hydrogen concentration for stressed steel is 1.01 times that
of non-stressed steel. At corrosion loss of 1.0 mm, hydrogen concentration for
stressed steel is 1.50 times that of non-stressed steel.
25
stessed
Hydrogen concentration (ppm)
non-stressed
20 Poly. (stessed)
Poly. (non-stressed)
15
[H] = -4.7607C2 + 14.981C + 0.4285
R² = 0.9686
10
5
[H] = -7.9825C2 + 14.963C + 0.3538
R² = 0.8795
0
0 0.5 1 1.5 2
Corrosion loss (mm)
The relationships between hydrogen concentration and corrosion loss for stressed
and non-stressed steel can be obtained by multiple regressions in Excel 2016 (see
88
Table 3-4). The models are reasonably fit for experimental data (R²>0.85 and
p<0.05).
There are two explanations for the higher hydrogen concentration in stressed
specimens. Corrosion is a combination of oxygen reduction and hydrogen evolution
reaction (Revie, 2008). Hydrogen gas is released in hydrogen evolution reaction,
which can diffuse in steel and cause the increase of hydrogen concentration.
Hydrogen evolution reaction plays a more dominant role in the corrosion reaction at
pits and cracks. There are more pits and cracks formed on the surface of stressed
steel, which subsequently enhances hydrogen evolution reaction and creates more
hydrogen release and diffusion into steel (Eggum, 2013). Hydrogen atoms in the acid
solutions can diffuse into steel and, subsequently, increase hydrogen concentration.
Stress contributes to the diffusion of hydrogen atoms by increasing the surface
energy. Stress also creates more dislocations and voids inside steel that traps
hydrogen atoms (Eggum, 2013). This thesis is the first study to undertake
comparison of the hydrogen concentration of stressed steel and non-stressed steel at
the same degree of corrosion and develop models to predict the hydrogen
concentration for stressed and non-stressed steel.
Element composition of corroded steel over time for stressed and non-stressed
specimens are shown in Figure 3-16. It can be seen that iron content is 93.01% and
oxygen content is 5.92% before corrosion, which is different from that in Table 3-1
(provided by manufacturer). This is because the formation of passive oxide film at
steel surface reduces the proportion of iron and increases the proportion of oxygen.
89
Iron content reduces from 93.01 to 48.36% and oxygen proportion increases from
5.92 to 47.26% after 28 days corrosion in 3 M HCl (pH=-0.5) solution with corrosion
loss from 0 to 1.82 mm. Figure 3-16 shows that, for stressed and non-stressed
samples, the proportion of iron reduces and the proportion of oxygen increases with
as corrosion increases. The reduction of iron proportion is due to the reaction
between iron and acid and the increase of oxygen proportion is due to the formation
of corrosion products. Iron attributes to the ductility of steel while corrosion products
containing oxygen make steel brittle. Therefore, the reduction of iron proportion and
increase of oxygen proportion during corrosion lead to the reduction in steel
ductility.
100
80
Fe% = 21.896C2 - 56.526C + 83.508
Iron (%)
60 R² = 0.7368
40
stressed
non-stressed
20 Poly. (stressed)
Poly. (non-stressed)
Poly. (Combined)
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(a) Iron proportion vs corrosion loss
90
100
stressed
non-stressed
80 Poly. (stressed)
Poly. (non-stressed)
Poly. (Combined)
Oxygen (%) 60
40
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(b) Oxygen proportion vs corrosion loss
Table 3-5: Relationship between Element Composition and Corrosion Loss (Lab
Specimens)
Stress condition Equation R² p
stressed Fe% 21.896C 2 56.526C 83.508 0.7368 8×10–4
non-stressed O% 19.918C 2 52.974C 13.497 0.7850 2×10–4
Although the literature suggests that stress initiates cracks and contributes to the
diffusion of oxygen into steel (Zhou, 2010), oxygen proportion does not become
remarkably higher for stressed steel than for non-stressed steel at the same corrosion
degree (corrosion loss). This may be because the diffusion coefficient (1.2×10–2
µm2/s) of oxygen in steel is very low (Yi & Lin, 1990). Even for stressed steel, there
is very limited oxygen diffused into steel during corrosion. The reaction between iron
91
and acid (leading to the reduction of iron proportion) and the formation of corrosion
products (leading to the increase of oxygen proportion) mainly occur at the surface
for stressed and non-stressed steel (Noor & Al-Moubaraki, 2008). Therefore, the
changes in iron and oxygen proportion are, in general, the same for stressed and non-
stressed steel at the same corrosion degree. This thesis is the first study to compare
the level of changes in element composition between stressed and non-stressed steel
and develop models to predict the iron and oxygen proportion based on corrosion
loss.
92
(b) Non-stressed (c) Stressed
Figure 3-17: Change in Grain Size after 28 Days Corrosion
Figure 3-18 shows grain size versus corrosion loss. It can be seen that the average
grain size for stressed and non-stressed steel reduces with corrosion loss increment.
Overall, the level of reduction is higher for stressed steel than for non-stressed steel
at the same immersion time. The difference in grain size between stressed and non-
stressed steel increases with the increase of corrosion loss. For example, at 0.5
mm of corrosion loss, the percentage of grain size reduction for stressed steel is 1.02
times that of non-stressed steel. At 1.0 mm of corrosion loss, the percentage of grain
size reduction for stressed steel is 1.42 times that of non-stressed steel.
16
stressed
14 non-stressed
Poly. (stressed)
12 Poly. (non-stressed)
Grain size (μm)
10
gz = 8.0508C2 - 13.918C + 10.688
8 R² = 0.704
6
4
2 gz= 5.2076C2 - 12.055C + 10.182
R² = 0.7888
0
0 0.5 1 1.5 2
Corrosion loss (mm)
Figure 3-18: Grain Size vs Corrosion Loss
93
The relationships between grain size and corrosion loss can be obtained by multiple
regression (see Table 3-6). The models are reasonably fit for experimental data
(R2>0.70 and p<0.05).
Table 3-6: Relationship between Grain Size and Corrosion Loss (Lab
Specimens)
Stress condition Equation R² p
stressed gz 5.2076C 2 12.055C 10.182 0.7888 1×10–4
There is a higher level of reduction in grain size in stressed steel compare to that in
non-stressed steel at the same degree of corrosion loss. This is because IGSCC leads
to a higher level of reduction in grain size for stressed steel than in non-stressed
steel (Huang et al., 2002; Ralston & Birbilis, 2010). This thesis is the first study
to undertake monitoring and quantification of the reduction of grain size for
stressed and non-stressed steel during corrosion.
Figure 3-19 presents the changes in phase composition of steel samples with
corrosion loss (3 M HCl solution as representative). It can be seen that ferrite
proportion is around 85% and cementite proportion is around 2% with corrosion loss
increasing. There were no dramatic changes in phase composition for stressed and
non-stressed steel during corrosion.
94
100
90
80
70 Ferrite stressed
Phase (%)
60 Ferrite non-stressed
50 Cementite stressed
Cemenitie non-stressed
40
Others stressed
30 Others non-stressed
20
10
0
0 0.5 1 1.5 2
Corrosion loss (mm)
Although the corrosion resistance of cementite is larger than that of ferrite, corrosion
mainly occurs at the boundaries of ferrite grains where cementite particles are
located (Chisholm et al., 2016). As a result, cementite can easily be washed away by
solutions. The composition of other phases, including graphite, austenitic and
impurities within steel, can also be washed away by solutions since they are
located at the boundaries of ferrite grains (Chisholm et al., 2016). Consequently, the
level of reduction of ferrite, cementite and others is similar and there are no
significant changes in their proportion. With the presence of stress, cementite is
more likely to be fractured than ferrite (Umemoto, Todaka & Tsuchiya, 2003).
However, the IGSCC also contributes to the corrosion of ferrite grains and other
phases being washed away by solutions earlier (Arioka et al., 2006). As a result,
the level of reduction of ferrite, cementite and others is still similar. This thesis is
the first study to monitor the changes in phase composition for stressed and non-
stressed steel during corrosion.
The largest corrosion pits observed close to the boundaries of samples 2, 3 and 4 are
shown in Figure 3-20 as representatives. The width and the depth of these corrosion
pits were measured through ImageJ. For non-stressed steel, the width of the
95
corrosion pits is 0.07 mm, 0.54 mm, 0.47 mm and 1.21 mm for 0.00001 M HCl
(pH=0), 0.003 M HCl (pH=2.5), 1 M HCl (pH=0) and 3 M HCl (pH=-0.5)
respectively. The depth of corrosion pits is 0.11 mm, 1.17 mm, 1.73 mm and 2.45
mm respectively. For stressed steel, the width of the corrosion pits is 0.32 mm, 0.79
mm, 0.98 mm and 0.97 mm for 0.00001 M HCl (pH=0), 0.003 M HCl (pH=2.5), 1
M HCl (pH=0) and 3 M HCl (pH=-0.5) respectively. The depth of corrosion pits
is 1.63 mm, 2.30 mm, 2.74 mm and 2.85 mm respectively
Corrosion pit
Corrosion pit
96
Corrosion pit Corrosion pit
Corrosion pit
Corrosion pit
Figure 3-21 shows micro-cracks detected in steel after corrosion. The cracks are
formed next to corrosion pits where there are residual stress concentrations. The
blisters discovered next to cracks are signs of inner pressure increment due to the
accumulation of molecular hydrogen within steel (Li et al., 2018). Additionally, there
are more cracks in stressed steel than in non-stressed after immersion. These findings
indicate a stress and corrosion combined environment contributes to the initiation
and propagation of pits and cracks, as the hydrogen concentration is higher for
stressed steel than non-stressed steel at the same degree of corrosion.
97
(a) Un-corroded samples
98
(d) 1 M HCl (non-stressed left, stressed right)
3.4.5 Discussion
The following sections discuss the observations from the test results.
99
3.4.5.1 Effect of Stress on Corrosion
The effect of stress on corrosion derived in this thesis is compared with that derived
by Ren et al. (2012) and Wang et al. (2014) (see Table 3-7). It is expected that the
corrosion loss for steel stressed up to 70% of yield strength is 1.44 times that of non-
stressed steel (Ren et al., 2012), and 1.08 and 1.07 times that of non-stressed steel
(Wang et al., 2014). However, in the present research the corrosion loss for stressed
steel is 1.50, 1.85, 1.87 and 1.34 times that for of non-stressed steel in 0.00001 M,
0.03 M, 1 M and 3 M HCl respectively after 28 days.
Table 3-7: Ratio of Corrosion Loss for Stressed and Non-Stressed Steel at Same
Corrosion Timea
Corrosive condition Steel grade mstressed Reference
mNon stressed
3.5 μA/cm2 stray current X80 pipeline 1.07 Wang et al. (2014)
and simulated soil steel
solutions
0.00001 M HCl G250 mild steel 1.50 This thesis
0.003 M HCl G250 mild steel 1.85 This thesis
1 M HCl G250 mild steel 1.87 This thesis
3 M HCl G250 mild steel 1.34 This thesis
At the same corrosion time, the difference in corrosion loss between stressed and
non-stressed steel in this research (0.00001 M, 0.03 M and 1 M HCl) is, in general,
larger than that in Ren et al. (2012) and Wang et al. (2014). This is because more
hydrogen atoms diffuse into steel by immersing steel in HCl, which affects the
stability of passive oxide film, initiating cracks and increasing corrosion loss.
However, in very aggressive solution (e.g., 3 M HCl), the reaction between steel and
solution occurs rapidly for stressed and non-stressed steel. Therefore the related
corrosion loss difference is lower than that of the other three solutions (Revie, 2008).
Also, although all studies listed in Table 3-7 used mild steel, phase composition and
100
element composition vary for different grades of mild steel which affects the
diffusion coefficient of hydrogen and causes the different level of acceleration effect
of stress on corrosion.
The level of increase in corrosion loss determined in this chapter is compared with
that determined by Gutman’s assumptions (Equation 2.9 in Section 2.3.4). Under the
applied stress of 220 MPa, is –7.98 mV. Ren et al. (2012) monitored corrosion
current density under different electrochemical potentials. An experimental
relationship between electrochemical potential ( ) and corrosion current density ( I P
) for mild steel can be determined through regression analysis, which is shown as
follows (Ren et al., 2012):
According to Equation 3.5, the corrosion current density increases by 0.82 µA/cm2
when electrochemical potential reduces by 7.98 mV. The differences in corrosion
loss can be determined by the changes in current density, with equations shown as
follows:
0.001cr t
C (3.7)
st
where cr is the change in corrosion rate ( g/m2 day ), I P is the change in
After 28 days, corrosion loss increases by 0.0007 mm and corrosion current density
increases by 0.82 µA/cm2. However, in this thesis, corrosion loss increases by 0.03
mm, 0.06 mm, 0.40 mm and 0.45 mm in 0.00001 M, 0.03 M, 1 M and 3 M HCl
respectively when steel is stressed after 28 days of immersion (see Figure 3-22). The
level of corrosion loss increment determined in this thesis is higher than that
101
determined based on Gutman’s assumption. This is because the stress concentration
at corrosion pits and cracks makes the stress the steel subjected to larger than the
applied stress. Also, for steel immersed in HCl, Chloride atoms can break the passive
film formed during corrosion and lead to the formation of pits and cracks during
corrosion (Revie, 2008).
0.8
3M HCl
0.7 1M HCl
0.003M HCl
Corrosion loss (mm)
As mentioned in Section 2.4.4, hydrogen effect on steel has been widely estimated
through hydrogen charging. Hydrogen charging can underestimate the level of
hydrogen embrittlement (Banerjee & Chatterjee, 2001; Eggum, 2013). This
conclusion was supported by comparing the growth of hydrogen concentration
between the immersion test in this study and the hydrogen charging test conducted
by Eggum (2013). The maximum hydrogen concentration achieved by charging is
around 1.84 ppm. However, in the current study, the hydrogen concentration after 28
days immersion reached 5.94 ppm for non-stressed steel and 12.10 ppm for stressed
steel. There are three explanations. First, hydrogen gas and atoms mainly reside at
voids and defects within steel. Corrosion creates more defects and pits that contribute
to the residence of hydrogen atoms. However, steel undergoes passive corrosion in
charging test and there are limited defects and pits initiated during charging. Second,
compared with the charging test, the chloride atoms in the solution affect the stability
of passive films formed before and during corrosion by immersing steel into the HCl
102
solutions. More hydrogen diffuses and is absorbed into the steel during immersion
test (Revie, 2008). Third, the reduction of grain size during immersion test weakens
the bonding stress between steel grains. This contributes to hydrogen accumulation
and initiation of hydrogen-induced cracking (Finšgar & Jackson, 2014). As such,
immersion of steel specimens in acid, as carried out in this thesis, is more a realistic
corrosion simulation and provides a more accurate hydrogen concentration and
corrosion loss relationship.
The test results raise the concern of protecting steel from intergranular corrosion and
IGSCC to prevent the subsequent loss of tensile properties. The results also indicate
that intergranular corrosion and IGSCC can occur for mild steel immersed in HCl. In
most previous studies, intergranular corrosion and IGSCC have mainly focused on
stainless steels (Shanmugam et al., 2007; Gonzaga, 2013; Sun et al., 2014; Zhou &
Yan, 2016), nickel base alloys (Turnbull, 2014) and aluminium/magnesium alloys
(Djukic, Zeravcic, Bakic, Sedmak & Rajicic, 2014) and mild steel in some particular
solutions (e.g., Clark solution and CO2+NaNO2 solution). No published studies have
reported the concern of intergranular corrosion and IGSCC for mild steel immersed
in HCl, which is a common corrosive environment mild steel is exposed to.
The applied stress can also lead to the redistribution of residual stress among steel
grains and, subsequently, affect the corrosion behaviour of steel (corrosion is more
severe at locations with high residual stress concentration). It is necessary to measure
103
the distribution of residual stress of steel under stress and corrosion combined
environment. Measurement can be performed through portable x-ray diffraction in-
situ or high resolution EBSD in the lab (Ma, Tschauner, Beckett, Rossman & Liu,
2012; Calcagnotto, Ponge, Demir & Raabe, 2010). Also, the corrosion resistance of
mild steel is related to its nanostructure (Aliofkhazraei, Rouhaghdam & Hassannejad,
2009). Therefore, to further understand the stress effect on corrosion it is imperative
to observe how the nanostructure is changed during corrosion for stressed and non-
stressed steel. This can be achieved by Raman spectroscopy technique and
transmission electron microscopy analysis using a Focused Ion Beam (Aliofkhazraei
et al., 2009).
Pits and cracks are formed during corrosion. The stress and corrosion combined
environment contributes to the initiation of pits and cracks. These pits and cracks can
be prevented by controlling pHs, chloride concentration and temperature in the
corrosive environment. Also, to make sure these control strategies are implemented
on a steel structure in time, the initiation and propagation of pits and cracks in
structural components subjected to corrosion needs to be monitored in-situ. In-situ
measurement can be achieved by using the direct current potential drop or alternating
current potential drop methods. In these methods, a constant or alternating current is
applied to the structural component and the change in potential drop used to
determine the dimension of pits and cracks.
Field tests were designed to 1) determine the correlation factors of corrosion rate
between atmosphere corrosion and the lab simulated corrosion; 2) develop the
relationship between corrosion loss and hydrogen concentration, grain size and
element composition for natural corrosion; 3) estimate the effect of field corrosion on
mechanical properties (hardness, yield strength, ultimate strength, failure strain and
fatigue strength limitation). The results of the first two aims are detailed in this
chapter, while the results of the third are discussed in Chapter 4.
104
3.5.1 Test Design
Field tests were conducted on two viaducts in Australia. Both are made of G250 mild
steel and located in the same corrosive zone as categorised by AS 4312 (Australian
Standard, 2008b). Both destructive and non-destructive tests were conducted.
The non-destructive test was conducted on a historical steel viaduct in Australia. The
viaduct is made up of two separated steel bridges, the northern and southern bridges.
The northern bridge was constructed in 1889 and the southern bridge was
constructed in 1919. The viaduct has 45 spans and site inspection was conducted on
Span 45 because visual inspection showed more corrosion there. There are five
girders in this span.
The non-destructive tests included ultrasonic thickness measurement (to estimate the
corrosion loss of the flanges and webs, discussed below) and hardness measurement
(to estimate the changes in tensile properties of steel, discussed in Section 4.2.1). The
test locations of ultrasonic thickness measurement are shown in Figure 3-23 and
Table 3-8. Photos of site measurement are shown in Figure 3-24.
105
Figure 3-24: Site Measurement
Ultrasonic instrument measures the transit time of the ultrasonic pulse through the
steel plates. The thickness of the steel plates can then be determined as follows:
Vv t
d (3.8)
2
where d is the material thickness, Vv is the velocity of sound in the material and t is
the transit time between the initial pulse and the backwall echo (or between
successive backwall echoes). The Vv values of materials are programmed so that the
material thickness can be read directly from the instrument screen.
106
Before the thickness measurement, the test locations were polished to remove paints,
scales, pits or other surface coatings (ASTM, 2012). The polished area was roughly
100×100 mm. The procedure of ultrasonic thickness measurement followed ASTM
E797/E797M−15 (ASTM, 2015c). Scanning was performed three times for each test
location and the thickness was determined based on the average value.
Specimens were cut from Girder 1 and Girder 2 of a demolished viaduct (see Figures
3-25 to 3-28). The cut area was 160×50 mm (see Figure 3-27). The thicknesses of the
specimens were measured by ultrasonic thickness device. The specimens were then
classified according to the level of corrosion loss (see Table 3-9).
107
Figure 3-27: Dimension of Specimens Cut from Viaduct
12 13
Median corrosion Top flange 12 14
At each level of corrosion, the specimens were further cut into three samples that
were used for hydrogen concentration measurement and microstructural studies
(phase analysis was not conducted as lab tests confirmed corrosion has limited effect
on phase composition), three samples for tensile tests and 10 samples for fatigue
testes. The tensile and fatigue testes are discussed in Chapter 4. Samples for
hydrogen concentration measurement and microstructural studies were 52×14 mm.
108
All cut samples were rinsed in rust removal solution and then grounded with
superfine sandpaper (600 grit) to remove rust. The rust removal solution was
prepared according to ASTM G1-03 (ASTM, 2017b). They were then cleaned
thoroughly with bidistilled water followed by acetone and dried with air. Photos of
samples before and after cleaning are shown in Figure 3-29. Hydrogen concentration
tests were carried out on the samples following the procedure introduced in Section
3.3.2.
After hydrogen measurement, the samples were further cut into sample 1 of 36×14×6
mm for element composition analysis and samples 2, 3 and 4 of 14×6× 4 mm for
grain size analysis. In particular, hydrogen concentration was measured for three trial
corroded samples. It was found that there are no changes in grain size and element
composition before and after hydrogen measurement. Therefore, this measurement
does not affect the samples.1
Figure 3-29: Specimens before (left) and after (right) Rust Removal
Thickness measurements were conducted on the two case study viaducts. One
viaduct contains two bridges constructed in 1889 and 1919, and the other viaduct
was constructed in 1908. Therefore, average corrosion loss can be determined for
steel subjected to natural corrosion after 98, 109 and 128 years (see Figure 3-30).
1
These measurement procedures were detailed in Section 3.3.3.
109
1.4
Thus, the relationship between corrosion loss and time for laboratory and field
specimens can be expressed as follows:
where C is the corrosion loss (mm) and t is time (days), ac1 and bc1 are the
regression parameters. ac1 and bc1 for different corrosive conditions are shown in
Table 3-10.
According to Draẑić and Vaŝĉiẑ (1989), the changes bc1 in Equation 3.9 mean only
the vertical translation of the function in the diagram, while the correlation factor
between simulated corrosion and natural corrosion is the ratio of ac1 for simulated
corrosion test to that for natural corrosion. The ratio indicates how many times faster
the corrosion process in simulated corrosion is than that under natural corrosion. The
correlation factors have been determined based on the simulated corrosion test of
non-stressed steel according to Draẑić and Vaŝĉiẑ (1989). Based on Table 3-10, the
factors are 65, 120, 815 and 2,500 for 0.00001 M HCl (pH=5), 0.003 M HCl
(pH=2.5), 1 M HCl (pH=0) and 3 M HCl (pH=-0.5) respectively for non-stressed
steel.
110
1 M HCl (pH=0) 0.0163 –0.0314 815.00
Based on the correlation factors, the relationship developed from simulated corrosion
can be correlated to that in natural corrosion through dividing the regression
parameters by their correlation factors.
Hydrogen concentration was measured for Type 1 samples under various corrosion
losses (1 mm, 2 mm and 3 mm in Table 3-9). The hydrogen concentration increased
from 0 to 2.65 ppm when corrosion loss increased to 3 mm. The relationship between
hydrogen concentration and corrosion loss for natural corrosion is shown in Figure 3-
31.
3
Hydrogen concentration
1.5
1
0.5
0
0 1 2 3 4
Corrosion loss (mm)
[ H ] aH C 2 bH C cH (3.10)
111
corrosion and natural corrosion are the ratios of aH and bH respectively for
simulated corrosion to that of natural corrosion (see Table 3-11). The correlation
factors for aH and bH are 128.96 and 13.93 respectively.
The element composition analysis indicates that the sample surface is mainly
composed of iron and oxygen under natural corrosion, which is the same as that of
simulated corrosion. The changes in the proportion of iron and the oxygen are shown
in Figure 3-32. When corrosion loss increases from 0 to 3 mm, the proportion of iron
reduces from 93.01 to 69.78%, while the proportion of oxygen increases from 5.92 to
27.08%.
100
80
Iron (%)
60
Fe% = 2.6025C2 - 15.579C + 93.051
40 R² = 0.9999
20
0
0 1 2 3 4
Corrosion loss (mm)
112
100
Oxygen (%)
80
60
O% = -1.5095C2 + 11.529C + 5.9997
40 R² = 0.9995
20
0
0 1 2 3 4
Corrosion loss (mm)
where aFe , aO , bFe , bO , cFe and cO are the regression parameters. Following a
similar procedure as that in Section 3.5.2.1, the correlation factors between simulated
corrosion and natural corrosion can be determined (see Table 3-12). The correlation
factors are 8.41 for aFe and 3.63 for bFe . There is no noticeable difference in element
composition between stressed and non-stressed steel at the same degree of corrosion.
Therefore, the correlation factors are suitable for stressed and non-stressed
conditions.
113
3.5.2.4 Grain Size Measurement
Figure 3-33 shows OM images of samples for different corrosion levels. It can be
seen that grain size is reduced due to intergranular corrosion long grain boundaries.
Also, fracture surfaces were found at the grain boundaries. The fracture surfaces are
likely slip lands caused by the dislocations among grain boundaries (Ren et al., 2012;
Pineau, 2015). The steel was subjected to cyclic loading at the time of bridge
operation. The fatigue damage caused by cyclic loading leads to this dislocation
among grain boundaries (Ni et al., 2010). The number of slip bands increases with
the increase of corrosion loss. This is because the intergranular corrosion weakens
the bonding force among steel grains and contributes to the initiation of dislocations.
114
(c) Corrosion loss 3 mm
The average grain size reduces from 10.12 to 6.83 μm when corrosion loss increases
from 1 mm to 3 mm. The model to predict grain size by knowing corrosion loss for
natural corrosion is shown in Figure 3-34.
14
12
10
Grain size (µm)
8
6
4 gz = 0.2006C2 - 2.4476C + 12.369
R² = 1
2
0
0 1 2 3 4
Corrosion loss (mm)
The relationship of grain size ( gz ) and corrosion loss ( C ) can be expressed for
laboratory and field specimens as follows:
gz ag C 2 bg C cg (3.13)
115
where a g , bg and cg are the regression parameters. Following a similar procedure as
that in Section 3.5.2.1, the correlation factors between simulated corrosion and
natural corrosion can be found (see Table 3-13). The correlation factors are 25.96 for
a g and 4.93 for bg for stressed steel, and 40.13 for a g and 5.69 for bg for non-
stressed steel.
3.5.2.5 Comparison
Hydrogen concentration, element composition and grain size are compared between
simulated corrosion and natural corrosion at the same degree of corrosion (i.e., 1 mm
of corrosion loss) (see Table 3-14). It can be seen that, at the same degree of
corrosion, the changes in microstructural features (hydrogen concentration, element
composition and grain size) are more significant for steel subjected to simulated
corrosion than natural corrosion. There are three explanations. First, there is more
hydrogen penetrating into steel by immersing steel in HCl compared with natural
corrosion. Second, the chloride atoms in the solutions break the passive films for
steel subjects to simulated corrosion, which contributes to the permeation of
hydrogen and oxygen and the reaction between iron and solutions. Third, steel
specimens suffer from more severe intergranular corrosion in simulated corrosion
tests due to their direct exposure to hydrogen and chloride atoms.
Stressed Non-stressed
[H] 10.65 ppm 7.33 ppm 1.01 ppm
Fe% 48.88% 48.88% 80.08%
116
O% 46.55% 46.55% 16.02%
gz 3.34 μm 4.82 μm 10.13 μm
3.6 Summary
This chapter has presented the simulated corrosion tests and field studies that
investigated the changes in hydrogen concentration and microstructural features
(element compositions, grain size, steel phase and morphology) during corrosion for
stressed and non-stressed steel. It was found that elastic stress can lead to the
acceleration in corrosion rate. Gutman’s model can underestimate this acceleration
effect. It was also found that corrosion can lead to the increase of hydrogen
concentration within steel. Therefore, steel can be more vulnerable to hydrogen
embrittlement during corrosion. The test results indicated that corrosion can change
element composition, reduce grain size and cause initiation of cracks and pits.
It was also found that the level of changes in hydrogen concentration and grain size
are higher for stressed steel than for non-stressed steel at the same degree of
corrosion. At corrosion loss of 1 mm, hydrogen concentration of stressed steel is
1.50 times that of non-stressed steel. Also, the percentage of grain size reduction of
stressed steel is 1.42 times that of non-stressed steel. There are no noticeable changes
in phase composition during corrosion for stressed and non-stressed steel. The
models to predict hydrogen concentration, element composition and grain size based
on corrosion loss were developed based on simulated corrosion tests. The correlation
factors between simulated corrosion and natural corrosion for corrosion loss,
hydrogen concentration and grain size were developed by field tests.
117
Chapter 4: Effect of Corrosion on Mechanical Properties
This section investigates the changes in tensile properties of stressed and non-
stressed steel subjected to simulated corrosion. For simulated corrosion, the solutions
and immersion process are the same as that in Section 3.3. Tensile properties are the
most widely used mechanical properties of steel in construction industry. In this
thesis, the relationships between tensile properties and corrosion loss are correlated
to the natural corrosion through field investigations.
4.2 Introduction
According to the literature review in Chapter 2 and test results presented in Chapter
3, corrosion leads to the initiation of pits and cracks due to the accumulation of
hydrogen within the steel. Corrosion also changes the element composition and
causes intergranular corrosion, which degrades the mechanical properties of steel
including tensile properties (e.g., yield strength, ultimate strength and ductility) and
fatigue properties (e.g., fatigue strength limitation).
Hydrogen concentration and intergranular corrosion are more severe for stressed
steel than for non-stressed steel at the same degree of corrosion. The coupled effect
of stress and corrosion also contributes to the initiation and propagation of corrosion
pits and cracks (Li & Cheng, 2008). Therefore, it is expected that the reduction of
mechanical properties is more severe for stressed steel than for non-stressed steel at
the same degree of corrosion (Revie, 2008).
Corrosion affects the fatigue properties of steel (i.e., the S-N curve) (Adasooriya &
Siriwardane, 2014). S magnitude for corroded steel can be smaller than that for
uncorroded steel at the same N to failure (Revie, 2008; Ni et al., 2010). With
corrosion, the parameters to determine the S-N curve change with corrosion loss and
time.
A review of existing literature suggests that few studies have investigated the
corrosion effect on mechanical properties, in particular, with stress, and almost no
118
research has compared the level of reduction of yield strength, ultimate strength and
ductility between stressed steel and non-stressed steel at the same degree of
corrosion. Also, there is no model that predicts the changes in S-N curve based on
corrosion loss (Pipinato & Modena, 2009; Ni et al., 2010; Adasooriya &
Siriwardane, 2014). In this chapter, the corrosion effect on mechanical properties for
stressed steel and non-stressed steel is studied through comprehensive laboratory and
field investigations. The changes in S-N curves due to corrosion are also investigated
through fatigue tests.
Tensile tests were conducted using a Shimadzu 500 kN test machine located at the
heavy structure lab of RMIT University (Figure 4-1). The machine was warmed up
to normal operating temperatures and then calibrated. Afterward, specimens were
loaded into the machine and extensometers were fit to the tested area of the specimen
to measure elongation (ASTM, 2016).
119
Figure 4-1: Set-Up of Tensile Test
The cross-head speed of the loading was set at 1.1 mm/s for the entire test. The full
range stress and strain curve were measured for each specimen in the tensile tests, on
which the yield strength, ultimate strength and failure strain of the specimens were
determined.
Conventionally, the yield strength and ultimate strength are determined by original
specimen’s cross-sectional area Acs and gauge length L0 :, which can be shown as
follows:
Pl
e (4.1)
Acs
e (4.2)
L0
where is the displacement of the specimen measured in the tensile test and Pl is
the load corresponding to the displacement. An engineering stress-strain curve is
obtained by plotting e against e .
120
The yield strength (the stress at which 0.2% plastic deformation occurs), ultimate
strength (maximum stress in tension test) and failure strain (the strain when the
failure occurs) determined based on engineering stress-strain curve are generally
used in design application to calculate the capacity of structural component.
However, for mild steel, a substantial change in the cross-sectional area and length of
the specimen is expected during the tensile test. In this way, the engineering stress
and strain determined based on Equations 4.1 and 4.2 are less accurate to investigate
the corrosion effect on the metallic tensile properties of steel. According to Garbatov
et al. (2014), the true stress t and strain t should be used to consider the change in
cross-section area and the length of the specimens, expressed as follows:
Pl P
t l (1 e ) e (1 e ) (4.3)
Aacs Acs
L
1 L
t dL ln ln(1 e ) (4.4)
L0
L L0
where Aacs is the actual cross-sectional area of the specimen and L is the actual
gauge length.
From true stress-strain curve, true yield strength, ultimate strength and failure strain
can be determined (Garbatov et al., 2014). In this thesis, the changes in tensile
properties were estimated based on both engineering stress-strain curve and true
stress-strain curve.
For non-destructive test, hardness measurement was conducted at the same locations
as that of ultrasonic thickness measurement (see Section 3.5.1, Figure 3-25 and Table
3-9). After thickness measurement, the test locations were further polished to remove
121
rust for Leeb hardness test. The test was conducted using a portable impact device
and the device was held firmly and vertically on the girder surface. Five hardness
measurements were conducted for each chosen location (ASTM, 2012) and the
hardness of the steel was determined based on the average value. The device and
measurement are shown in Figure 4-2.
As mentioned in Section 3.5.1, for destructive tests specimens were classified into
three corrosion levels after cutting—slight, median and heavy corrosion. At each
corrosion level, specimens were cut into three samples for tensile tests. The
thicknesses of the samples for tensile test were 15 mm for slight corrosion, 12 mm
for median corrosion and 10 mm for heavy corrosion. The dimension of the samples
was shown in Figure 3-1.
Samples were rinsed in rust removal solution and then ground with superfine
sandpaper (600 grit) to remove the rust following ASTM G1-03 (ASTM, 2017b) (see
Figure 4-3).
122
Figure 4-3: Specimens before (left) and after (right) Rust Removal
Both Leeb hardness measurement and tensile tests were conducted on these samples.
The empirical relationships between Leeb hardness and tensile properties were
established (see Figure 4-4).
440 Engieering
True
420 Linear (Engieering)
Yield strength (MPa)
Linear (True)
400
σyt = 1.1529HL - 135.77
380 R² = 0.7694
360
340
σye = 1.1051HL - 121.2
320 R² = 0.7236
300
400 410 420 430 440
Hardness (HL)
123
650
Engieering
600 True σut = 4.9673HL - 1580.4
500
450
400
σue = 2.0664HL - 446.84
350 R² = 0.9081
300
400 410 420 430 440
Hardness (HL)
0.6
Engieering
True
0.5 εfe= 0.0028HL - 0.8068
Linear (Engieering)
Linear (True) R² = 0.5726
0.4
Failure strain
0.3
0
400 410 420 430 440
Hardness (HL)
(c) Ductility
Notes. HL=Leeb hardness, ɛfe=engineering failure strain, ɛft=true failure strain, σut=engineering
ultimate strength, σue=true ultimate strength.
The regression models in Figure 4-4 can be used to predict yield strength, ultimate
strength and failure strain by knowing the Leeb hardness. As tensile test and
hardness measurements were conducted on steel at different levels of corrosion (in
Section 3.5), the changes in tensile properties can be determined as a function of
corrosion loss for natural corrosion (in Section 4.2.3).
124
4.2.3 Results and Discussion
As a representative, the engineering and true stress-strain curves for steel before and
after corrosion of 28 days in 1 M HCl and 3 M HCl are shown in Figures 4-5 and 4-6
respectively. It can be seen, from both engineering and true stress-strain curve, that
there is a noticeable reduction in ultimate strength and ductility for mild steel after
corrosion. The reduction of ultimate strength and failure strain is caused by
corrosion-induced microstructural changes which includes the initiation of hydrogen-
induced cracks (hydrogen embrittlement), changes in element composition,
intergranular corrosion (reduction in grain size) and stress concentration at corrosion
pits (Revie, 2008). The reduction is more obvious for stressed steel than non-stressed
steel. The specific changes in yield strength, ultimate strength and ductility during
corrosion and the related mechanism are further analysed below.
600 600
Engineering stress (MPa)
Engineering stress (MPa)
400 400
300 300
200 200
non-corrosion non-corrosion
1M HCl stressed 1M HCl stressed
100 100 1M HCl non-stressed
1M HCl non-stressed
0 0
0 0.2 0.4 0.6 0 0.05
Engineering strain Engineering strain
125
600 600
Enlarged area
400 400
300 300
200 200
non-corrosion
non-corrosion
1M HCl stressed
100 1M HCl stressed 100 1M HCl non-stressed
1M HCl non-stressed
0 0
0 0.2 0.4 0 0.05
Engineering strain Engineering strain
600 600
Enlarged area
Engineering stress (MPa)
500 500
400 400
300 300
200 200
non-corrosion non-corrosion
3M HCl stressed 3M HCl stressed
100 100
3M HCl non-stressed 3M HCl non-stressed
0 0
0 0.2 0.4 0.6 0 0.05
Engineering strain Engineering strain
126
(a) Engineering stress-strain curve (b) Stress-strain curve, enlarged from
(a)
600
400
Enlarged area
500
300
400
300 200
200
non-corrosion non-corrosion
3M HCl stressed 100 3M HCl stressed
100 3M HCl non-stressed 3M HCl non-stressed
0 0
0 0.2 0.4 0 0.05
True strain True strain
The level of reduction of tensile properties is higher for stressed steel than for non-
stressed steel at the same immersion time. This is because stressed steel has a faster
corrosion rate and is subjected to more significant changes in microstructural features
than non-stressed steel (see Section 3.4). After 28 days, the level of reduction of
engineering ultimate strength for stressed steel is 5.95, 2.60, 1.79 and 2.58 times that
127
of non-stressed steel in 0.00001 M HCl (pH=5), 0.003 M HCl (pH=2.5), 1 M HCl
(pH=0) and 3 M HCl (pH=-0.5) respectively. The level of reduction of true ultimate
strength of stressed steel is 1.85, 1.48, 1.81 and 1.98 times that of non-stressed steel
in these four solutions after 28 days.
For failure strain, the level of reduction of engineering failure strain for stressed steel
is 2.01, 1.52, 1.78 and 1.07 times that of non-stressed steel in 0.00001 M HCl
(pH=5), 0.003 M HCl (pH=2.5), 1 M HCl (pH=0) and 3 M HCl (pH=-0.5) after 28
days. The level of reduction of true failure strain of stressed steel is 2.06, 1.56, 1.90
and 1.08 times that of non-stressed steel in these four solutions after 28 days.
50
Reduction of engineering ultimate
45
40
35
30
strength (%)
25
20
15
10
5
0
0 10 20 30
Time (Days)
0.00001M stressed 0.00001M non-stressed
0.003M stressed 0.003M non-stressed
1M stressed 1M non-stressed
3M stressed 3M non-stressed
128
50
100
Reduction of engineering failure strain (%)
90
80
70
60
50
40
30
20
10
0
0 10 20 30
Time (Days)
0.00001M stressed 0.00001M non-stressed
0.003M stressed 0.003M non-stressed
1M stressed 1M non-stressed
3M stressed 3M non-stressed
129
100
The relationships between the yield strength and corrosion loss for stressed and non-
stressed steel are shown in Figure 4-9. It can be seen that there is a slight reduction in
yield strength for non-stressed steel. The reduction of engineering yield strength
increases from 0 to 4.18% when corrosion loss reaches 1.36 mm. In the meanwhile,
the reduction of true yield strength increases from 0 to 6.17%. The reduction of yield
strength is caused by microstructural changes discussed earlier in this section. The
level of reduction is higher for true yield strength than engineering yield strength due
to the loss of ductility during corrosion (Garbatov et al., 2014).
For stressed steel, yield strength unexpectedly increases when corrosion loss
increases from 0 to 0.86 mm. Then, yield strength reduces as corrosion loss
increases. This is because strain hardening occurs when steel is subjected to plastic
deformation caused by prestress. The plastic deformation leads to the dislocations
and slips among grain boundaries and subsequently, increases yield strength by
adding energy to the steel (Jin & Lee, 2009). In this thesis, although steel is subjected
130
to elastic stress (70% of yield strength), stress concentration at corrosion pits, defects
and grain boundaries can lead to the plastic deformation and dislocation/slip among
grain boundaries.
20
stressed
Reduction of engineering yield strength
15 non-stressed
Poly. (stressed)
Δσye = -3.1491C2 + 6.9428C + 0.4846
10 Poly. (non-stressed) R² = 0.813
5
(%)
0
0 0.5 1 1.5 2
-5
-10
Δσye= 10.151C2 - 16.48C - 1.9644
-15 R² = 0.6806
-20
Corrosion loss (mm)
20
stressed
Reduction of true yield strength (%)
15 non-stressed
Poly. (stressed)
10 Poly. (non-stressed) Δσyt = -3.5148C2 + 8.4161C + 1.1155
R² = 0.746
5
0
0 0.5 1 1.5 2
-5
Δσyt = 9.2749C2 - 14.468C - 0.9045
-10 R² = 0.6353
-15
-20
Corrosion loss (mm)
131
Notes. C = corrosion loss, σyt = true yield strength, σye = engineering yield strength.
Yield strength starts to reduce when corrosion loss is larger than 0.86 mm.
Engineering yield strength is reduced by 1.38% and true yield strength by 3.34%
when corrosion loss reaches 1.82 mm. The effect of corrosion-induced
microstructural changes can play a more dominant role over the strain hardening
effect when corrosion loss further increases and this eventually leads to the reduction
in yield strength. Then, after 28 days immersion in 3 M HCl, the reductions of
engineering and true yield strength of stressed specimens are respectively 1.08 and
1.85 times less than that of non-stressed specimens.
The relationship between the reduction in engineering yield strength, the true yield
strength and corrosion loss can be obtained by multiple regression (see Table 4-1).
The models are reasonably fit for experimental data (p<0.05).
Table 4-1: Relationship between Yield Strength and Corrosion Loss (Lab
Specimens)
Stress Equation R² p
condition
Engineering stressed ye 10.151C 2 16.4800C 1.9644 0.8130 2×10–5
yield strength
non-stressed ye 3.1491C 2 6.9428C 0.4846 0.6806 3×10–2
Notes. C = corrosion loss (mm), σyt = true yield strength (%), σye = engineering yield strength (%)
The relationship between ultimate strength and corrosion loss is shown in Figure 4-
10. When corrosion loss reaches 1.36 mm for non-stressed steel and 1.82 mm for
stressed steel, the engineering ultimate strength reduces by 16.96% and 11.08%, and
the true ultimate strength reduces by 28.65% and 33.65%. Also, the level of
reduction of ultimate strength is higher for stressed steel than non-stressed steel at the
same degree of corrosion. The difference in the reduction of ultimate strength
increases with corrosion loss increment. For example, at 0.5 mm of corrosion loss,
the reductions of engineering and true ultimate strength for stressed steel are 1.12
and 1.01 times that for non-stressed steel. At 1 mm of corrosion loss, the
132
reductions of engineering and true ultimate strength for stressed steel are 2.18 and
1.17 times that for non-stressed steel. As mentioned in Section 3.4.4, there is a
reduction in iron proportion, an increase in oxygen proportion and a reduction in
grain size during corrosion. These corrosion-induced changes in element
composition and grain size can lead to the reduction in ultimate strength.
37 stressed
Reduction of engineering ultimate
non-stressed
32 Poly. (stressed)
Poly. (non-stressed)
27
strength (%)
22
σue= -4.6326C2 + 17.664C + 0.38
17 R² = 0.9426
12
7
σue= -5.898C2 + 16.102C + 0.0977
2 R² = 0.9509
-3
0 0.5 1 1.5 2
Corrosion loss (mm)
(a) Engineering ultimate strength
40
Reduction of engineering yield strength
25
Δσut = -12.305C2 + 36.92C + 1.0271
(%)
R² = 0.9741
20
15
10
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(b) True ultimate strength
133
Figure 4-10: Ultimate Strength vs Corrosion Loss
Notes. C = corrosion loss, σu = true ultimate strength, σue = engineering ultimate strength
The level of reduction of ultimate strength is higher for stressed steel than for non-
stressed steel at the same degree of corrosion. This is because the reduction of grain
size is larger for stressed steel than non-stressed steel at the same degree of corrosion
(see Figures 3-17 and 3-18). Grain size reduces due to intergranular corrosion which
affects the bonding strength among steel grains and, subsequently, reduces the tensile
properties. Further, the interaction of stress and corrosion contributes to the initiation
of pits and cracks, leading to degradation of the tensile properties of steel (see
Figures 3-20 and 3-21). This indicates stressed steel is more vulnerable to
intergranular stress corrosion and grain fracture than non-stressed steel.
Figure 4-10 also shows, compared with yield strength, that corrosion-induced
microstructural changes play a more dominant role over the strain hardening effect
on the ultimate strength for stressed steel. Therefore, unlike yield strength, ultimate
strength of the steel reduces continuously during corrosion. This is because cracks
can be initiated at corrosion pits and IGSCC can be propagated when applied stress
becomes larger than yield strength during the tensile test. Therefore, the corrosion-
induced changes in element composition and grain size have a more severe effect on
ultimate strength than yield strength.
The relationships among the reduction in engineering ultimate strength, true ultimate
strength and corrosion loss can be obtained by multiple regression (see Table 4-2).
The models are reasonably fit for experimental data (R2>0.94 and p<0.05).
Table 4-2: Relationship between Ultimate Strength and Corrosion Loss (Lab
Specimens)
Stress Equation R² p
condition
Engineering stressed ue 4.6326C 2 17.644C 0.38 0.9426 9×10–9
stress-strain
non-stressed ue 5.8980C 2 16.1020C 0.0977 0.9509 3×10–9
curve
True stress- stressed ut 13.3C 2 41.064C 2.2158 0.9618 6×10–10
strain curve
non-stressed ut 12.305C 2 36.92C 1.0271 0.9741 4×10–11
134
Notes. C = corrosion loss (mm), σut = true ultimate strength (%), σue = engineering ultimate strength
(%)
The relationships between the failure strain and corrosion loss are shown in Figure 4-
11. Failure strain, for stressed and non-stressed steel, reduces during corrosion. From
Figure 4-11, the engineering and true failure strains reduce by 94.82% and 88.90%
respectively for non-stressed steel when corrosion loss reaches 1.36 mm. Also, the
corresponding failure strains reduce by 93.83% and 86.94% respectively for stressed
steel when corrosion loss reaches 1.82 mm. The level of reduction of failure strain is
higher for stressed steel than for non-stressed steel at the same degree of corrosion.
At 0.5 mm of corrosion loss, the reductions of engineering and true failure strain for
stressed steel are 1.05 and 1.02 times that for non-stressed steel. At 1 mm of
corrosion loss, the reductions of engineering and true failure strain for stressed steel
are 1.18 and 1.14 times that for non-stressed steel.
stressed
Reudction engineering failure strain(%)
100 non-stressed
Poly. (stressed) Δεfe = -47.18C2 + 130.46C + 11.625
Poly. (non-stressed) R² = 0.9236
80
40
20
0
0 0.5 1 1.5 2
Corrosion loss (mm)
135
stressed
non-stressed
100 Poly. (stressed)
Poly. (non-stressed) Δεft = -43.825C2 + 125.12C + 9.7146
60
Δεft = -41.73C2 + 115.15C + 6.3647
R² = 0.9091
40
20
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(b) True failure strain
Notes. C = corrosion loss (mm), ɛfe = engineering failure strain (%), ɛft = true failure strain (%).
The mechanism of failure strain reduction for stressed and non-stressed steel is the
same as that for ultimate strength. The relationships among the reduction in
engineering failure strain, true failure strain and corrosion loss can be obtained by
multiple regression (see Table 4-3). The models are reasonably fit for experimental
data (R2>0.90 and p<0.05).
Table 4-3: Relationship between Failure Strain and Corrosion Loss (Lab
Specimens)
Stress Equation R² p
condition
Engineering stressed fe 47.18C 2 130.46C 11.625 0.9236 5×10–8
stress-strain
non-stressed fe 51.703C 2 129.31C 7.3848 0.9091 5×10–7
curve
True stress- stressed ft 43.825C 2 125.12C 9.7146 0.9425 8×10–9
strain curve
non-stressed ft 41.73C 2 115.15C 6.3647 0.9091 2×10–7
Notes. C = corrosion loss (mm), ɛfe = engineering failure strain (%), ɛft = true failure strain (%).
136
In summary, corrosion causes changes in element composition and reductions in
grain size, which leads to the degradation of tensile properties (yield strength,
ultimate strength and failure strain) of corroded steel. Applied stress accelerates
corrosion and its interaction with corrosion exacerbates the degradation of ultimate
strength and failure strain of corroded steel.
The field investigation (see Section 4.2.2) measured the hardness and tensile
properties of steel subjected to different levels of natural corrosion. The changes in
yield strength, ultimate strength and failure strain due to corrosion are shown in
Figures 4-12 to 4-14.
30
Reduction of engineering
25
yield strength (%)
20
15
Δσye = -0.3428C2 + 3.4498C - 0.6139
10 R² = 0.6841
5
0
0 1 2 3 4
Corrosion loss (mm)
30
Reduction of true yield
25
strength (%)
20
15 Δσyt = -0.0728C2 + 2.5753C - 0.6852
10 R² = 0.6243
5
0
0 1 2 3 4
Corrosion loss (mm)
137
Notes. C = corrosion loss (mm), σyt = true yield strength (%), σye = engineering yield strength
(%).
30
Reduction of engeering
ultimate strength (%)
20 Δσue = 0.0674C2 + 3.8084C + 0.5729
R² = 0.9545
10
0
0 1 2 3 4
Corrosion loss (mm)
30
Reduction of true ultimate
R² = 0.647
10
0
0 1 2 3 4
Corrosion loss (mm)
Notes. C=corrosion loss (mm), σut=true ultimate strength (%), σue=engineering ultimate strength
(%).
138
50
Reduction of engineering
40
20
Δεfe = -4.6596C2 + 23.813C + 2.2076
10 R² = 0.8188 P=1×10-2
0
0 1 2 3 4
Corrosion loss (mm)
50
Reduction of true failure
40
strain (%)
30
20
139
Table 4-4: Comparison of Changes in Tensile Properties at Corrosion Loss of 1
mm
Factor Simulated corrosion test Natural corrosion
Stressed Non-stressed
–8.37% 2.68% 0.87%
–6.10% 4.15% 1.55%
13.54% 6.21% 4.45%
29.98% 25.64% 9.01%
94.91% 84.99% 21.31%
91.01% 79.78% 17.45%
Notes. σye=engineering yield strength, σyt=true yield strength, σut=true ultimate strength,
σue=engineering ultimate strength, ɛfe=engineering failure strain, ɛft = true failure strain.
From Tables 4-1 to 4-3 and Figures 4-9 to 4-14, the relationship between the changes
in tensile properties ( ye , yt , ue , ut , fe and ft ) and corrosion loss
(C) (mm) can be expressed for laboratory and field specimens as follows:
ft a ft C 2 b ft C c ft (4.10)
Following a similar procedure to that in Section 3.5.2, the correlation factors between
simulated corrosion (non-stressed steel) and natural corrosion are determined as
follows:
140
ac
cora (4.11)
an
bc
corb (4.12)
bn
where cora and corb are the correlation factors determined based on the ratios of
and respectively for simulated corrosion to that for natural corrosion. ac represents
a ye , a yt , aue , aut , a fe and a ft for simulated corrosion and an represents all the
corresponding values for natural corrosion. Similarly, bc represents bye , byt , bue , but ,
b fe and b ft for simulated corrosion and bn represents all the corresponding values
Table 4-5: Correlation Factors for Regression Parameters for Tensile Properties
Factors Simulated corrosion test Natural corrosion Correlation factor
The results of the experiments and field investigations show that the reduction of
tensile properties of corroded steel are mainly caused by the changes in the
141
microstructure of steel, including element composition and the grain size. Taking the
data shown in Figures 3-18 to 4-14 as inputs, regression analysis can help to develop
the relationship between tensile properties, element composition and grain size. This
relationship has not been developed in the existing literature. Since phase
composition remains unchanged for steel during corrosion, it is not considered in the
regression analysis. Models are developed based on true stress-strain curve to show
the effect of element composition and grain size on steel tensile properties.
According to Figure 3-35 and Figures 4-12 to 4-14, the mechanical properties (true
yield strength, true ultimate strength and failure strain) decreases with the reduction
of grain size caused by corrosion. Also, according to Zhao et al. (2007), the values of
typical mechanical properties increase with the reciprocal root of the grain size.
Therefore, it is assumed that the relationship between grain size and mechanical
properties can be represented as follows in the multiple regress:
gz m1 a2
yt , ut , ft f ( gz m1 ) or yt , ut , ft f ( ) (4.13)
gz m 2 a1
Also, it can also be seen from Figure 3-33 and Figures 4-12 to 4-14 that the
mechanical properties increase with the increase of oxygen proportion and the
reduction of iron proportion caused by corrosion. Therefore, it can be presented by
the relationship as follows:
O%m 4 a3
yt , ut , ft f ( ) (4.14)
gz m 2 a4
Combing Equation 4-13 and 4-14 and using the multiple regression analysis in
Excel 2016, for true yield strength ( yt ), the relationship can be written as
follows:
142
gz
gz 3 O%0.8 150 2.5
yt 5.83 103 ( )0.01 1.86(3 ) 6174.33 (4.15)
gz O% Fe%
0.1
gz 3 Fe%
1 8.5 O%
ut 4.60 107 ( ) 144.63 545.35 (4.16)
gz Fe%
1 8.5 O%
ft 8.62 105 ( ) 0.27 0.35 (4.17)
gz Fe%
The fit of Equations 4.13–4.15 are shown in Table 4-6 based on analysis of
variance (ANOVA) carried out by Excel 2016. Based on Table 4-6, the adjusted
coefficients of determination ( ) are 0.51, 0.95 and 0.94 accordingly for Equation
4.15, 4.16 and 4.17. It indicates that Equation 4.15 explains 51% of variance of yt ,
Equation 4.16 explains 95% of variance in ut and Equation 4.17 explains 94%
of variance in ft .
Also, the P value obtained for test statistic are 4.99×10–12, 1.45×10–12 and 7.84×10–
12
for Equations 4.15-4.17. A low p-value (< 0.05) indicates that, Equations 4.15-
4.17 provide a statistically significant explanation for the differences in yt , ut
and ft . Therefore the developed models are fit for experimental data.
acceptable. Therefore, the models can provide a reliable estimation of the tensile
properties based on microstructural features (Giustolisi & Savic, 2006).
143
Table 4-6: Fit of Equations
Equation p Std Error
4.15 0.51 4.99×10–12 12.94
4.16 0.95 1.45×10–12 15.61
–12
4.17 0.94 7.84×10 0.031
The field studies conducted out in Section 3.5 show that corrosion mainly occurs at
bridge components subjected to normal stress. Therefore, this section only discusses
the corrosion effect on the S-N curve of steel subjected to normal stress instead of
shear stress. In this thesis, a model was developed to predict the changes in S-N
curves of corroded steel. The developed model was then verified through fatigue
tests conducted in the laboratory.
N AS B (4.18)
Both and are related to tensile properties of steel, for example, engineering
ultimate strength ( ), which can be expressed as follows (British Standards
Institution, 2014; Zhao, 1995):
A N1 (kr ue ) B (4.19)
log N 0 log N1
B (4.20)
log(kr ueo )
where N 0 is the largest load cycle in the low load cycle region, N1 is the minimum
load cycle in the high load cycle region, ue is the engineering ultimate strength of
steel, k r is the ratio of fatigue strength limitation to ultimate strength and ueo is the
original engineering ultimate strength of steel (Zhao et al., 1994; Bandara et al.,
144
2015). A log plot of Equation 4.18 is the well-known S-N curve subjected to normal
S (see Figure 2-14).
Due to the corrosion of steel, the tensile properties of steel, (e.g., ue ) will change
(Li et al., 2018). As such, the S-N curve of corroded steel will change with corrosion
over time. A relationship between engineering ultimate strength of corroded steel and
corrosion loss can be developed based on Figure 4-10 and expressed as follows:
Equations 4.21 and 4.22), and can be determined from Equations 4.19 and 4.20
as a function of corrosion loss and time, from which the corrosion affected S-N curve
can be determined. Figure 4-15 shows an example of S-N curve of G250 steel plates
with no corrosion ( =0 mm) and with various corrosion losses ( =3 mm and 6 mm),
where N 0 is taken as 105, N1 as 107, ueo 420 MPa and kr 0.31 , based on BS
7608 (British Standards Institution, 2014).
145
1000
C=0mm
Line 1: σueo=420.00
C=3mm
Stress range (S) MPa
C=6mm
Line 3:
krσᵤeo=136.42
Line 2:
100 logk 𝑟 − logN krσᵤeo=102.17
logS =
m krσᵤeo=71.06
Nₒ=105 N1=107
10
1.0E+04 1.0E+05 1.0E+06 1.0E+07 1.0E+08
Number of cycles to failure (N)
Figure 4-15: S-N Curve for Steel with and without Corrosion
The corrosion effect on S-N curve estimated based on Equations 4.19 to 4.22 can be
verified by carrying out constant amplitude axial fatigue tests on corroded steel. The
design of the fatigue test follows ASTM E466-15 (ASTM, 2015b). As mentioned in
Section 3.5, specimens were cut from an existing bridge and were classified into
three corrosion categories—sight, median and heavy corrosion. At each corrosion
level, these specimens were further cut into 10 samples for fatigue tests. The
thickness was 15 mm for slightly corroded specimens, 12 mm for median corroded
specimens and 10 mm for heavily corroded specimens. The geometry of the test
samples were designed based on their thickness (see Figure 4-16) (ASTM, 2015b).
Before relevant tests were carried out, all samples were rinsed in rust removal
solution and then ground with superfine sandpaper (600 grit) to remove the rust. The
rust removal solution was prepared according to ASTM G1-03 (ASTM, 2017b).
They were then cleaned thoroughly with bidistilled water followed by acetone and
dried with air. Both ends of the samples were polished to fit the clamp systems in the
fatigue test machine (see Figure 4-17).
146
Polished ends
Polished ends
147
The samples were subjected to axial constant cyclic load with a frequency of 30 Hz.
According to the tensile test results (Section 4.2.3), the tested samples have an
ueo
original engineering ultimate strength ( ) of 420 MPa. Therefore, the stress
ranges exerted on the samples were 0–100, 0–120, 0–220, 0–320 and 0–420 MPa
(ASTM E466-15 2015). There were three samples under each stress range and the
number of cycles to failure of each sample was recorded (ASTM E466-15 2015).
The stress ranges S and the number of cycles to failure N for each sample were then
plotted. The details of the fatigue test plan are shown in Table 4-7.
ueo 420 MPa and kr 0.31 (British Standards Institution, 2014) for the corroded
specimens in the fatigue test. Therefore, A and B can be calculated based on
Equation 4.19 and 4.20 by knowing corrosion loss (C). A and B are 3.41×1014 and
3.62 respectively when C=1 mm, 1.64×1014 and 3.50 respectively when C=2 mm and
8.04×1013 and 3.38 respectively when C=3 mm.
The S-N curve determined from prediction based on corrosion loss is compared with
that from the fatigue test results (see Figure 4-18). The R2 value between the test
148
results and the predicted S-N curve is 0.94 for slight corrosion condition, 0.96 for
median corrosion and 0.95 for heavy corrosion, and the p values are 2×10–5, 1×10–8
and 3×10–9 for these three conditions respectively. Therefore, the predicted S-N curve
agrees well with the test results, thus, the developed model (i.e., Equations 4.19 to
4.22) is verified.
1000
Stress range (S)
100
predicted-slight corrosion
test results
10
1.0E+04 1.0E+05 1.0E+06 1.0E+07 1.0E+08
Number of cycles to failure (N)
(a) Slight corrosion
1000
Stress range (S)
100
predicted-median corrosion
test results
10
1.0E+04 1.0E+05 1.0E+06 1.0E+07 1.0E+08
Number of cycles to failure (N)
(b) Median corrosion
149
1000
10
1.0E+04 1.0E+05 1.0E+06 1.0E+07 1.0E+08
Number of cycles to failure (N)
(c) Heavy corrosion
The following were observed from the model and fatigue results. Corrosion affects
the fatigue failure of steel structure primarily in two ways. First, corrosion leads to
loss of the cross-sections and reduction of the sectional properties (e.g., area and
effective section modulus). Due to the sectional area loss, the S that the beam is
subjected to under certain cyclic loading increases with time (Adasooriya &
Siriwardane, 2014). Second, the S-N curve for steel changes with corrosion loss,
including the fatigue strength coefficient and fatigue strength exponent. Therefore
the S magnitude for corroded steel is smaller than that for uncorroded steel at the
same N to failure (Revie, 2008; Ni et al., 2010). It is imperative to consider the
corrosion effect on both stress range and S-N curve during fatigue assessment to
prevent the underestimation of fatigue damage (i.e., damage accumulation index in
Equation 2.34).
Further, in this thesis a model was developed to predict the changes in S-N curves by
knowing the tensile properties and corrosion loss. This model allows damage
accumulation index to be treated as a stochastic process which provides the basis to
predict the fatigue life of a corroded structure using time-dependent reliability
analysis. Also, models have been developed to predict the full range S-N curves of
corroded steel based on its ultimate strength and corrosion loss. It is much easier to
test the ultimate strength than S-N curve of steel. In this way, the models developed
150
in this thesis simplify the experimental work for the construction of S–N curves for
corroded steel.
4.4 Summary
It was also found that stress can expedite the reduction of tensile properties of
corroded steel, more significantly for ultimate strength and failures strain than yield
strength, due primarily to stress concentration at corrosion pits and IGSCC. Further,
this chapter presented empirical models developed to predict the changes in tensile
properties based on corrosion loss. The models, which were determined based on
simulated corrosion tests, were correlated to those from natural corrosion and their
correlation factors were determined.
Finally, a model was developed to determine the effect of corrosion on the S-N curve
of corroded steel, which was verified with experimental results. Through these
developed models, the resistance of a bridge under different failure criteria (e.g.,
flexure, shear, deflection and fatigue) can be treated as a stochastic process.
Therefore, the time-dependent reliability analysis can be conducted to predict the
service life of a bridge under each failure criterion.
151
Chapter 5: Corrosion-Induced Delamination
5.1 Introduction
In this chapter, the causes and effects of preferred corrosion of steel are
quantitatively determined through a comprehensive experimental program and
detailed analysis of steel microstructure. A large number of simulated corrosion tests
were conducted by immersing stressed and non-stressed steel in acidic solutions to
acquire a new understanding of the corrosion process of continuously cast steel.
Three microstructural characteristics that affect the corrosion resistance of steel were
quantified across the steel thickness—grain size, phase composition and distribution
152
of impurities. Thus, this chapter provides a novel understanding of the causes and
effects of preferred corrosion on continuously cast steel, advances knowledge on the
corrosion science of steel and suggests improvements for the manufacturing process
of steel.
Test specimens were made of G250 mild steel, which was continuously cast without
any defects as confirmed by the supplier. The geometries for non-stressed specimens
are shown in Figure 3-2 and that for stressed specimens are shown in Figure 3-3. The
stressed specimens were wrapped with acid-resistant tape at both ends and the stress
exerted on the specimens was 70% of yield strength. The stressed system (clamp and
jig system) used in this experiment was the same as that used for immersion tests in
Section 3.2.1 (see Figure 3-4).
The set-up of the immersion test followed the same procedure as that in Section 3.2
and is shown in Figure 3-5b. The specimens were immersed in 3 M HCl (pH=-0.5), 1
M HCl (pH=0), 0.003 M HCl (pH=2.5) and 0.00001 M HCl (pH=5) for accelerating
corrosion. The immersion times were 7, 14 and 28 days and there were three stressed
and non-stressed specimens at each immersion stage. After each stage of immersion,
the protections of the specimens against corrosion were removed (for stressed
specimens). The specimens were washed thoroughly with bidistilled water followed
by acetone and dried with air to remove rust and stop corrosion. Corrosion losses of
the specimens were measured following Equations 3.1 and 3.3. Specimens were then
cut into three samples of 14×6×4 mm (same as samples 2, 3, 4 in Figures 3-2 and 3-
3) for microstructural studies.
153
5.2.1.2 Delamination Measurement
The effect of preferred corrosion is steel delamination which was measured on each
cut sample using two parameters—the opening depth into the steel body and the
opening width of splitting layers of cross-section (see Figure 5-1).
OM and ImageJ were used to measure the width and depth of the delamination. The
depth and width were measured on the three cut samples after each stage of corrosion
to determine the average value. After each period of immersion, samples 2, 3 and 4
were hot mounted and polished following the procedures mentioned in Section 3.3.3.
They were then etched with 2% Nital for 30 seconds. OM scanning was then
conducted on each cut sample using 5× magnification (scan location shown in
Figure 5-1). ImageJ analysis was conducted on each scanned image to calculate
the width and depth of delamination, following the procedure in Section 3.4.4.
The microstructural studies include the measurement of grain size, steel phase and
number of impurities. Three locations along the thickness of the cross-section were
selected for grain size and phase measurement—the edge, one-quarter thickness (1.5
mm from the edge) and the middle. The measuring locations were chosen as close to
steel/solution interface as possible (see Figure 5-1). Grain size and phase
measurement were conducted on the three cut samples after each stage of corrosion
to determine the average value at each measurement locations.
154
Grain size was quantified for etched samples using OM at 100× magnification and
average grain size was calculated using linear intercept procedure as mentioned in
Section 3.3.3.2. Steel phase analysis was conducted by EBSD. The two phases
selected for EBSD analysis were ferrite and cementite (relevant procedures shown in
Section 3.3.3.3).
Corrosion loss over time for stressed and non-stressed specimens is shown in Figure
5-2. Corrosion loss is larger for stressed steel than non-stressed steel at the same
immersion time for all HCl solutions.
155
4
0.00001M stressed
0.00001M non-stressed
3.5 0.003M stressed
0.003M non-stressed
3 1M stressed
1M non-stressed
1.5
0.5
0
0 10 Time (Days) 20 30
From Figures 5-4c and 5-4d, delamination for stressed and non-stressed specimens in
the middle sections can be seen after 14 days immersion in 0.003 M HCl (pH=2.5)
solution. For specimens immersed in 1 M (pH=0) and 3 M (pH=-0.5) HCl solutions,
delamination initiates much earlier, within 7 days of immersion (Figs 5-3e, f, g and
h) and becomes more prominent with the increase of immersion time. Figures 5-3 to
5-5 show that higher concentration of acid contributes to the initiation and
development of delamination due to the accelerated corrosion.
156
(a) 0.00001 M HCl, non-stressed (b) 0.00001 M HCl, stressed
157
(g) 3 M HCl, non-stressed (h) 3 M HCl, stressed
Delamination
158
(e) 1 M HCl, non-stressed (f) 1 M HCl, stressed
159
(c) 0.0003 M HCl, non-stressed (d) 0.003 M HCl, stressed
The OM images of steel cross-sections after 7, 14 and 28 days immersion are shown
in Figures 5-6 to 5-8. It can be seen that during immersion, the middle region of steel
160
undergoes more corrosion loss than edge regions. The corrosion loss in the middle
region is more observable for stressed steel than for non-stressed steel at the same
corrosion time. Also, although delamination cannot be seen in visual inspection for
non-stressed steel after 14 and 28 days immersion in 0.00001 M HCl, it can be seen
under magnification in OM analysis (see Figures 5-6 and 5-7).
161
(e) 1 M HCl, non-stressed (f) 1 M HCl, stressed
Voids
Figure 5-6: Optical Microscope Images of Samples after 7 Days Immersion (5×
Magnification)
162
(c) 0.003 M HCl, non-stressed (d) 0.003 M HCl, stressed
Voids
163
(a) 0.00001 M HCl, non-stressed (b) 0.00001 M HCl, stressed
Steel Sample
Steel Sample
164
Voids
Figure 5-8: Optical Microscope Images of Samples after 28 Days Immersion (5×
Magnification)
Stressed 7
0.003 M HCl (pH=2.5) Non-stressed 7
Stressed 7
Stressed 7
Stressed 7
165
Figure 5-9 shows that steel (non-stressed) in 3 M HCl completely delaminates
after 125 days. Based on Table 3-11, the acceleration factor from immersion in 3
M HCl (pH=-0.5) solution to that in 0.00001 M HCl (pH=5) solution is
2500/65=38.46. Based on this acceleration factor, it is possible to conceptually
extrapolate that steel in pH=5 solution may be completely delaminated after 13.6
years. This very serious for structural steel in general, given the service life of
properly designed steel structures can be 100 years, and in particular for steel
structures where the working environment has pH values less of than 5, such as
steel wells in an oil field (Finšgar & Jackson, 2014) and underground steel water
tanks (Sheoran, Sheoran & Choudhary, 2010). Also, complete delamination may
occur faster than expected since most steel structures are working under stress.
The severity of steel delamination in such a short time, relative to service life,
highlights the significance of these results.
Figure 5-10 shows the changes in the depth and width of delamination with corrosion
time. It can be seen that the depth and width of delamination increases with corrosion
time. Also, the depth and width of delamination for stressed steel is higher than that
of non-stressed steel at the same corrosion time. After 28 days, the depth of
delamination for stressed steel is 1.21, 1.25, 1.32 and 1.33 times that of non-stressed
steel in 0.00001 M HCl (pH=5), 0.003 M HCl (pH=2.5), 1 M HCl (pH=0) and 3 M
HCl (pH=-0.5) respectively. The width of delamination for stressed steel is 1.05,
166
1.12, 1.29 and 1.26 times that for non-stressed steel in these four solutions
respectively after 28 days. Therefore, the combined effect of stress and corrosion
contributes to the propagation of delamination due to accelerated corrosion.
4
0.00001M stressed
0.00001M non-stressed
3.5
Preferred corrosion depth (mm)
0.003M stressed
0.003M non-stressed
3 1M stressed
1M non-stressed
3M stressed
2.5 3M non-stressed
1.5
0.5
0
0 10 Time (Days) 20 30
(a) Depth
4
0.00001M stressed
0.00001M non-stressed
3.5
Preferred corrosion width (mm)
0.003M stressed
0.003M non-stressed
3 1M stressed
1M non-stressed
3M stressed
2.5 3M non-stressed
1.5
0.5
0
0 10 Time (Days) 20 30
(b) Width
Figure 5-11 shows the changes in the depth and width of delamination with corrosion
loss for stressed and non-stressed samples. After 28 days in 3 M HCl, the depth of
delamination increases from 0 to 1.67 mm for non-stressed samples when corrosion
167
loss reaches 1.36 mm and from 0 to 2.08 mm for stressed samples when corrosion
loss reaches 1.82 mm. The width of delamination increases from 0 to 1.06 mm for
non-stressed samples and from 0 to 1.72 mm for stressed samples when corrosion
loss reaches 1.36 mm and 1.82 mm respectively. The depth and width of
delamination are larger for stressed samples at the same corrosion degree (corrosion
loss). For example, at corrosion loss of 0.5 mm, the depth and width of
delamination for stressed steel are 1.10 times and 1.26 times those of non-stressed
steel. At corrosion loss of 1 mm, the depth and width of delamination for stressed
steel are 1.22 times and 1.48 times those of non-stressed steel.
3
stressed
non-stressed de = -1.3861C2 + 3.6629C - 0.0273
Preferred corrosion depth (mm)
1.5
de = -1.709C2 + 3.5806C - 0.0391
R² = 0.9702
1
0.5
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(a) Depth
3 stressed
Preferred corrosion width (mm)
2.5 non-stressed
wi= -1.1154C2 + 2.9905C - 0.0446
2 R² = 0.9841
1.5
1
wi= -1.2833C2 + 2.5409C - 0.0221
0.5 R² = 0.9773
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(b) Width
Figure 5-11: Changes in Depth and Width of Delamination with Corrosion Loss
Notes. de=depth of delamination (mm), wi = width of delamination (mm), C = corrosion (mm).
168
Relationships between the delamination geometry—depth and width—and corrosion
loss were developed based on multiple regression (see Figure 5-11 and Table 5-2).
The models are reasonably fit for experimental data (R2>0.096 and p<0.05).
Notes. de = depth of delamination (mm), wi = width of delamination (mm), C = corrosion loss (mm).
Delamination occurs due to the intrinsic differences in grain size, phase and number
of impurities between the edge and middle across the thickness. These differences
are discussed in detail in this section.
169
the middle than in the edge for stressed and non-stressed steel after 28 days. This
indicates that intergranular corrosion is more severe in the middle. It can also be seen
that the level of reduction of grain size, both in the middle and edge, is higher for
stressed steel than non-stressed steel at the same immersion time.
(b) After 7 days corrosion, non-stressed specimen (edge left, middle right)
(c) After 7 days corrosion, stressed specimen (edge left, middle right)
170
(d) After 14 days corrosion, non-stressed specimen (edge left, middle right)
(e) After 14 days corrosion, stressed specimen (edge left, middle right)
(f) After 28 days corrosion, non-stressed specimen (edge left, middle right)
171
(f) After 28 days corrosion, stressed specimen (edge left, middle right)
The grain size can be determined by linear intercept procedure (see Figure 3-10) for
samples after each stage of immersion. Figure 5-13 shows the changes in grain size
at each scanned location with corrosion loss. It can be seen that the average grain
size before corrosion is 12.18 μm in the middle, 10.60 μm at one-quarter of sample
thickness and 6.36 μm at the edge. After 28 days corrosion, the average grain size for
non-stressed samples reduces by 46.5% in the middle, 45.6% at one-quarter
thickness and 45.2% in the edge when corrosion loss reaches 1.36 mm. For stressed
steel, the grain size reduces by 59.7% in the middle, 52.7% at one-quarter thickness
and 51.9% in the edge when corrosion loss reaches 1.82 mm. The reduction of grain
size is caused by intergranular corrosion, as mentioned in Section 2.4.2.
The difference in grain size between the middle and edge also increases with the
increase of corrosion loss. For example, before corrosion, the grain size in the
middle is 1.91 times that in the edge. However, the ratio increases to 1.22 for
non-stressed steel when corrosion loss reaches 1.36 mm and to 1.27 for stressed
steel when corrosion loss reaches 1.82 mm. Therefore, grain size reduces faster in
the middle where corrosion is more advanced than at the edge, which leads to
more corrosion loss in the middle and, subsequently, delamination.
172
16
stressed
14 non-stressed
Poly. (stressed)
12 Poly. (non-stressed)
16
stressed
14 non-stressed
Poly. (stressed)
12 Poly. (non-stressed)
Grain size (μm)
10
gzq = 5.5377C2 - 9.6975C + 9.2241
8 R² = 0.5347
6
4
gzq = 4.1533C2 - 9.7262C + 9.2987
2 R² = 0.718
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(b) One-quarter thickness
173
16
stressed
14 non-stressed
12 Poly. (stressed)
Notes. gze = grain size at edge (μm), gzq = grain size at one-quarter thickness (μm), gzm = grain
size at middle (μm), C = corrosion loss (mm).
The results also confirm that the small grain size at the edge has a higher corrosion
resistance than the large grain size found in the middle, consistent with the literature
(Ralston & Birbilis, 2010; Marcus, 2011). Grain refinement increases the number of
grain boundaries and these boundaries improve the stabilities of passive films since
they have a higher energy than the bulk grain (Ralston & Birbilis, 2010). In addition,
grain refinement improves corrosion resistance by decreasing the compositional
difference between bulk grain and grain boundaries, which helps neutralise the
galvanic reactions. This study is the first to analyse the effect of grain size on
corrosion-induced delamination.
The relationships between grain size and corrosion loss across the thickness of steel
were developed from multiple regression (see Table 5-3). The models are reasonably
fit for experimental data (p<0.05).
174
Table 5-3: Relationship between Grain Size and Corrosion Loss
Location Stress Equation R² p
condition
Middle stressed gzm 5.2076C 2 12.055C 10.182 0.7888 1×10–4
Notes. gze = grain size at edge (μm), gzq = grain size at one-quarter thickness (μm), gzm = grain size
at middle (μm), C = corrosion loss (mm).
The effect of stress on grain size reduction is more significant in the middle than
in the edge. There are four reasons. First, large grains have a higher hydrogen
diffusion coefficient (Revie, 2008). As a result, the hydrogen concentration in the
middle is higher than the edge. IGSCC is, therefore, more likely to occur in the
middle (Revie, 2008). Second, passive oxide film can prevent grains from
intergranular corrosion (Ralston & Birbilis, 2010). Large grain size makes the
passive oxygen film less stable and it is more likely to be ruptured in the middle
with the presence of stress. Third, large grains have lower ductility than small
grains. Therefore, the grains in the middle are more likely to be fractured than
those in the edge with the presence of stress (Arioka et al., 2006). Fourth, for steel
specimens subjected to axial tension, the stress in the middle of the thickness can
be larger than that in the edge due to the localised corrosion at the surface (Huang
et al., 2002).
Figure 5-14 presents the changes in steel phase composition at three measured
locations with corrosion loss (3 M HCl solution as representative). Ferrite (α- )
proportion is around 85% in the middle, 80% at one-quarter thickness and 75% in the
edge with corrosion loss increasing. Cementite ( ) proportion is around 2% in
the middle, 3% at one-quarter thickness and 5% in the edge. These results indicate
175
that ferrite proportion is higher and cementite proportion is lower in the middle of
thickness compared to that at the edge.
100
90
80
70 Ferrite stressed
Phase (%)
60 Ferrite non-stressed
50 Cementite stressed
Cemenitie non-stressed
40
Others stressed
30 Others non-stressed
20
10
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(a) Middle
100
90
80
70 Ferrite stressed
Phase (%)
60 Ferrite non-stressed
50 Cementite stressed
Cemenitie non-stressed
40
Others stressed
30 Others non-stressed
20
10
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(b) One-quarter thickness
176
90
80
70
Ferrite stressed
60 Ferrite non-stressed
Phase (%)
50 Cementite stressed
Cemenitie non-stressed
40
Others non-stressed
30 Others non-stressed
20
10
0
0 0.5 1 1.5 2
Corrosion loss (mm)
(c) Edge
177
The EBSD scanning images of the steel confirm that there are more ferrite and less
cementite in the middle compared to the edge for both un-corroded and corroded
samples (see Figure 5-15). They also show that there are more cementite particles
dispersed into the ferrite grains at the edge compared to into the middle. Also, there
are more cementite particles dispersed into the ferrite grains for stressed samples
since the stress causes rupture of ferrite grains (Revie, 2008).
(b) Non-stressed sample after 28 days immersion in 3 M HCl (edge left, middle
right)
178
(c) Stressed sample in after 28 days immersion in 3 M HCl (edge left, middle right)
BSE images of each cut sample (samples 2, 3 and 4) before corrosion are shown in
Figures 5-16a, b and c and one corroded sample is shown in Figure 5-16d. Impurities
mainly reside in the middle of steel cross-sections. At the edge, one impurity was
observed in samples 2 and 4, and two impurities were observed in sample 3. In the
middle, however, sample 2 has six impurities, sample 3 has seven impurities and
sample 4 has five impurities. These impurities are corroded away after corrosion, as
shown in Figure 5-16d. Additionally, corrosion pits are formed at the steel/solution
interface for corroded samples. These corrosion pits are likely formed due to
galvanic corrosion when impurities are exposed to acid (Szklarska-Śmialowska,
Szummer & Janik-Czachor, 1970).
179
(a) Un-corroded sample 2 (edge left, middle right)
180
(d) Non-stressed sample after 28 days immersion in 3 M HCl (edge left, middle
right)
The compositions of the observed impurities are shown in Table 5-4. EDS analysis
indicates that impurities 1–3, 5–7, 9–11 and 13–21 primarily contain Aluminium and
Oxygen, while impurities 4, 8, 12 and 22 primarily contain Aluminium, Oxygen,
Manganese and Sulphur. Moreover, impurity 15 contains Chromium, Nickel and
Oxygen. Based on the literature (Garet et al., 1998; Zhang & Thomas, 2003;
Payandeh & Soltanieh, 2007) and EDS analysis, impurities 1–3, 5–7, 9–11 and 13–
21 are likely to be Aluminium Oxides, impurities 4, 8, 12 and 22 are −
and impurity 15 is Cr-Ni-O compounds. Impurities accelerate corrosion by
inducing galvanic reactions and stress concentration (Marcus, 2011). Therefore,
the larger number of impurities in the middle of steel cross-section is due to
preferred corrosion.
181
Table 5-4: Chemical Composition of Impuritiesa
Impurity O Al Si S Cr Ni Mn Fe Other Total
Number
1 26.10 35.12 0.00 0.00 0.00 0.00 0.84 37.93 0.01 100
2 36.02 37.01 0.00 0.04 0.00 0.00 0.00 26.93 0.00 100
3 26.39 32.96 0.00 0.67 0.00 0.00 1.73 33.63 4.62 100
4 32.88 38.76 0.00 3.07 0.00 0.00 6.86 18.66 0.00 100
5 13.74 18.12 0.00 0.00 0.00 0.00 0.36 67.78 0.00 100
6 11.09 12.89 0.00 0.00 0.00 0.00 0.41 75.04 0.57 100
7 20.31 28.71 0.00 0.17 0.00 0.00 0.72 49.57 0.52 100
8 31.26 39.92 0.00 2.92 0.00 0.00 6.52 19.38 0.00 100
9 15.25 18.08 0.42 0.00 0.00 0.00 0.00 63.39 2.86 100
10 31.31 35.83 0.00 0.76 0.00 0.00 0.73 28.81 2.56 100
11 16.39 19.35 0.35 0.00 0.00 0.00 0.00 59.85 4.06 100
12 25.98 31.43 0.00 1.35 0.00 0.00 5.65 35.58 0.01 100
13 55.42 43.53 0.00 0.00 0.00 0.00 0.00 1.05 0.00 100
14 26.39 33.44 0.00 0.56 0.00 0.00 1.36 37.87 0.38 100
15 39.01 0.10 0.32 0.00 10.32 4.81 0.00 45.44 0.00 100
16 35.20 38.12 2.02 0.00 0.00 0.00 3.52 21.14 0.00 100
17 21.49 19.86 0.00 0.00 0.00 0.00 0.00 56.74 1.91 100
18 36.01 36.1 0.00 1.56 0.00 0.00 2.26 23.80 0.27 100
19 30.43 37.30 0.00 0.73 0.00 0.00 1.39 30.16 0.01 100
20 42.18 45.2 0.00 1.41 0.00 0.00 3.88 4.48 2.85 100
21 37.42 45.92 0.00 0.00 0.00 0.00 0.64 16.02 0.00 100
22 27.27 27.56 0.00 5.49 0.00 0.00 11.48 22.71 5.49 100
Notes. a All values are percentages. O=Oxygen, Al=Aluminium, S=Sulfur, Cr=Chromium, Ni=Nickel,
Mn=Manganese, Fe=Iron.
Preferred corrosion occurs in the middle of the thickness of steel and is caused by
variations in grain size, phase composition and distribution of impurities between the
middle and edge across the thickness of a steel cross-section. The effect of preferred
corrosion is delamination, which is severe as it destroys the integrity of steel. Since
preferred corrosion is more severe under stress and corrosion combined environment,
it is also imperative to control the level of stress that steel is subjected to.
182
The following suggestions could be beneficial in preventing preferred corrosion
during the manufacturing of steel.
183
`
The grain size of steel can also be refined by adding alloying elements (Maalekian,
2007). For example, Boron concentrated at grain boundaries to form Boron Carbide
during manufacturing prohibits the growth of the grain (Maalekian,
2007). Additionally, Niobium and Vanadium precipitate during the rolling process of
continuous casting and also hinder the growth of grains (Maalekian, 2007). These
alloying elements are likely to concentrate in the middle of the thickness of steel
during continuous casting (Thomas, 2002). Therefore, it is recommended that Boron,
Niobium and Vanadium be added to steel during the manufacturing process to refine
grain size in the middle and prevent preferred corrosion.
Eliminating the difference in phase composition across the thickness of the cross-
section of steel also prevents preferred corrosion. The literature reports that this can
be achieved by normalising continuously cast steels products (Digges, Rosenberg &
Geil, 1966; Shrestha, Alsagabi, Charit, Potirniche & Glazoff, 2015). In this process,
the steel is heated to austenitisation temperature which makes ferrite and cementite
grains recrystallise and transform into austenite (Digges et al., 1966; Shrestha et al.,
2015). After heating, the steel is held at the austenitisation temperature for a
sufficient time to form homogenous microstructure. Rapid cooling is then undertaken
to decompose the austenite into ferrite, cementite and undefined oxides, resulting in
them being uniformly distributed across the thickness of the steel (Digges et al.,
184
1966; Shrestha et al., 2015). Additionally, performing thermo-mechanical treatments
on continuously cast steel products (see Section 5.4.1) can also help make steel phase
composition homogenous across its thickness (Junior et al., 2012).
For stressed steel, the stress in the middle of the thickness can be larger than that
in the edge due to the localised corrosion at the surface (Huang et al., 2002),
which enhances the formation of delamination. It is imperative to prevent the
initiation and propagation of localised corrosion (i.e., pitting corrosion). The
localised corrosion can be prevented by controlling pHs, chloride concentration
and temperature in the corrosive environment (see Section 3.5). Applying
cathodic protection in these locations can also protect steel from localised
corrosion. Residual stress is likely to be distributed in the middle of steel
thickness during continuous casting. It is imperative the perform heat treatment
on steel to eliminate the residual stress (Thomas, 2002).
5.5 Summary
185
Observation and analysis of test results showed that preferred corrosion occurs as a
result of the variation in microstructure within steel, in locations where larger grain
size, less cementite and more ferrite contents exist. For continuously cast steel,
preferred corrosion occurs in the middle of steel cross-section due to the slower
solidification during manufacturing process. It was found that intergranular
corrosion and localised stress concentration are triggers for the initiation of steel
delamination. The width and depth of the delamination increases with corrosion
progress, up to 1.06 mm and 1.66 mm respectively after 28 days immersion in 3 M
HCl.
It was also found that stress contributes to the initiation and propagation of
delamination. This is because film rupture can be more serious in the middle than in
the edge when steel is stressed. Also, hydrogen embrittlement, intergranular
corrosion and IGSCC are more severe in the middle with the presence of stress.
186
Chapter 6: Service Life Prediction
6.1 Introduction
Corrosion can affect the service life of a bridge by causing reduction of cross-section
and mechanical properties (i.e., yield strength, ultimate strength, failure strain and
fatigue strength limitation) (see Chapters 1 and 2). It is imperative to treat both cross-
section and mechanical properties of bridge elements as time dependent to make a
better prediction of the service life of a bridge. However, few studies have
considered the degradation of mechanical properties caused by corrosion in service
life prediction. The majority of research has estimated the service life of a bridge
based on design codes, which is less accurate than time-dependent reliability
methods (Melchers, 1999). Also, there is a high auto-correlation among the corrosion
damage at each point in time (Zhao et al., 1994; Ni et al., 2010). However, this auto-
correlation has not been considered in existing reliability-based assessment methods
on bridges subjected to corrosion.
The relationships between mechanical properties, cross-section loss and time have
been developed in Chapters 3 and 4. The relationship between changes in parameters
in the S-N curve, corrosion rate and time have been developed in Chapter 4. These
relationships provide the input data for reliability analysis, through which the
stochastic models of bridge resistance can be determined for different failure modes,
the time-dependent probability of failure can be computed by outcrossing method
and the service life of a bridge can be predicted based on system reliability analysis.
187
6.2 Formulation of Service Life
In this section, the definition of service life, the time-dependent reliability methods
and the stochastic models are presented. These models provide the basis for
conducting service life prediction of a bridge.
The service life of a structure is defined as the time period at the end of which the
structure stops performing its intended functions (Li & Mahmoodian, 2013). In
assessing the risk of failures for a structure, a performance criterion should be
established. In the theory of structural reliability, this criterion is expressed in the
form of a limit state function as follows:
G( R, Sl , t ) R(t ) Sl (t ) (6.1)
where Sl (t ) is the load or its effect and R(t ) is the resistance at time t.
With the limit state function of Equation 6.1, the probability of structure failure p f
failure ( Pa ), it is the time the structure (or system) fails. This can be determined as
follows:
p f (Tc ) Pa (6.3)
where Tc denotes the time at which a phase of service life ends. In principle, Pa can
be determined from risk-cost optimisation of the structure during its whole service
life. Also, most codes and standards have specified Pa . For example, Pa is 0.023% in
American Association of State Highway and Transportation Officials (1998) and
0.011% in AS 5100.1 (Australian Standard, 2017a) and BS EN 1990 (British
188
Standards Institution, 2002). In this section, Pa is 0.011%, as the case study bridge is
located in Australia.
limit, R(t ) , sometime during the service life of the steel structure. Equally, the
probability of failure of the structure, p f (t ) , can be determined based on the
probability of the first occurrence of such an excursion during that period. This is
known as ‘first passage probability’ and under the assumption of Poisson processes it
can be expressed as follows (Li and Melchers 1993)
t
vdt
p f (t ) 1 [1 p f (0)]e 0
(6.4)
where p f (0) is the probability of structural failure at time t = 0 and υ is the mean
rate for the process Sl (t ) to up-cross critical limit R(t ) . Based on Li and Melchers
(2005), the mean up-crossing rate is very small in practical problems. Thus, Equation
(6.4) can be rewritten as follows:
t
p f (t ) p f (0) vdt (6.5)
0
The up-crossing rate in Equation (6.5) can be determined by Rice formula as follows
(Rice 1944):
. . .
v vR ( Sl R) f . ( R, Sl )dR (6.6)
Sl Sl
Sl
189
where vR is the up-crossing rate of the stochastic process Sl (t ) relative to the
. .
threshold R(t ) , R is the slope of R with respect to time, Sl is the time derivative
.
process of Sl and f . is the joint probability density function for Sl and Sl .
Sl Sl
The solution to Rice formula, Equation 6.6, is rare and only possible under certain
conditions. An analytical solution to Equation 6.6 has been derived by Li and
Melchers (1993) when Sl (t ) is a Gaussian process and the threshold R is
deterministic as follows:
. . .
.
R Sl R . R . R .
v vR det ( ){ ( ) (
Sl Sl Sl Sl Sl Sl Sl Sl
)} (6.7)
S l
S l
. . .
Sl Sl Sl Sl Sl Sl
where v vRdet denotes the up-crossing rate when the threshold R is deterministic,
() and () are the standard normal density and distribution functions respectively,
.
and denote the mean and standard deviation of Sl and Sl , represented by
subscripts and ‘ | ’ denotes the condition. For a given Gaussian stochastic process
with mean function Sl () and auto-covariance function Csl sl (ti , t j ) , all terms in
.
.
E[ Sl S l R] . S l (R S )
Sl
(6.8)
S
.
Sl Sl Sl l
l
. [ . 2 (1 Sl 2 )]1/2 (6.9)
Sl Sl Sl
where,
d Sl (t )
. (6.10)
Sl d (t )
2Csl sl (ti , t j )
[ ]1/2 (6.11)
ti t j
.
i j
Sl
190
C . (ti , t j )
S [
sl sl
]1/2 (6.12)
l
Csl sl (ti , t j ) C . . (ti , t j )
sl sl
Csl sl (ti , t j )
C (ti , t j ) (6.13)
t j
.
sl sl
It is acknowledged that Sl (t ) may not be Gaussian process nor stationary, for which
analytical solutions are rare. Li et al. (2016) recently derived an up-crossing solution
for non-Gaussian and non-stationary process. As the purpose of this thesis is not
deriving analytical solution for Rice formula but application of first-passage
probability theory to predict time-dependent probability of fatigue failure, please see
Li et al. (2016) for more information on analytical solutions for non-Gaussian and
non-stationary process.
It is reasonable to assume that the R(t ) is more than Sl at the beginning of structural
. . .
t .
R Sl R . R . R .
p f (t ) ( ){ ( ) (
Sl Sl Sl Sl Sl Sl Sl Sl
)}dt (6.14)
0
S l
S l
. . .
Sl Sl Sl Sl Sl Sl
191
Sl (t ) f ( X1 , X 2 ,..., X n , t ) (6.15)
where X1 , X 2 ,..., X n are the basic random variables, the probabilistic information of
which are (presumed) available. With this treatment, the mean Sl () and standard
Melchers, 2005).
introduced, which can be defined in such a way that its mean is unity (i.e., E ( Sl ) 1
S
) and its coefficient of variation is Sl (t ) (Li & Melchers, 2005). Thus, Sl (t )
l
S l
Sl (t ) Slc (t ) sl (6.16)
where Slc (t ) is treated as a pure time function of residual load effect which can be
calculated based on Sections 6.3 and 6.4. The mean and auto-covariance function of
Sl (t ) can be determined as follows:
192
bridge girder under different performance criteria (failure modes) are presented in
this section.
G(M R , M Sl , t ) M R (t ) M Sl (6.19)
flexural moment due to the load effect, which is treated as a deterministic parameter
in this research. Moreover, the M R (t ) and M Sl are resistance and load effect at the
M R (t ) ye (t )Ze (t ) (6.20)
where Z e (t ) is the effective section modulus of mid girder of the span and ye (t ) is
The bending moment M Sl is case specific and can be calculated based on the
geometry, the support condition and load distribution on the girders. For the case
study girder, M Sl is presented in Section 6.5.
For shear failure, the limit state function can be written as follows:
where VR (t ) is the shear resistance of girder at time t. VSl is the shear force due to
VR (t ) 0.6 ye (t ) Aw (t ) (6.22)
193
The shear force VSl is also case specific and can be calculated based on the
geometry, support condition and load distribution of the girders. For the case study
girder, VSl is presented in Section 6.5.
G( R , Sl , t ) R Sl (t ) (6.23)
where R is the allowable deflection under live load, that is, the deflection at the
middle of the span due to live load effect at time t. According to AS 5100.2
(Australian Standard, 2017b), maximum mid-span deflection under live load should
not exceed 1/640 of the span length and Sl is case specific, which can be calculated
based on the elastic modulus of steel, the second moment of inertia of girder cross-
section, corrosion rate and time. For the case study girder, Sl is presented in Section
6.5.
G( D, , t ) D(t ) (6.24)
where D(t ) is the Miner’s damage accumulation index at time t , which can be
determined as follows:
n(t )
D(t ) E[ S (t ) B (t ) ] (6.25)
A(t )
where A(t ) and B(t ) can be determined based on Equations 4.15 and 4.16 (in
Section 4.3). n(t ) is the total number of fatigue stress cycles applied during time t,
which can be determined as follows (Zhao 1995):
194
where fload is the load frequency in the lifetime and is time in years.
B(t )
E[ S (t ) B (t ) ] [ 2S0 (t )]B (t ) ( 1) (6.27)
2
2
S0 (t ) E[ S (t )] (6.28)
In Equation 6.28, E[S (t )] is the mean stress effect at time , which can either be
normal stress effect or shear stress effect. To consider the worst-case scenario, the
largest load cycle to which a structure is subjected is used to determine the mean
stress effect. The stress concentration at the changes in the cross-sectional area of the
structural components is not considered according to BS 7608 (British Standards
Institution, 2014). Thus, the mean stress effect can be determined based on the
maximum stress and minimum stress in largest load cycle as follows (Ni et al.,
2010):
max g (t ) min g (t )
E[ S (t )] (6.29)
2
where max g (t ) and min g (t ) are the maximum and minimum stress on the girders
which are case specific and determined in Section 6.5 for the case study bridge.
Most of existing steel bridges have riveted connections. For riveted connections, the
limit state function can be defined by calculating the tensile and fatigue failure of the
connections plates and the fatigue failure of rivets (Czarnecki & Nowak, 2006). The
limit state functions of connection plates and rivets under different performance
criteria are presented in this section.
195
6.4.1 Tensile Failure
In most cases, connection plates are subjected to tension for a riveted connection,
based on which the limit state function is calculated as follows:
the maximum normal stress on the connection plates at time t. max c (t ) is case
specific and is presented in Section 6.5 for the case study bridge.
The limit state function for the fatigue failure of connection plates is the same as
Equation 6.24. The Miner’s damage accumulation index D(t ) in the limit state
function can be determined by Equations 6.25–6.29.
For connection plates, E[S (t )] can be determined based on the maximum normal
stress max c (t ) and minimum normal stress min c (t ) at the critical location of the
connection plates (locations where the tensile failure occurs), following Equation
6.30 (Kulak, Fisher & Struik, 2001).
The limit state function for fatigue failure of rivets is also the same as Equation 6.24.
For most existing bridges, corrosion has a negligible effect on the geometry and
mechanical properties of the rivets (Chen & Duan, 2014). Therefore, A , B and
E ( S ) in Equation 6.25 are not time dependent. The mean shear stress effect, E ( S ) ,
can be determined based on the maximum shear stress, max r , and minimum shear
stress, min r , of rivets at the critical locations, which are as follows (Kulak et al.,
2001):
max r min r
E (S ) (6.31)
2
max c (t ) , min c (t ) , max r (t ) and min r (t ) for the case study connection plates are
determined in Section 6.5.
196
All the limit state functions presented in Sections 6.3 and 6.4 can be rewritten into
the formats in which Sl (t ) is a Gaussian process and up-crosses the deterministic
threshold R (see Table 6-1). Therefore, the time-dependent probability of failure for
each performance criteria of the bridge can be calculated based on Equations 6.4–
6.14.
In this section, a steel railway viaduct subjected to corrosion is used as a case study.
The name and location of the viaduct are not mentioned due to confidentiality. The
structural configuration, load condition and corrosion models of the case study
viaduct are introduced. Based on the information presented in this section, the basic
variables to carry out time-dependent reliability analysis of the viaduct are
determined.
The following assumptions were made when carrying out reliability analysis on the
case study viaduct based on recommendations provided by the Department of
Railway Transportation and site inspections:
The viaduct is subjected to uniform loss of thickness at the bottom flange and
the bottom half of the web
197
The superimposed dead load is assumed to be uniformly distributed along the
girders
Dead loads and live loads (axle load) are directly exerted on the girders
The average train frequency is constant
The average corrosion rate is constant
In fatigue analysis, mean stress effect is calculated based on the maximum
and minimum load effects at mid-span (normal stress) and at connections (shear
stress)
The failures of the decks and piers are not considered in system reliability
analysis
In system reliability analysis, events of failure of the system are uncorrelated.
The structural configuration of the case study viaduct is presented in this section. The
viaduct was constructed in 1889 and was rehabilitated in the 1970s. The entire length
of the viaduct is approximately 760 m and the length of each span varies from 10 m
to 20 m. The superstructure of each span has steel through decks supported by three
riveted mild steel girders. Girders between each span are connected by rivets at the
webs. End connections of the girders are expansion joints. The piers of the viaduct
are made of solid masonry.
The reliability analysis presented in this section focused on the longest span (span
length of 20 m) of the viaduct since it is subjected to the largest load effect and most
severe corrosion. The structural configuration of this span is shown in Figure 6-1.
The flanges of girder have five 14 mm-thick steel plates. The thickness of the web is
16 mm (see Figure 6-1a). The thicknesses of flanges and webs were considered as
normally distributed random variables with coefficient of covariance (COV) of 0.01.
The viaduct steel is G250 steel, with average engineering yield strength of 320 MPa
and ultimate strength of 420 MPa. Engineering yield strength ( ye ) and ultimate
There were no cracks inspected on the girders. Therefore, the fatigue classification of
girders belongs to Class B based on Figure 2-15 (in Section 2.5.4). Girders between
198
each span are connected by rivets at the webs and, therefore, the fatigue classification
of connection plates belongs to Class D (see Figure 2-15). The average fatigue
strength limitation is 130.20 MPa for girders and 75.02 MPa for connection plates.
Therefore, the average ratio k r between fatigue strength limitation and ultimate
strength is 0.31 for girders and 0.18 for connection plates, based on which A(t ) and
B(t ) in Equation 6.25 can be determined from Equations 4.17 and 4.18 for girders
and sections plates. As suggested by BS 5100.10 (British Standards Institution,
1980), the COV of k r is 0.14 to consider its randomness due to the stress
concentration caused by corrosion and manufacturing. The mean area of rivet hole is
453.0 mm2 and COV of the area was assumed to be 0.05 to consider its randomness.
For rivets, the rivets of the connection belong to Class 100. Therefore, S5 is 180
MPa and S8 is 46 MPa (Australian Standard, 2017d), based on which A(t ) and B(t )
in Equation 6.25 can be determined from Figure 2-16 for rivets. The COV for both
S5 and S8 is 0.08 to consider the randomness.
199
(b) Steel trough deck
200
V/2
(e) Connection between girders
Notes. Units for Figures 6-1a–c and e are in millimetres. Units for Figure 6-1d are in metres.
Based on Figure 3-29 (in Section 3.5), corrosion rate can be assumed as constant
based on the site inspections (same as the standards model in Section 2.4.1). The
average corrosion rate ( cr ) for the case study viaduct is mm , or
8.7 mm . The COV of corrosion rate is 0.15 to consider its randomness
(Bai & Jin, 2015). Corrosion was assumed to occur at the bottom flange, half of the
web close to the bottom flange (see red lines in Figure 6-1a) and connection plates.
This assumption was based on site inspections. The thicknesses of bottom flange,
corroded web and connection plates were treated as a function of corrosion rate and
time, shown as follows:
d (t ) d0 C d0 cr t (6.32)
where d (t ) (mm) is the thickness/width of the steel plate at time t (in years), d 0
201
The tensile properties of the steel in corrosion affected area are time dependent. As
mentioned in Section 4.2.3, the relationships between tensile properties and time are
shown as follows:
There were no signs of corrosion on rivets. Therefore, the area of rivet hole was
considered as constant.
The dead load and live load of girders are determined according to AS 5100.2
(Australian Standard, 2017b). The self-weight of the girder is 11 kN/m.
Superimposed dead load on the girders is deterministic (see Table 6-2).
Each span contains four pairs of tracks (see Figure 6-1c). Based on the data from the
Department of Railway Transportation, trains operating on the viaduct have four
motor cars and two trailer cars. Axle load is 236 kN for a motor car and 217 kN for a
trailer car. The COV of axle load is 0.13 for a motor car and 0.16 for a trailer car.
This considers the difference of an empty versus a full train (i.e., fully occupied with
passengers). From the dead load and live load information presented above and the
axial load system show in Figure 6-2, the load effects on the viaduct girder can be
determined.
202
Figure 6-2: Axial Load System of Train Carriages on the Case Study Viaduct
The resistance and load effect of the viaduct for each failure mode can be determined
based on the basic random variables presented in Table 6-3.
203
Symbol Parameter Mean Covariance
2
Aeq Area of rivets 453 mm 0.05
Based on Table 6-3, for flexural and shear failure, M R and VR can be determined by
Equations 6.20 and 6.22 based on the yield strength before corrosion, the dimensions
of flanges and webs, average corrosion rate and time. M Sl and VSl can be
ql 2 1 P1 l P l P l
M Sl Pl
1 ( x1 ) 2 ( x2 ) 2 ( x3 ) (6.35)
8 4 2 2 2 2 2 2
ql l x1 l x2 l x3
VSl P1 P1 ( ) P2 ( ) P2 ( ) (6.36)
2 l l l
where l is the length of the girder, P1 and P2 are the axle force of motor car and
trailer cars exerted on girders, q is the uniform disturbed load calculated based on
204
self-weight and Table 6-2. x1 , x2 and x3 are determined based on Figure 6-2, and
are 2.2 m, 6.9 m and 9.1 m respectively.
For deflection failure, R can be determined based on span length and Sl which is
calculated as follows:
=
( ) ( ) ( )
(6.37)
where E is the elastic modulus of steel and I (t ) is the second moment of inertia of
girder cross-section at time t.
For the tensile capacity, ye (t ) can be determined based on Equation 6.33 and
2 2
VSl VSl ey1
max c (t ) (6.38)
4 Aaa (t ) 4 I bb (t )
where VSl is the shear force determined based on Equation 6.36, e is the horizontal
distance between force and centre of rivet group (centre O), y1 is the vertical
distance between the critical location and neutral axis (see Figure 6-1e), Aaa (t ) is the
net area of the connection plate at section A-A at time (see Figure 6-1e), I bb (t ) is
the moment of inertia of the connection plate at section B-B about neutral axis
(moment of inertia of solid plate minus moment inertia of all rivets holes) at time t
(see Figure 6-1e). Sections A-A and B-B are the horizontal and vertical cross-section
of the connection plate across the critical location (see Figure 6-1e).
For fatigue failure, the damage accumulation index, D(t ) , can be determined by
Equation 6.25 for the girder, connections plates and rivets by knowing their
geometries, ultimate strength, dead load, live load, train frequency, ratio between
fatigue strength limitation and ultimate strength, corrosion rate and time. For girders,
the maximum and minimum normal stress can be determined as follows:
205
max g (t ) M max / Ze(t ) (6.39)
where M max and M min are the maximum and minimum bending moments. M max can
ql 2
M min (6.41)
8
For connection plates, the maximum normal stress can be determined as Equation
6.38 and the minimum normal stress can be determined as follows:
where Vmin is the minimum shear force which the connection is subjected to and can
be determined as follows:
1
Vmin ql (6.43)
2
For mid girder, the stress ranges that girder and connection plates subjected to are
136.42 MPa and 114.25 MPa respectively, which are larger than their fatigue
strength limitation (130.20 MPa for girder and 75.02 MPa for connection plates). The
girder and connection plates can, therefore, be vulnerable to fatigue failure (Ni et al.,
2010). For edge girders, however, the stress ranges that girder and connection plates
subjected to (68.21 MPa and 57.13 MPa) are smaller than their corresponding fatigue
strength limitation. In this way, the edge girder and its connection plates are not
vulnerable to fatigue damages. Therefore, the two failure modes—fatigue failure of
girder and connection plates—were not considered when carrying out time-
dependent reliability analysis of edge girders.
For the rivets, the maximum shear stress, max , and minimum shear stress, min , of
rivets at the critical locations are as follows (Kulak et al., 2001):
206
1 V V er1 2 V V er1
max ( max )2 ( max
nt
) 2( max )( max
nt
) cos (6.44)
Aeq 4nt 4nt
4 rnt 2
4 rnt 2
1 1
1 V V er1 2 V V er1
min ( min )2 ( min
nt
) 2( min )( min
nt
) cos (6.45)
Aeq 4nt 4nt
4 rnt 2
4 rnt 2
1 1
where Vmax is determined by Equation 6.36. Aeq is the cross-section area of rivets.
Aeq is constant for the case study rivets since they are subjected to negligible
corrosion effect. rnt is the distance between the rivet nt and centroid ( r1 as
representative in Figure 6-1e). nt is the total number of rivets in the rivets group.
is the angle between the lines in Figure 6-1e.
Based on Table 6-3, the statistics of Sl (t ) for different failure modes were obtained
0.0E+00
0 200 400 600 800
-2.0E+06 Num=10
Num=1000
-4.0E+06 Num=10000
Num=1000000
-6.0E+06
-MR (Nm)
-8.0E+06
-1.0E+07
-1.2E+07
-1.4E+07
Time (Years)
(a) Flexural, girder
207
0.0E+00
0 200 400 600 800
-5.0E+05
Num=10
Num=1000
-1.0E+06 Num=10000
Num=1000000
-1.5E+06
-VR (N)
-2.0E+06
-2.5E+06
-3.0E+06
-3.5E+06
-4.0E+06
Time (Years)
(b) Shear, girder
0.0E+00
-2.0E-03 0 200 400 600 800
Num=10
-4.0E-03
Num=1000
-6.0E-03 Num=10000
ΔS (m)
Num=1000000
-8.0E-03
-1.0E-02
-1.2E-02
-1.4E-02
-1.6E-02
-1.8E-02
-2.0E-02
Time (Years)
(c) Deflection, girder
208
1.5E+00
Num=10
1.0E+00 Num=1000
Num=10000
5.0E-01 Num=1000000
D
0.0E+00
0 200 400 600 800
-5.0E-01
-1.0E+00
-1.5E+00
Time (Years)
(d) Fatigue, girder
From the above, the probability of failure due to corrosion can be determined by
Equation 6.13 for each failure mode (see Figure 6-4). From Figure 6-4, it can be seen
that the probability of failure is sensitive to the auto-correlation between two points
in time (ρ value) for each failure mode. The difference in the probability of failure at
given time (t = 360 years as representative) for different p values is shown in Table
6-4.
209
1
ρ=0.9
ρ=0.5
0.8 ρ=0.1
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
(a) Flexural, girder
1
ρ=0.9
ρ=0.5
0.8 ρ=0.1
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
(b) Shear, girder
210
1
ρ=0.9
ρ=0.5
0.8
Probability of failure ρ=0.1
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
1
ρ=0.9
ρ=0.5
0.8
ρ=0.1
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
211
1
ρ=0.9
ρ=0.5
0.8
ρ=0.1
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
1
ρ=0.9
ρ=0.5
0.8
ρ=0.1
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
212
1
ρ=0.9
ρ=0.5
0.8
Probability of failure ρ=0.1
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
213
From Table 6-4, the probability of failure is higher for 0.9 than 0.5 and 0.1
at the same time. According to Revie (2008), corrosion damage at any time points
depends on the corrosion loss and corrosion effect in previous times. It is reasonable
0.9
to assume that there is high auto-correlation between different time ( ) for
0.9
each failure mode. Therefore, the time-dependent probability of failure at
was used for system analysis for each failure mode. Table 6-4 also shows that the
probability of failure of the structural component can be underestimated by ignoring
the auto-correlation between different times, which agrees with Melchers (1999).
As shown in Sections 6.3 and 6.4, the corrosion effect on mechanical properties
(yield strength and fatigue strength limitation) was considered when calculating the
time-dependent probability of failure for flexural failure, shear failure, fatigue failure
of girders, tensile failure and the fatigue failure of connection plates. Figure 6-5
shows the comparison of time-dependent probability of failure determined for these
failure modes with and without considering the corrosion effect on mechanical
properties. To compute the probability of failure without corrosion impact on
mechanical properties, ye and ue were kept constant and equal to 320 MPa and
With t = 360 years as a representative, the probability of failure is 1.09, 1.28, 1.04,
1.07 and 1.12 times that of not considering the corrosion impact on mechanical
properties for each failure mode in Figure 6-5. In most previous studies, the
corrosion effect on mechanical properties was ignored in failure assessment. Clearly,
this will lead to underestimating the failure probability of structural components.
Figure 6-5 shows that it is essential to consider the corrosion effect on mechanical
properties to prevent unexpected failure of the steel viaduct within its service life.
214
1
Corrosion impact considered
Corrosion impact not considered
0.8
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
1
Corrrosion impact considered
Corrosion impact not considered
0.8
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
(b) Shear, girder
215
1
Corrosion impact considered
Corrosion impact not considered
0.8
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
1
Corrosion impact considered
Corrosion impact not considered
0.8
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
216
1
Corrosion impact considered
Corrosion impact not considered
0.8
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
Figure 6-6a shows the time-dependent probability of failure for different failure
modes for mid girder at ρ = 0.9. The probability of fatigue failure of the girder is
higher than others. At t = 360 years, the probability of fatigue failure of the girder is
2.11 times that of the fatigue failure of connection plates, 13.42 times that of
deflection failure, 26.84 times that of flexural failure, 67.10 times that of tensile
failure of connection plates, 234.83 times that of shear failure and 1,409 times that of
fatigue failure of rivets.
As mentioned in Section 6.5.5, the two failure modes (fatigue failure of girder and
connection plates) were not considered when carrying out time-dependent reliability
analysis of edge girders. The time-dependent probability of failure for different
failure modes for edge girders is shown in Figure 6-6b. For edge girders, the
probability of failure determined based on flexural failure is higher than others. At t
= 360 years, the probability of failure determined based on flexural failure is 1.44
times that of the deflection failure, 603.80 times that of tensile failure of connection
217
plates, 4,846.72 times that of shear failure and 199,220.54 times that of fatigue
failure of rivets.
1
Flexture-girder Shear-girder
Deflection-girder Fatigue-girder
0.8 Tensile-connection Fatigue-connection
Probability of failure
Fatigue-rivets
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
(a) Mid girder
0.04
Flexture-girder Shear-girder
Deflection-girder Tensile-connection
0.03 Fatigue-rivets
Probability of failure
0.02
0.01
0
0 200 400 600 800 1000 1200
Time (Years)
(b) Edge girder
Notes. ρ = 0.9.
Overall, the probability of fatigue failure of mid girder and its connection plates are
higher than other failure modes. Therefore, these two failure modes are more critical
than others and more care should be taken to protect the mid girder and its
218
connection plates from fatigue failure. The results of sensitivity analysis conducted
on these two failure modes are presented in the next section.
For fatigue failure of mid girder and its connection plates, the damage accumulative
index, D(t ) , can be affected by several factors including corrosion rate, train
frequency and load effect. The sensitivity analysis helps to identify the factors that
have the most effect on fatigue failure so that future research can focus on those
factors. Reliability-based sensitivity analysis is a tool that helps to quantify the
changes in the probability of failure with regard to variations in the input parameters
(Ditlevsen & Madsen, 1996; Mahmoodian & Li, 2016). In this way, the effect of
each variable on fatigue failure can be estimated by reliability-based sensitivity
analysis (Ditlevsen & Madsen, 1996).
In the sensitivity analysis, the relative contribution ( X2 ) of each random variable (X)
to the variance of the limit state function (G) can be introduced as follows (Ditlevsen
& Madsen, 1996):
G
2
X
X
X2 (6.46)
G2
where X is the standard deviation of the random variable and G is the variance of
the limit state function.
For the mid girder in case study, there are 10 random variables, including corrosion
rate ( cr ), flange width ( b f ), flange thickness ( d f ), web depth ( bw ), web thickness
( d w ), axle load for motor car ( P1 ), axle load for trailer car ( P2 ), load frequency
( fload ), the engineering ultimate strength ( ueo ) and ratio ( k r ) between fatigue
strength limitation and ultimate strength before corrosion.
For the connection plates, there are also 10 random variables, including the initial net
area of section A-A ( Aaa Aoaa ), corrosion rate ( cr ), horizontal distance between
force and centre O ( e e ), train frequency ( fload ), the initial moment of inertia of the
219
connection plate at section B-B ( Ibb I obb ), ratio ( k r ) between fatigue strength
limitation and engineering ultimate strength, axle load for motor car ( P1 ), axle load
for trailer car ( P2 ), vertical distance between the critical rivet hole and neutral axis
( y1 ) and the initial engineering ultimate strength ( ueo ). Due to the effect of
corrosion, the contribution of each variable to fatigue failure may change with time
(Ni et al. 2010). Therefore, the contribution of each variable was estimated at various
points in time at t 0 (no corrosion effect), t 100 , t 300 and t 500 years using
Equation (6.46). The selection of time scale was based on Ni et al. (2010). The
results of sensitivity analysis are shown in Figure 6-7.
For the probability of fatigue failure determined at the mid girder (Figure 6-7a), the
contribution of the ratio between fatigue strength limitation and ultimate strength ( k r
) is the largest (72.61%) among all the variables at t = 0. After 500 years, the
contribution of k r reduces to 27.68% and the contribution of corrosion rate increases
from 0.004% to 60.77%. The contribution of corrosion rate becomes the largest after
500 years. For the probability of fatigue failure determined at the connection plates
(see Figure 6-7b), the contribution of the ratio between fatigue strength limitation
and ultimate strength ( k r ) is also the largest (38.52%) initially (t = 0). After 500
years, the contribution of k r becomes 27.68% (second largest) and the contribution
of corrosion rate increases from 0.0005% to 62.41% (largest).
In both failure modes, the contribution of corrosion rate becomes the largest after
500 years, which indicates a very significant effect of corrosion in the longer term
despite the average corrosion rate being extremely small in the case study. The
randomness of corrosion rate, therefore, has an indispensable effect on the accuracy
of the fatigue life prediction and this effect becomes more significant over time. For
any steel bridge, the effect of k r on fatigue life is more important at earlier stages
and the effect of corrosion rate is more important for an old bridge. Also, more care
should be taken to determine this factor in both design and assessment of fatigue life.
220
100
t=0
40
20
0
cr bf df bw dw fload P1 P2 σᵤₑₒ kr
Random Variables
100
t=0
Relative Contribution (%)
80 t=100 years
t=300 years
60 t=500 years
40
20
0
cr Aoaa Iobb y1 fload P1 P2 σᵤₑₒ kr e
Random Variables
Notes. cr = corrosion rate, bf = bottom flange width, df = bottom flange thickness, bw = web
depth, dw = web thickness, Aoaa = initial net area of section A-A, Iobb = initial moment of inertia
of the connection plate at section B-B, y1 = vertical distance between the critical rivet hole and
neutral axis, fload = train frequency, P1 = axel load for motor care, P2 = axel load for trailer car,
σueo = engineering ultimate strength, kr = ratio between fatigue strength limitation and ultimate
strength before corrosion, e = horizontal distance between force and centre O.
Since k r and cr are identified as the most effective factors for fatigue failure at mid
girder and the connection plates, further analysis was conducted on these two
parameters. They were varied by ±3 standard deviations to measure the subsequent
221
changes in time-dependent probability of failure. Figure 6-8 shows that, at t = 360
years, the probability of failure increases by 57.27% for girder when k r reduces by
41.7% to 0.18. Also, probability of failure increases by 65.43% for connection plates
when k r reduces by 44.4% to 0.10. The probability of failure increases by 70.79%
and 59.98% when k r increases by 41.9% to 0.44 for girder and by 44.4% to 0.26 for
connection plates.
For cr at t = 360 years, the probability of failure increases by 26.01% for girder and
by 32.77% for connection plates when corrosion rate rises by 45.98% to 12.7
μm/year. Probability of failure reduces by 20.43% for girder and 29.13% for
connection plates when corrosion rate decreases by 47.12% to 4.6 μm/year (see
Figure 6-9). These results further confirm that the values of k r and cr have a major
effect on the accuracy of fatigue life prediction, and more care should be taken to
determine these two factors in both design and assessment of fatigue life.
1
kr=0.44
kr=0.31
0.8 kr=0.18
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
222
1
kr=0.26
kr=0.18
0.8 kr=0.10
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
Notes. kr = ratio between fatigue strength limitation and ultimate strength before corrosion. ρ =
0.9.
1
cr=12.7μm/year
cr=8.7µm/year
0.8 cr=4.6μm/year
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
(a) Fatigue, girder
223
1
cr=12.7μm/year
cr=8.7µm/year
0.8 cr=4.6μm/year
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
(b) Fatigue, connection plate
According to Miner (1945), the critical damage accumulation index, , may not be
exactly one due to the modelling error associated with Miner’s rule and the variation
in mechanical properties. Therefore, it is necessary to investigate the effect of
changes in on probability of fatigue failure. For this, sensitivity of the results was
examined by two other values of (0.7 and 1.3). Figure 6-10 shows the probability
of fatigue failure for various values. It can be seen that, at given time (t = 360
years), the probability of failure at certain years reduces by 57.37% for mid girder
and 90.29% for connection plates when increases from 0.7 to 1.3. This indicates
that the uncertainties in affect the probability of fatigue failure propositionally.
This finding calls for further investigations of the effect of on the probability of
failure.
224
1
Δ=0.7
Δ=1.0
0.8
Probability of failure Δ=1.3
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
1
Δ=0.7
Δ=1.0
0.8
Δ=1.3
Probability of failure
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200
Time (Years)
225
6.5.8 System Reliability Analysis
As mentioned in Section 6.5.2, the reliability analysis of the case study bridge
focused on the longest span. This span includes one mid girder and two edge girders.
The mid girder was analysed against seven failure modes—the flexural, shear,
deflection and fatigue failure of the girder; the tensile and fatigue failure of
connection plates; and the fatigue failure of the rivets. The edge girder was analysed
against five failure modes—the flexural, shear and deflection failure of the girder;
the tensile failure of connection plates; and the fatigue failure of the rivets. It is
imperative to analyse how various failure modes of different girders interact with
each other and, subsequently, how the overall system fails. Therefore, system
reliability analysis was conducted on this critical span.
It was assumed that all failure modes are uncorrelated. Therefore, the probability of
failure for the viaduct system can be calculated as follows:
226
nt
p f 1 [1 P( Fi )] (6.47)
i 1
where P( Fi ) is the probability of failure for each failure mode of each girder
calculated using outcrossing method, p f is the probability of failure for the viaduct
span system and nt is the total number of failure modes, which is 17 for the case
study viaduct.
Based on Equation 6.47, the time-dependent probability of failure calculated for the
viaduct system is shown in Figure 6-12.
1
system reliability analysis
0.8
Probability of failure
0.6
0.4
0.2
0.011%
0
0 200 400 600 800
Time (Years)
(Australian Standard, 2017a), the service life ( Tc ) for the viaduct is 143 years. The
viaduct was constructed in 1889, so the remaining life of the viaduct is 14 years.
Therefore, rehabilitation should be conducted immediately on the case study viaduct
to prevent any potential damage.
227
6.6 Summary
This chapter has proposed a new methodology for prediction of the service life of
corroded steel structures. The models were developed from Chapters 3 and 4 to
predict the changes in mechanical properties (yield strength, ultimate strength, failure
strain and fatigue strength limitation) during corrosion. The resistances of bridges
under different failure modes were modelled as stochastic processes in predicting the
probability of failure for corroded steel structures using the first-passage probability
method.
Based on the time-dependent probability of failure for each failure mode, it was
found that not considering the effect of corrosion on mechanical properties would
also lead to underestimation of the probability of failure for corroded steel structures,
which can be catastrophic. It has also been found that ignoring the auto-correlation of
resistance at each point of time can lead to the underestimation of probability of
failure demands a time-dependent reliability method for prediction of structural
failure. It was found that the service life is 143 years for the case study bridge (based
on its critical span) and fatigue failures at the mid girder and connection plates of the
mid girder are more critical than other failure modes.
A sensitivity analysis was conducted on fatigue failure at the mid girder and
connection plates of the mid girder to identify the factors that affect the service life
the most. It was found that, for both failure modes, the ratio between fatigue strength
limitation and ultimate strength can affect the probability of fatigue failure
significantly at earlier stages and the corrosion rate can affect the probability of
228
fatigue failure considerably in longer term. The strengthening options of the bridge
against corrosion and fatigue damage were also summarised in this chapter. It can be
concluded that the time-dependent reliability method is a rational tool for the
prediction of the service life of corrosion affected steel structures.
229
Chapter 7: Conclusions and Future Work
7.1 Conclusions
This thesis had five research objectives—1) study the effect of corrosion on
mechanical properties, 2) investigate the stress effect on corrosions, 3) investigate
corrosion-induced delamination, 4) investigate the effect of corrosion on fatigue
damage and 5) carry out time-dependent reliability analysis on a corroded bridge.
The first and second objectives were fulfilled in Chapters 3 and 4 through
comprehensive lab and field investigations. The third objective was achieved through
simulated corrosion tests and microstructural studies in Chapter 5. The fifth objective
was achieved in Chapter 6 by out-crossing method and system reliability analysis.
Conclusions drawn from the research and results are below.
Factors that affect the tensile properties during corrosion were studied—
hydrogen concentration, element composition, grain size, phase
composition and the existence of corrosion pits. Models were developed to
predict the changes in tensile properties based on element composition and
grain size. It was found that mild steel is vulnerable to hydrogen
embrittlement. In addition, there is a reduction in iron proportion and
increase in oxygen proportion during corrosion. It was also found that
intergranular corrosion occurs for mild steel during corrosion, which leads
230
to the loss of grain size and attributes to degradation of tensile properties.
In summary, the degradation of tensile properties is caused by 1) the changes
in microstructure, which includes the element component changes and
intergranular corrosion, 2) the hydrogen embrittlement and 3) residual stress
concentration at corrosion pits.
The effect of stress on corrosion was studied. It was found that the elastic
stress exerted in steel can accelerate the corrosion progress by 34–87% in
various corrosive environments. There are three main mechanisms for this
acceleration which can also interact—1) stress can break down the protective
passive oxide film and enhance the dissolution rate of steel; 2) stress causes
deformation of the steel surface, which increases the surface energy and
makes the corrosive solution easier to penetrate the steel surface; and 3) stress
causes the micro-plastic deformation at grain boundaries which leads to
dislocations and slips among grains and, subsequently, reduces the
electrochemical potential in steel.
231
In this thesis, models to predict the changes in hydrogen concentration,
microstructural features and tensile properties based on corrosion loss and
time were developed for stressed and non-stressed steel during simulated
corrosion. Correlations factors were developed for each relationship, which
indicates how many times larger the changes in each factor are in simulated
corrosion than in natural corrosion.
The corrosion effect on the S-N curve was investigated in this thesis. A model
was developed to predict the S-N curve of corroded steel based on corrosion
loss. Fatigue tests were also conducted to develop the full range S-N curves of
steel at different corrosion levels. The S-N curve predicted from the model
was compared with that directly from fatigue test results. The predicted S-N
curve agreed well with the test results, thus the developed model was
verified. Based on the models, the reduction in S at the same N can be up to
23.56% for steel when corrosion loss (thickness loss) increases to 3 mm.
Therefore, it is imperative to consider the changes in S-N curve during
corrosion to prevent the underestimation of fatigue damage.
232
significant in the middle than that in the edge. The delamination can be
prevented by making the steel homogeneous during the manufacturing stage.
Models were developed to predict the sectional area loss and changes in
tensile properties based on corrosion loss within time. Time-dependent
reliability analysis was applied to a case study bridge and the resistances of
bridges under different failure modes were modelled as stochastic processes.
Seven failure modes were considered—flexure, shear, deflection and fatigue
failure of the girder; the tensile and fatigue failure of connection plates; and
the fatigue failure of the rivets.
233
Therefore, further experiments can be conducted to investigate the corrosion
behaviour and corrosion effect for steel subjected to different levels of elastic
stress. For steel under stress and corrosion, stress may cause misorientation of
the grain boundaries and, subsequently, affect the corrosion behaviour. The
effect of stress on grain boundaries misorientation needs to be further
investigated through high resolution EBSD.
234
are hot rolled into different structural shapes (i.e., L shape, I shape and Z
shape) after continuous casting. The hot rolling process leads to the
redistribution of residual stress and recrystallisation of grains, which
subsequently affects the initiation and propagation of corrosion-induced
delamination. Further research needs to be conducted to monitor preferred
corrosion-induced delamination for different structural shapes of steel.
In this research, the effect of corrosion on the S-N curve of steel was studied.
However, further research needs to be conducted to estimate the changes in
microstructural features for steel at different corrosion levels after fatigue
tests. This can help us better understand how cracks are propagated and how
steel grains and phase may change under the combined environment of
corrosion and cyclic loading.
There are two methods to predict the fatigue life of a structure—the S-N
curve approach and linear-elastic fracture mechanics (LEFM) approach. In
this thesis, the S-N curve approach was used to predict the fatigue life of a
steel structure subjected to corrosion It is recommended that the LEFM
approach be conducted for the case study structure in further research to
verify the results from the S-N curve approach in this thesis. It is worth noting
that the time-dependent changes in fracture toughness due to corrosion needs
to be measured beforehand when using the LEFM approach to predict the
probability of fatigue failure of a steel structure. Therefore, further
experiments are suggested to investigate the effect of corrosion on fracture
toughness of steel.
235
consider the correlation between different failure modes and structural
components. Computing system reliability when system components are
correlated is a challenging task and requires further research.
236
References
Abuzeid, O., Zour, M. A., Aljoboury, A. & Alzafin, Y. (2012). Stress corrosion
from Nigeria oil and gas industry. American Journal of Materials Science,
1(2), 108–112.
Lincoln, Nebraska.
28(5), 454–459.
237
character and crystallographic texture studies. Corrosion Science, 51(1), 119–
128.
Aravinthan, A. & Nachimani, C. (2011). Analysis of spot weld growth on mild and
Arioka, K., Yamada, T., Terachi, T. & Staehle, R. W. (2006). Intergranular stress
ASTM International. (2012). ASTM A956-12. Standard test method for Leeb
ASTM International. (2016). ASTM E8/E8M-16a. Standard test methods for tension
www.astm.org.
www.astm.org.
238
ASTM International. (2013c). ASTM E2809-13. Standard guide for using scanning
www.astm.org.
ASTM International. (2017a). ASTM F1113-87. Standard test method for electro-
www.astm.org.
ASTM International. (2013d). ASTM G129-00. Standard practice for slow strain rate
technique. www.astm.org.
Arizio, E., Orsega, E. F., Sommariva, G. & Falcone, R. (2013). Tin amalgam mirrors:
Australian Standard.
239
Australian Standard. (2014). AS 2312.1. Guide to the protection of structural steel
Standard.
Australian Standard. (2009). AS 2758. Aggregates and rock for engineering purposes
Australian Standard.
Australian Standard. (2017a). AS 5100.1. Bridge design, part 1: Scope and general
Australian Standard.
Standard.
Australian Standard. (2017d). AS 5100.6. Bridge design, part 6: Steel and composite
Australian Standard.
240
Bai, Y. & Jin, W. L. (2015). Marine structural design (2nd ed.). Saint Louis:
Elsevier Science.
Bain, E. C. & Paxton, H. W. (1966). Alloying elements in steel. Metals Park, Ohio:
Beidokhti, B., Dolati, A. & Koukabi, A. H. (2009). Effects of alloying elements and
Bentiss, F., Traisnel, M., Gengembre, L. & Lagrenée, M. (1999). A new triazole
studies, weight loss determination, SEM and XPS. Applied Surface Science,
152(3), 237–249.
241
British Standards Institution. (2014). BS 7608. Guide to fatigue design and
Standards Institution.
Calcagnotto, M., Ponge, D., Demir, E. & Raabe, D. (2010). Orientation gradients and
11), 2738–2746.
UK
Chisholm, C., Mook, W., Bufford, D. C., Hattar, K. M., Jungjohann, K. L., Hayden,
242
Collings, D. (2005). Steel-concrete composite bridges. Thomas Telford.
Dordrecht.
of the phases of the corrosion process and use of a modified accelerated test.
De la Fuente, D., Díaz, I., Simancas, J., Chico, B. & Morcillo, M. (2011). Long-term
Digges, T. G., Rosenberg, S. J. & Geil, G. W. (1966). Heat treatment and properties
Bureau of Standards.
243
Djukic, M. B., Zeravcic, V. S., Bakic, G., Sedmak, A. & Rajicic, B. (2014).
Science, 3, 1167–1172.
El Sarraf, R., Iles, D., Momtahan, A., Easey, D. & Hicks, S. (2013). Steel-concrete
Fang, B. Y., Han, E. H., Wang, J. Q. & Ke, W. (2007). Stress corrosion cracking of
X-70 pipeline steel in near neutral pH solution subjected to constant load and
123–129.
acidic media for the oil and gas industry: A review. Corrosion Science, 86,
17–41.
244
Frangopol, D. M., Kallen, M. J. & Van Noortwijk, J. M. (2004). Probabilistic models
Ferhat, M., Benchettara, A., Amara, S. & Najjar, D. (2014). Corrosion behaviour of
Garbatov, Y., Soares, C. G., Parunov, J. & Kodvanj, J. (2014). Tensile strength
303.
Garet, M., Brass, A., Haut, C. & Guttierez-Solana, F. (1998). Hydrogen trapping on
non metallic inclusions in Cr-Mo low alloy steels. Corrosion Science, 40(7),
1073–1086.
Infrastructure Australia.
222.
1–8.
Science Press.
245
Graedel, T. E. & Frankenthal, R. P. (1990). Corrosion mechanisms for iron and low
Huang, C. S., Yeh, Y. K., Liu, G. Y., Hu, H. T., Tsai, K. C., Weng, Y. T., … Wu, M.
Hou, Y., Lei, D., Li, S., Yang, W. & Li, C. Q. (2016). Experimental investigation on
Horner, D. A., Connolly, B. J., Zhou, S., Crocker, L. & Turnbull, A. (2011). Novel
Hu, C., Xia, S., Li, H., Liu, T., Zhou, B., Chen, W. & Wang, N. (2011). Improving
13, 237–244.
assisted stress corrosion cracking of carbon steel in mixed and pure cultures
246
of sulfate reducing bacteria. International Biodeterioration &
Jia, Y. Z., Wang, J. Q., Han, E. H. & Ke, W. (2011). Stress corrosion cracking of
with and without preload. Journal of Materials Science & Technology, 27(11),
1039–1046.
Junior, A. M. J., Guedes, L. H. & Balancin, O. (2012). Ultra grain refinement during
Kreysa, G. & Schütze, M. (Eds.) (2008). Corrosion handbook (2nd ed.). Dechema.
Kerekes, E. & Petrovics, J. (2000). Fatigue estimation for beam structures using
Hungary.
28, 211–284.
Kulak, G. L., Fisher, J. W. & Struik, J. H. (2001). Guide to design criteria for bolted
247
Krupp, U., Kane, W., Pfaendtner, J. A., Liu, X., Laird, C. & McMahon Jr, C. J.
35–41.
Landolfo, R., Cascini, L. & Portioli, F. (2010). Modeling of metal structure corrosion
2404.
Li, C. Q., Firouzi, A. & Yang, W. (2016). Closed-form solution to first passage
119, 2354–2361.
Li, L., Mahmoodian, M., Li, C. Q. & Robert, D. (2018). Effect of corrosion and
248
Li, M. & Cheng, Y. (2008). Corrosion of the stressed pipe steel in carbonate–
Lino, R. E., Marins, Â. M. F., Marchi, L. A., Mendes, J. A., Penna, L. V., Neto, J. G.
50–56.
Lisnianski, A., Frenkel, I. & Ding, Y. (2010). Multi-state system reliability analysis
and optimization for engineers and industrial managers. Springer Science &
Business Media.
Liu, Y., Wang, Z. & Ke, W. (2014). Study on influence of native oxide and corrosion
176.
Luo, H., Dong, C. F., Liu, Z. Y., Maha, M. T. J. & Li, X. G. (2013). Characterization
of hydrogen charging of 2205 duplex stainless steel and its correlation with
Lu, B. T., Luo, J. L., Norton, P. R. & Ma, H. Y. (2009). Effects of dissolved
41–49.
249
Paper presented at the 5th International Conference on Durability of Concrete
Ma, C., Tschauner, O., Beckett, J. R., Rossman, G. R. & Liu, W. (2012). Panguite,
(Ti4+, Sc, Al, Mg, Zr, Ca) 1.8 O3, a new ultra-refractory titania mineral from
Man, O., Pantělejev, L. & Pešina, Z. (2009). EBSD analysis of phase compositions
Mahadevan, S. & Dey, A. (1997). Adaptive Monte Carlo simulation for time-variant
642.
250
Melchers, R. E. & Jeffrey, R. (2008). The critical involvement of anaerobic bacterial
A159–A164.
Moriber, N. (2015). Failed anchor rods on the San Francisco-Oakland Bay Bridge:
http://www.materialsperformance.com/articles/material-selection-
design/2015/11/failed-anchor-rods-on-the-san-francisco-oakland-bay-bridge-
a-corrosion-discussion.
Nelson, L. D., Spence, P. R. & Lachlan, K. A. (2009). Learning from the media in
Nieto, P. J. G., Suárez, V. M. G., Antón, J. C. Á., Bayón, R. M., Blanco, J. Á. S. &
251
Noor, E. A. & Al-Moubaraki, A. H. (2008). Corrosion behavior of mild steel in
Nussbaumer, A., Borges, L. & Davaine, L. (2012). Fatigue design of steel and
Olasolo, M., Uranga, P., Rodriguez-Ibabe, J. & López, B. (2011). Effect of austenite
(6), 2559–2569.
Pardo, A., Merino, M. C., Coy, A. E., Viejo, F., Arrabal, R. & Matykina, E. (2008).
252
Petersen, R. & Melchers, R. E. (2012). Long-term corrosion of cast iron cement lined
no. 23).
Pigozzo, J. C., Espinosa, M. M., Junior, C. C., Arroyo, F. N., Christoforo, A. L., &
Pineau, A. (2015). Crossing grain boundaries in metals by slip bands, cleavage and
p.20140131.
Priestley, M. N., Seible, F., Calvi, G. M. & Calvi, G. M. (1996). Seismic design and
7808.
253
Rancho. R. (2011). Case studies on the rehabilitation of historic bridges. American
Ren, R. K., Zhang, S., Pang, X. L. & Gao, K. W. (2012). A novel observation of the
low carbon steel in 3.5 wt% NaCl solution. Electrochimica Acta, 85, 283–
294.
Revie, R. W. (2008). Corrosion and corrosion control. John Wiley & Sons.
Román, J., Vera, R., Bagnara, M., Carvajal, A. M. & Aperador, W. (2014). Effect of
Rubinstein, R. Y. & Kroese, D. P. (2016). Simulation and the Monte Carlo method
Business Media.
Shanmugam, S., Ramisetti, N. K., Misra, R. D. K., Mannering, T., Panda, D. &
254
properties of Nb-microalloyed steels. Materials Science and Engineering: A,
460, 335–343.
23, 1073–1100.
Shrestha, T., Alsagabi, S. F., Charit, I., Potirniche, G. P. & Glazoff, M. V. (2015).
Shi, Z., Liu, M. & Atrens, A. (2010). Measurement of the corrosion rate of
588.
Shimada, M., Kokawa, H., Wang, Z. J., Sato, Y. S. & Karibe, I. (2002). Optimization
Shin, D. H., Kim, I., Kim, J. & Park, K. T. (2001). Grain refinement mechanism
49(7), 1285–1292.
255
Siriwardane, S., Ohga, M., Dissanayake, R. & Taniwaki, K. (2008). Application of
64(2), 228–237.
Sly, C. (2001). An initial look at the Lowe’s motor speedway pedestrian bridge
Sun, F., Li, X. & Cheng, X. (2014). Effects of carbon content and microstructure on
256
Turnbull, A. (2014). Corrosion pitting and environmentally assisted small crack
p.20140254.
Dekker.
Tromans, D. (1994). On surface energy and the hydrogen embrittlement of iron and
Turnbull, A., McCartney, L. N. & Zhou, S. (2006). A model to predict the evolution
Uhlig, H. H. (2011). Uhlig’s corrosion handbook (vol. 51). John Wiley & Sons.
426(432), 859–864.
51(7), 881–981.
257
Varias, A. G. & Feng, J. L. (2004). Simulation of hydride-induced steady-state crack
Wang, J. J., Zhang, J. P., He, L. & Zhao, Z. M. (2006). Influence of long-term
Wang, X., Tang, X., Wang, L., Wang, C. & Guo, Z. (2014). Corrosion behavior of
X80 pipeline steel under coupling effect of stress and stray current.
Mexico.
258
Zhao, Z. (1995). Primary and deformation-induced high and low cycle fatigue
Zhao, Z., Haldar, A. & Breen Jr, F. L. (1994). Fatigue-reliability evaluation of steel
Zhao, M. C., Yin, F., Hanamura, T., Nagai, K., & Atrens, A. (2007). Relationship
between yield strength and grain size for a bimodal structural ultrafine-
Zhao, Y. Karimi, A. R., Wong, H. S., Hu, B., Buenfeld, N. R. & Jin, W. (2011).
Zhiyuan, Y. Z., Bohai, J., Muye, Y., Zhongqiu, F. & Yuan, T. (2016). Study on
Zhou, Y. & Yan, F. (2016). The relation between intergranular corrosion and
artificial sea water in the presence of sulphide ions. Corrosion Science, 48(2),
522–530.
259