Numerical Tool For Fatigue Life Prediction of Corroded Steel Riveted Connections Using Various Damage Models

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/310482153
NUMERICAL TOOL FOR FATIGUE LIFE PREDICTION OF CORRODED STEEL

RIVETED CONNECTIONS USING VARIOUS DAMAGE MODELS
Thesis · April 2012
CITATIONS READS
12 956
1 author:
Ihab M. El Aghoury
Ain Shams University
30 PUBLICATIONS 109 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Fatigue Life Prediction View project
“Behavior and Design of Ultra-Large Capacity End-Plate Bolted Connections” View project
All content following this page was uploaded by Ihab M. El Aghoury on 19 November 2016.
The user has requested enhancement of the downloaded file.

NUMERICAL TOOL FOR FATIGUE LIFE
PREDICTION OF CORRODED STEEL RIVETED
CONNECTIONS USING VARIOUS DAMAGE MODELS
Ihab El Aghoury
A Thesis
in
The Department of
Building, Civil and Environmental Engineering
Presented in Partial Fulfillment of the Requirements

for the Degree of Doctorate of Philosophy (Civil Engineering) at
CONCORDIA UNIVERSITY
Montréal, Québec, Canada
April 2012
© Ihab El Aghoury, 2012

CONCORDIA UNIVERSITY
SCHOOL OF GRADUATE STUDIES
This is to certify that the thesis prepared

By: Ihab El Aghoury
Entitled: Numerical Tool for Fatigue Life Prediction of Corroded Steel Riveted
Connections Using Various Damage Models
and submitted in partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY (Civil Engineering)
complies with the regulations of the University and meets the accepted standards with
respect to originality and quality.
Signed by the final examining committee:
____________________________________Chair
Dr. R. Raut
____________________________________External Examiner
Dr. J. J. Cheng
____________________________________External to Program
Dr. R. Sedaghati
____________________________________Examiner
Dr. O. Pekau
____________________________________Examiner
Dr. A. Bagchi
____________________________________Thesis Supervisor
Dr. K. Galal
Approved by_______________________________________________________
Dr. K. Ha-Huy, Graduate Program Director
April 5, 2012_______________________________________________________
Dr. Robin A.L. Drew, Dean
Faculty of Engineering & Computer Science
ABSTRACT
Fatigue of structural elements has been a major cause of many catastrophic failures
of steel bridges. Corrosion is considered to be an influential factor that significantly
contributes to the reduction of the fatigue life of steel structures. The objective of this
research is to develop a numerical tool that is capable of predicting the fatigue life of
steel members and connections while accounting for the effects of corrosion. To reach
this goal, the research is divided into three phases.
In the first phase, a stress-life damage accumulation model (called the VTLC, for
Virtual Target Life Curves) is proposed and validated using the experimental work
available in the literature. The model has the advantage of including the effect of
overloading. The VTLC model was verified using a case study of a riveted railway
bridge.
In the second phase, a numerical tool is developed (named CorrFLP for Corrosion
Fatigue Life Predictor). It acts as an add-on and uses available FEM packages as solvers.
This tool is useful in predicting the fatigue life of railway or roadway steel bridges with
riveted or bolted connections. CorrFLP uses the strain-life approach along with the theory
of critical distances. CorrFLP is validated using several test results in the literature with
and without the effect of corrosion.
To account for corrosion, a new fatigue strain-life model based on the Smith-
Watson-Topper model is proposed in the third phase of this research. The model takes
into account the corrosivity of the environment, the stress level, and the corrosive
behaviour of the material used. A method is proposed to standardize a testing protocol to
evaluate some new material constants that describe its behaviour in corrosive
iii
environments. The proposed strain-life method is implemented in CorrFLP and the
resulting fatigue life predictions matched well with reported experimental results of
twenty-four steel beams subjected to various fatigue and weathering conditions.
It is seen that the developed numerical tool, along with the corrosion-fatigue strain-
life based model, would help structural engineers by generating several stress-life design
charts to predict the fatigue life of several commonly-used structural components while
accounting for different levels of corrosion.
iv
To my beloved wife, son, daughter, and family
v
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to my supervisor, Dr. Khaled Galal,
for his guidance, encouragement and patience during my research. I believe that this
research wouldn‟t have been completed without his continuous support. His efforts in
reviewing and correcting this thesis drafts are greatly appreciated.
The financial support of the Canadian National Railway (CN) through their awards
is greatly appreciated. Encouraging awards from Concordia University are also
acknowledged.
Of course I wouldn‟t have reached this stage without the continuous help,
guidance, support and care of my wife, my mother and father in every single stage of my
life. I would like to thank all my professors in Ain Shams University, Cairo, Egypt. Their
great help and support during my undergraduate study really helped me to have a strong
academic background and to continue my studies here in Canada to achieve what I aimed
for. I want to thank my friend Wael El-Haddad for continuously inspiring me during our
studies together. I would like to thank my friend Ahmed Eweda for helping me so much
on the final stages of presenting this thesis. I would also like to thank all my friends in
Montréal for their support for me during my study.
Finally, I dedicate this thesis to my wife, my mother and my father, my sisters and
all my friends and fellow colleagues for their invaluable support during my study. I also
dedicate it to my lovely son and my newly born daughter. Their presence in my life was
the main motivator for me to work hard on my research and achieve my goals.
vi
TABLE OF CONTENTS
CHAPTER 1 INTRODUCTION 1
1.1 Background, problem definition, and need for research 1
1.2 Deterioration of Structures and the Need for Fatigue Life Prediction 2
1.2.1 Methods of Determining Fatigue Life 3
1.2.2 The Need for a Fatigue Life Prediction Software with Corrosion 4
1.2.3 Fatigue of Metals 4
1.2.4 Corrosion of Metals 5
1.2.4.1 Introduction 5
1.2.4.2 Corrosion-Fatigue 7
1.2.4.3 Factors Influencing Corrosion 8
1.2.4.4 Effects of Corrosion on Structural Components 9
1.3 Motivations for this Research 10
1.4 Objective and Scope of Research 12
1.5 Thesis Organization 13
CHAPTER 2 LITERATURE REVIEW 14
2.1 History of Fatigue 14
2.2 Fatigue Testing Approaches 15
2.2.1 Fatigue Evaluation of Riveted Connections 15
2.2.1.1 Fatigue testing without corrosion 16
2.2.1.2 Fatigue testing with corrosion 18
2.3 Evaluation of corrosion 21
2.3.1 Corrosion damage functions 21
2.3.2 Time of Wetness (TOW) 22
2.4 Fatigue Life Prediction Approaches 23
2.4.1 Stress-life method 24
2.4.2 Strain-life method 25
2.4.2.1 Principal strain criterion: 27
2.4.2.2 Von Mises equivalent strain criterion: 28
2.4.2.3 Smith-Watson-Topper relationship: 29
2.4.2.4 Maximum shear strain criterion: 29
2.4.2.5 Brown-Miller combined strain criterion: 30
vii
2.4.2.6 Universal Material Method: 31
2.4.3 Energy-based method 32
2.4.4 Fracture Mechanics Approach 32
2.4.5 Theory of Critical Distances 33
2.5 Cycles Counting 34
2.5.1 Level-Crossing Counting 35
2.5.2 Peak Counting 35
2.5.3 Simple Range Counting 35
2.5.4 Rainflow Cycle Counting 35
2.6 Corrosion fatigue Life Prediction 36
2.7 Review on cumulative fatigue damage models 37
2.8 Review of Current Standards 38
2.9 Remaining Fatigue Life 41
2.10 Review on available finite element fatigue software 42
2.10.1 Introduction 42
2.10.2 Object Oriented Programming 43
2.10.3 Fe-Safe™ (2011) 43
2.10.3.1 MSC.Fatigue™ (2011) 44
2.10.3.2 FemFat™ (2011) 44
2.10.3.3 Ansys™ nCode Design Life (2011) 45
2.10.3.4 Comparison and Proposed tool 46
CHAPTER 3 DEVELOPMENT OF A FATIGUE ACCUMULATIVE-DAMAGE
MODEL 48
3.1 Introduction 48
3.2 Review on Cumulative Damage Models 49
3.2.1 Miner‟s Linear damage Rule (LDR) 49
3.2.2 Marco-Starkey theory 50
3.2.3 Damage theories based on endurance limit reduction 51
3.2.3.1 Corten and Dolan theory 52
3.2.3.2 Manson‟s theories 53
3.2.3.3 S-N Curve modification theories 54
3.3 Factors Affecting Damage Models 55
3.3.1 Mean Stress Effects 55
3.3.2 Effect of Frequency 56
viii
3.3.3 Effect of temperature 56
3.3.4 Effect of corrosion 56
3.3.5 Other effects from previous damage models 57
3.4 Proposed Damage Model 57
3.4.1 Discussion 57
3.4.2 Derivation 60
3.4.3 Overloading effect 65
3.4.4 Proposed procedure 66
3.5 Remaining life calculations 67
3.6 Validation 67
3.6.2 Observations and Conclusions 70
3.7 Extending the methodology 71
3.8 Case Study 71
3.8.1 Geometry 72
3.8.2 Formation of the stress range spectrum 73
3.8.3 Fatigue Life Prediction using VTLC 76
3.8.4 Discussion and Conclusions 77
CHAPTER 4 NUMERICAL FORMULATION OF THE FATIGUE LIFE
PREDICTION TOOL 79
4.1 Introduction 79
4.2 Pre-Processing Module 81
4.2.1 Model Preparation Section 81
4.2.2 Simulation Section 82
4.2.3 CorrFLP‟s User Interface 83
4.2.4 Material Editor 85
4.2.5 Spectrum Editor 86
4.3 Analysis Section 87
4.3.1 CorrFLP‟s Object Oriented Structure 87
4.3.2 Main Mesh Object 87
4.3.2.1 Fatigue Management Object 88
4.3.2.2 Corrosion Management Object 89
4.3.2.3 Temperature Management Object 89
4.3.2.4 Interaction Management Object 90
ix
4.3.2.5 Refinement Management Object 90
4.3.2.6 Element Damage Management Object 91
4.3.2.7 Spectrum Management Object 91
4.3.2.8 Input/Output Management Object 91
4.3.2.9 FE Components Object 92
4.3.2.10 FE Containers Object 93
4.3.2.11 Searching Management Object 93
4.3.2.12 Statistics Management Object 94
4.3.2.13 Mesh Info Management Object 94
4.3.3 Cycle Jumps 94
4.3.3.1 The cycle jump concept 95
4.3.3.2 Calculation of Local Cycle jump 96
4.3.3.3 Choice of the Global Cycle jump 97
4.3.4 CorrFLP‟s Fatigue calculations: 98
4.3.4.1 Fatigue Life Prediction 98
4.3.4.2 Analysis Procedure 101
4.4 Post-Processing Phase 102
4.5 Verification with experimental results 102
4.5.1 Case Study 1: Plate with a circular hole 103
4.5.1.1 Test Description 103
4.5.1.2 Specimen Dimensions and Material Properties 103
4.5.1.3 Finite Element Model 104
4.5.2 Case Study 2: Riveted plates with staggered circular holes 107
4.5.2.1 Test description and geometry 107
4.5.2.2 Materials and Properties 107
4.5.2.4 Interaction surfaces 109
4.5.2.6 Discussion of results 115
4.5.3 Case Study 3: Built-up beams 116
4.5.3.1 Introduction 116
4.5.3.2 Experimental Tests 117
4.5.3.3 Materials 119
x
4.5.3.4 Experimental Test Procedure 120
4.5.3.6 Boundary Conditions and Loading 121
4.5.3.7 Modeling material properties 122
4.5.3.9 Summary 125
4.6 Discussion 125
CHAPTER 5 EFFECT OF CORROSION ON FATIGUE LIFE 127
5.1 Introduction 127
5.2 Implementing the effect of corrosion 128
5.2.1 Geometrical representation of corrosion 128
5.2.2 Proposed corrosion fatigue strain-life model 131
5.2.3 Calibration of the proposed model 135
5.2.4 Implementation in CorrFLP 139
5.3 CorrFLP (Corrosion Module) Verification 140
5.3.2 Material 142
5.3.3 Finite Element Model 143
5.3.4 Fatigue Life Prediction 145
5.3.5 Discussion of results 147
5.3.5.1 Discussion of group 1 results 149
5.3.5.4 Conclusions 150
5.3.6 Standardizing a procedure for determining new material constants 152
5.3.7 Summary 152
5.4 Guidelines for developing corrosion-fatigue design charts 153
5.4.2 Method for generating environmental S-N curves for various riveted details 154
5.4.3 Splice Plates 155
5.4.4 Angles connected to Gusset Plates 157
5.4.5 Angles in built-up Plate Girders 158
5.5 Future work 160
xi
CHAPTER 6 CONTRIBUTIONS, CONCLUSIONS AND
RECOMMENDATIONS 161
6.1 Summary and Contributions of this research 161
6.2 Features of the developed numerical tool: 163
6.2.1 CorrFLP Applications: 163
6.2.2 CorrFLP Advantages: 163
6.2.3 CorrFLP Current Limitations: 164
6.3 Significance of this Research 165
6.4 Conclusions 166
6.5 Recommendations for future work 168
REFERENCES 169
APPENDIX A 177
APPENDIX B 187
APPENDIX C 189
xii
LIST OF FIGURES
Figure ‎2.1 Stable cyclic stress-strain hysteresis loops using Ramberg-Osgood equation
(Dowling, 1999) ................................................................................................................ 26
Figure ‎2.2 Cycle closure and material memory effect (Drapper, 2008) ........................... 27
Figure ‎2.3 Fatigue life of several riveted fatigue details from literature (adopted from
Kulak (2005)) .................................................................................................................... 40
Figure ‎3.1 Possible trends in fatigue damage models (Hwang et al., 1986) ..................... 49
Figure ‎3.2 Load sequence effect in Marco-Starkey theory (1954) ................................... 51
Figure ‎3.3 Corten‟s hypothesis of the progress of fatigue damage (Corten et al., 1956) 52
Figure ‎3.4 Modified S-N curve approach ......................................................................... 54
Figure ‎3.5 The proposed Virtual Target Life Concept ..................................................... 58
Figure ‎3.6 Comparison between the proposed VTLC and previous S-N curve
modification approaches ................................................................................................... 59
Figure ‎3.7 VTLC as a measure of deterioration of structures........................................... 60
Figure ‎3.8 Change in slope in multi-stress level loading using a linear trend ................. 63
Figure ‎3.9 VTLC as a measure of deterioration of structures......................................... 64
Figure ‎3.10 Schematic curve for the expected loss in life at each stress level ................ 64
Figure ‎3.11 Testing Loading Pattern (Corten et al., 1956) .............................................. 67
Figure ‎3.12 Analytical predictions of fluctuating stress amplitude experiments with high
stress of 96,000 psi using three methods: VTLC, Corten, and Miner. ............................. 69
Figure ‎3.13 Analytical predictions of fluctuating stress amplitude experiments with high
stress of 86,000 psi using three methods: VTLC, Corten, and Miner. ............................. 69
Figure ‎3.14 Structural System of the truss bridge showing the investigated member D-36
........................................................................................................................................... 72
Figure ‎3.15 VTL curves for the investigated member D-36 showing the deterioration in
fatigue life with years (assuming the detail is category D) ............................................... 76
Figure ‎3.16 VTL curves for the investigated member D-36 showing the deterioration in
fatigue life with years (assuming the detail is category C) ............................................... 77
Figure ‎4.1 Illustration for the three main modules of CorrFLP and the interaction
between applying damage and cycle jumps ...................................................................... 80
Figure ‎4.2 Schematic drawing for the CorrFLP and ABAQUSTM communication process
........................................................................................................................................... 83
Figure ‎4.3 CorrFLP‟s user interface snapshot .................................................................. 84
Figure ‎4.4 CorrFLP‟s material editor (showing the list of built-in materials library) ...... 85
Figure ‎4.5 CorrFLP‟s spectrum editor (a) Input window, and (b) History graph window 86
Figure ‎4.6 Schematic drawing for the mesh managers in CorrFLP .................................. 88
Figure ‎4.7 Supported formats for the Input Output Management Object ......................... 92
Figure ‎4.8 Components managed by the FE Components Management Object .............. 92
Figure ‎4.9 Containers managed by the FE Containers Management Object .................... 93
Figure ‎4.10 The super cycle and cycle jump concept Paepegem et al. (2001) ................. 95
Figure ‎4.11 Damage extrapolation and corresponding cycle jump .................................. 96
Figure ‎4.12 Choice of global cycle jump based on frequency distribution (Paepegem et
al., 2001) ........................................................................................................................... 97
xiii
Figure ‎4.13 Sub-meshing of critical (parent) elements into fine (children) elements ..... 99
Figure ‎4.14 Propagation of Damage in (children) elements .......................................... 100
Figure ‎4.15 Schematic flow chart for the analysis process showing the “Management
Object” responsible for each task ................................................................................... 101
Figure ‎4.16 Test specimens from Sehitoglu (1983) ........................................................ 103
Figure ‎4.17 The finite element model (Loading on the left, meshing on the right) using
CorrFLP .......................................................................................................................... 104
Figure ‎4.18 Damage during the crack initiation life using CorrFLP .............................. 105
Figure ‎4.19 Predicted lives using CorrFLP vs. experimentally observed by Sehitoglu
(1983) using three different criteria ................................................................................ 106
Figure ‎4.20 Test specimens of Josi et al. (1999)............................................................. 108
Figure ‎4.21 The Finite Element model for series (S0) from Josi et al. (1999) (Meshing on
the left and boundary conditions on the right) using CorrFLP ....................................... 109
Figure ‎4.22 Locations of interaction surfaces in the Finite element model .................... 111
Figure ‎4.23 Section cut in the CorrFLP FE model showing the stress distribution and the
relative sliding between interaction surfaces (Scaled deformed shape for illustration) . 112
Figure ‎4.24 Damage during the crack initiation life using CorrFLP surfaces (Scaled
deformed shape for illustration) ...................................................................................... 112
Figure ‎4.25 Predicted lives using CorrFLP vs. experimentally observed by Josi et al.
(1999) using three different criteria ................................................................................ 115
Figure ‎4.26 Riveted Girders tested by Fisher et al. (1988) ............................................. 118
Figure ‎4.27 Cross-Section of tested girders Fisher et al. (1988) .................................... 118
Figure ‎4.28 Extraction of modeled part in CorrFLP analysis ......................................... 121
Figure ‎4.29 Finite Element Model for Santa Fe Girders (Loading and boundary
conditions on the left, meshing on the right) .................................................................. 122
Figure ‎4.30 Plot for Fatigue crack initiation lives for Santa Fe Girders from the NCHRP
report 302 (1988) vs. those from CorrFLP simulation ................................................... 123
Figure ‎4.31 Reported crack initiation and propagation (Fisher et al., 1988) vs. CorrFLP
damage initiation for specimen SF-2 .............................................................................. 124
Figure ‎5.1 Stress concentration factor due to pitting corrosion depending on pit shape
(Cerit et al., 2009) ........................................................................................................... 129
Figure ‎5.2 Assigning corrosion rate factors for mesh surfaces in CorrFLP ................... 130
Figure ‎5.3 Simulation of corrosion thickness loss on selected critical surfaces ............. 130
Figure ‎5.4 Relationship between total strain amplitude and endurance in a non-corrosive
environment .................................................................................................................... 132
Figure ‎5.5 Relationship between total strain amplitude and endurance in a highly
corrosive environment .................................................................................................... 133
Figure ‎5.6 Plot for the proposed relation between the log of the penetration versus the
proposed corrosion factor  corr . ..................................................................................... 136
Figure ‎5.7 Fatigue strength of two year weathered plain rolled SMA steel specimens
(Kunihiro et al., 1972) ..................................................................................................... 137
Figure ‎5.8 Kunihiro et al. (1972) data points converted into strain-life data points and
fitted ................................................................................................................................ 138
Figure ‎5.9 Corrosion of weathering steel compared with copper-bearing and mild steels
(Albrecht, 1983) .............................................................................................................. 139
Figure ‎5.10 Configurations of tested beams (Albrecht et al., 1994) ............................... 141
xiv
Figure ‎5.11 Test matrix and corrosion rates table adapted from Albrecht et al. (1994) . 141
Figure ‎5.12 Finite element model for the girder (loading and boundary conditions in the
top, meshing and cross-section in the bottom)................................................................ 143
Figure ‎5.13 Plot of  corr   values versus  max /  u . ...................................................... 145
Figure ‎5.14 Predicted lives using CorrFLP vs. experimentally observed by Albrecht et al.
(1994) .............................................................................................................................. 148
Figure ‎5.15 Sample S-N curves for a studied detail for several corrosion rates ............. 155
Figure ‎5.16 The finite element model for “Splice Plate” series ..................................... 156
Figure ‎5.17 Schematic drawing for the test variables ..................................................... 156
Figure ‎5.18 The finite element model for the “Angles connected to Gusset Plates” series
......................................................................................................................................... 157
Figure ‎5.20 The finite element model for the “Built-up Beam” series ........................... 158
xv
LIST OF TABLES
Table ‎2.1 TOW categories according to ISO standard 9223 (1992) ............................. 23
Table ‎2.2 Coefficients for deriving the cyclic coefficients using the Universal Material
Method........................................................................................................... 31
Table ‎2.3 A comparison of features offered by commercial fatigue postprocessors and
the proposed tool CorrFLP. (Maximum is marked with five stars, whereas No
stars corresponds to no implementation). ...................................................... 46
Table 3.1 Tables reported by Kühn, et al. (2008) for estimated damage from (a) 1895-
1980, and (b) 1980-2000 ............................................................................... 74
Table 3.2 Chronologically ordered data from Kühn et al. (2008) report ...................... 75
Table ‎4.1 Observed Results from Josi et al. (1999) vs. minimum and maximum
expected cycles to crack initiation from CorrFLP ....................................... 114
Table ‎4.2 Selected Specimens in this study from NCHRP Report 302 (1988) ........... 119
Table ‎4.3 Material properties of NCHRP report 302 (1988)....................................... 119
Table ‎4.4 Fatigue Crack initiation lives for Santa Fe Girders from the NCHRP report
302 (1988) and from CorrFLP simulation ................................................... 123
Table ‎5.1 Correlation between and the ISO-9224 categories ...................................... 136
Table ‎5.2 Input data for the finite element model ....................................................... 144
Table ‎5.3 Derived Corrosion factors and estimated number of cycles to failure ........ 146
xvi
LIST OF SYMBOLS
a [-] Universal Material method factor
b [-] Fatigue strength exponent
b' [-] Fatigue strength exponent in an NaCL 3.5% corrosive environment
bv [-] Virtual life curve slope
bvi [-] Virtual life curve slope at stress level i
c [-] Fatigue ductility exponent
c' [-] Fatigue ductility exponent in an NaCL 3.5% corrosive environment
C [-] Marco-Starkey power exponent
CL [-] Load cycles per truck
D [-] Accumulated fatigue damage
D0 [-] Previous damage state
e [-] Normal strain on given plane
E [MPa] Elasticity modulus in tension
F [N] External force
g [-] Shear strain on given plane
gr [-] Traffic growth rate
G [MPa] Elasticity modulus in torsion
K' [MPa] Cyclic strength coefficient
KC [MPa.m1/2] Critical stress intensity factor
L [m] Critical length
m [-] Slope of S-N curve
n' [-] Cyclic strength exponent
ni [-] Number of cycles applied under the ith constant-amplitude loading level
N [MPa] Normal stress on given plane
N'i [-] Modified life in Leipholz‟s model
Nf [-] Number of cycles to failure
Nf1 [-] Cycles to failure under constant-amplitude loading level for high stress level
Nf2 [-] Cycles to failure under constant-amplitude loading level for low stress level
Nacc [-] Total accumulated cycles
Nfi [-] Number of cycles to failure under the ith constant-amplitude loading level
NVTL [-] Number of cycles to reach the virtual target life
p [-] Material constant for overstress effect
pa [-] Average annual penetration due to corrosion
rb [-] Ratio of S2/S1
R [-] Stress ratio (low to high stress)
RL [-] Remaining life of a detail
S1 [MPa] High stress level
S2 [MPa] Low stress level
Su [MPa] Ultimate strength
Sy [MPa] Yield strength
Y [-] Number of years
 [-] Corten‟s high to low stress level number of cycles ratio
b [-] Ratio of b' / b
c [-] Ratio of c' / c
i [-] Frequency of cycles in Leipholz‟s model
bi [-] Change in virtual life curve slope in cycle block i.
bDi [-] Change in virtual life curve slope due to overstress loading
bvo [-] Total expected change in virtual life curve slope under constant amplitude loading
Nexpected [-] Difference between the VTL and the S-N curve life at a certain stress level
 [-] Strain range
1 [-] Principal strain range
EFF [-] Effective Von Mises strain range
N [-] Nominal strain range perpendicular to the plane of maximum shear strain range
xvii
max [MPa] Maximum shear strain range
th [MPa.m1/2] Fatigue threshold stress intensity factor
 [MPa] Stress range in fatigue
0 [MPa] Critical stress range in fatigue
 ’f [-] Fatigue ductility coefficient
n [-] Normal strain
 [-] Shear strain
corr [-] Environment corrosivity intensity factor
 [-] Correction factor depending on the maximum applied stress
 [-] Poisson‟s ratio
* [-] Equivalent Poisson‟s ratio
0 [MPa] Fatigue limit in repeated tension
a [MPa] Stress amplitude
1,max [MPa] Maximum principal stress
max [MPa] Maximum normal stress
n [MPa] Normal stress
’f [MPa] Fatigue strength coefficient
H [MPa] Hydrostatic stress
 [MPa] Shear stress
xviii
NOMENCLATURE
AASHTO American Association of State Highway and Transportation Officials
ADT Average Daily Traffic
ADTT Average Daily Truck Traffic
AREA American Railway Engineering Association
DCA Damage Curve Approach
DDCA Double Damage Curve Approach
DLDR Double Linear Damage Rule
ECCS European Convention for Constructional Steelwork
EPFM Elastic Plastic Fracture Mechanics
FE Finite Element
FEA Finite Element Analysis
FEM Finite Element Method
FT Percentage of Truck traffic
FL Percentage of Trucks in the main lane in highways
HCF High-Cycle Fatigue
LCF Low-Cycle Fatigue
LCM longest chord method
LDR linear damage rule
LEFM Linear Elastic Fracture Mechanics
LHS left hand side of equation
LPM longest projection method
MD critical plane set according to Maximum Damage criterion
MS mean stress
MSE mean stress effect
SWT Smith, Watson & Topper method of uniaxial fatigue damage calculation
TOW Time Of Wetness
VTL Virtual Target Life
VTLC Virtual Target Life Curves
xix
1 CHAPTER 1
INTRODUCTION
1.1 Background, problem definition, and need for research
North America has a large inventory of riveted railway and roadway steel bridges'
infrastructure that is rapidly aging and is in continuous need for regular inspection,
monitoring and maintenance. One of the main components of an effective structural
evaluation of steel bridges is to determine the remaining fatigue life of the structure.
Corrosion is considered to be an influential factor that significantly contributes to the
reduction of the fatigue life. The corrosion process is highly accelerated in cold climate
regions where de-icing salt is regularly used. Laboratory fatigue testing is the most
accurate method for determining the fatigue life of structural elements or small
assemblages. The drawback is that such tests are costly, time consuming, usually based
on constant amplitude loading, and being specific to a certain detail. Moreover,
simulating corrosion experimentally along with fatigue loading is challenging, expensive,
and has several limitations. On the other hand, the other alternative for fatigue life
prediction is to carry out stress or strain analysis using the finite element method (FEM)
along with selecting an appropriate damage law. Unfortunately, there is a limitation in the
damage models that account for fatigue-corrosion interaction, thus the fatigue life
prediction using the finite element is limited to cases where there is no severe corrosion.
The following sections provide a brief description on the amount of deterioration
in the existing infrastructure, as well as the needs and means for fatigue life prediction
taking into account the effect of corrosion.
1
1.2 Deterioration of Structures and the Need for Fatigue Life Prediction
Recent survey reported by the American Society of Civil Engineers (ASCE, 2005)
estimates that almost one-quarter of the North America‟s bridge inventory are rated
structurally deficient or functionally obsolete. Over 40% of Canada‟s Bridges are over 40
years old and a significant percentage of them are structurally or functionally deficient
(Lounis, 2007). Existing bridges are experiencing accelerated deterioration due to
increasing traffic demands, higher loads, harsh environment, along with inadequate
maintenance funding.
Fatigue has consistently been a major source of many bridge catastrophic failures
throughout history. On the other hand, corrosion resulted in similar catastrophic failures
of bridges. Back in 1967, Ohio silver bridge collapsed due to a miniature crack caused by
stress corrosion cracking in one of its main eye-bar connections. It is worth mentioning
that the loads at the time of collapse were tripled with respect to the service loads at the
time of construction in a period of 39 years.
There are several initiatives for predicting the fatigue life of different structural
details, yet they are -until now- uni-dimensional, with serious limitations in incorporating
the interaction between sources of fatigue, such as: loading amplitude, loading sequence,
mean stress, stress gradients, residual stresses, temperature, corrosion, and metallurgy of
the metal, pre-existence of micro-cracks. Although fatigue life is affected by several
factors, corrosion is considered to be one of the factors that significantly contribute to the
reduction of the fatigue life. The corrosion process is highly accelerated in snowy regions
where there is excessive use of de-icing salt like Canada. Therefore, there is a need to
2
have an accurate methodology to determine the remaining fatigue life of deteriorating
structures.
1.2.1 Methods of Determining Fatigue Life
Traditionally, fatigue life prediction is determined experimentally, but the
drawback is that fatigue tests are costly, time consuming, and are usually based on
constant amplitude loading and specific to a certain detail. Moreover, experimental work
is generally neither timely nor statistically comprehensive for a majority of designers.
Modern structural design codes of practice usually refer to well-known damage
accumulation rules such as Miner‟s rule (Miner, 1945) to predict the fatigue life of
specific details, by knowing its expected loading history in a simplified empirical way.
All civil engineering codes use the “stress-life” approach, which is based on the net
section stresses and fatigue category of the structural detail in study. That method could
be conservative in some cases or it could be detrimental if the engineer does not know the
rationale behind the choice of the code fatigue limits, or does not have detailed data
regarding the real/expected loading history of the structure. Moreover, codes of practice
do not have detailed provisions or guidelines for the predicting fatigue life of corroding
structural components.
Another alternative approach is to predict the fatigue life of a specific structural
critical component (e.g. a connection) by a strain analysis using the finite element
method. This could be achieved by applying the actual loading along with selecting an
appropriate damage law based on the “strain-life” method, or stress analysis along with
applying the “theory of critical distances”. Currently (2012), the software packages that
3
are capable of such analyses are not widely available and have several limitations that
will be discussed in details in Chapter 2.
1.2.2 The Need for a Fatigue Life Prediction Software with Corrosion
The advancements in the field of numerical simulations of structures using the
finite element method (FEM) in the last few decades encouraged engineers and
researchers to develop post-processing programs to estimate the fatigue life of structural
components. Despite this advancement, there is still a need for more research in the field
of fatigue life prediction of structural components using the FEM. A literature survey of
the advancements and limitations of fatigue life prediction tools will be discussed in
Chapter 2. Until now, up to the author‟s knowledge, there is no evaluation tool that
accounts for the structural response under the combined effects of fatigue and corrosion
using the finite element method. Corrosion is known to accelerate the fatigue process and
reduce the fatigue life of a given detail depending on the severity of corrosion and the
corrosion type (Du, 1998). Hence, there is a need to develop a fatigue life prediction
numerical tool that accounts for the effects of corrosion.
1.2.3 Fatigue of Metals
Fatigue cracking occurs even if the maximum applied stress is less than the elastic
limit of the metal. A fatigue crack usually originates at a location of stress concentration,
such as an existing material/manufacturing flaw or an abrupt geometrical discontinuity in
the material. While almost all metals can exhibit fatigue cracking, structural engineers are
often concerned with the fatigue performance of structural steel as most of fatigue-critical
structures such as bridges, offshore structures and towers are usually made of steel. The
4
dominant variables that influence the fatigue strength of structural steel are the applied
stress range, the number of cycles of applied stress, and the type of structural detail
(Fisher, 1977).
Fatigue failures typically involve minimal levels of plastic deformation.
Consequently, it can be difficult to detect fatigue cracks before fracture of the remaining
cross-section occurs. Many examples of catastrophic failures caused by fatigue cracking
have been documented by Fisher et al. (1987) illustrating that structural engineer must
have a firm understanding of the phenomenon of fatigue. Many civil engineering
structures, such as bridges, cranes or offshore structures are required to withstand the
effects of high cycle fatigue because of the nature of the moving or repeated loads that
they must carry.
1.2.4 Corrosion of Metals
1.2.4.1 Introduction
Corrosion is a process of degradation of a metal by an electrochemical reaction
with its environment. This means a loss of electrons of metals reacting with water and
oxygen. For example, weakening of iron due to oxidation of the iron atoms is a well-
known example of electrochemical corrosion which is commonly known as rust. This
type of damage usually affects metallic materials, and typically produces oxide(s) and/or
salt(s) of the original metal. Corrosion could be categorized into several types
(USACE(EM), 2001):
1- General atmospheric corrosion: is defined as a slow uniform corrosive attack
that results in a uniform thinning spread over a wide area that is not likely to cause
5
significant structural degradation in a short period of time. This type of corrosion can be
easily measured.
2- Crevice corrosion: occurs in narrow openings between two contact surfaces,
such as between adjoining plates or angles in a connection. It can lead to blistering and
failure of the paint system, which further promotes corrosion
3- Pitting corrosion: occurs on bare metal surfaces as well as under paint films. It
is characterized by small cavities penetrating into the surface over a localized area (at a
point). If pitting occurs, it can rapidly accelerate the fatigue damage.
4- Galvanic corrosion: can occur in when steels with different electrochemical
potential (dissimilar metals) are in contact. The corrosion typically causes blistering or
discoloration of the paint and failure of the paint system adjacent to the contact area of
the two steels and decreases as the distance from the metal junction increases.
5- Stray current corrosion: may occur when sources of direct current (i.e., welding
generators) are attached to the structure, or unintended fields from cathodic protection
systems are generated.
6- Filiform corrosion: occurs under thin paint films and has the appearance of fine
filaments emanating from one or more sources in random directions.
7- Erosion corrosion: is caused by removal of surface material by action of
numerous individual impacts of solid or liquid particles and usually has a direction
associated with the metal removal.
8- Cavitation corrosion: is caused by cavitations associated with turbulent flow. It
can remove surface films such as oxides or paint and expose bare metal, producing
rounded micro craters.
6
9- Fretting corrosion: is a combination of wear and corrosion in which material is
removed between contacting surfaces when very small amplitude motions occur between
the surfaces.
10- Stress Corrosion: Corrosion can cause many types of surface micro-cracking,
for example: Stress corrosion cracking (SCC) is the cracking induced from the combined
influence of tensile stress and a corrosive environment. The impact of SCC on a material
usually falls between dry cracking and the fatigue threshold of that material. The required
tensile stresses may be in the form of directly applied stresses or in the form of residual
stresses (KTS, 2010).
In civil engineering applications, there are some common causes for corrosion;
next sections will briefly discuss them.
1.2.4.2 Corrosion-Fatigue
Corrosion-fatigue is the result of the combined action of an alternating or cycling
stresses and a corrosive environment. The fatigue process is thought to cause rupture of
the protective passive film, upon which corrosion is accelerated. If the metal is
simultaneously exposed to a corrosive environment, the failure can take place at even
lower loads and shorter time. The fatigue fracture is brittle and the cracks are most often
transgranular, as in stress-corrosion cracking, but not branched. Usually corrosion-fatigue
cracks are widened by a secondary corrosion reaction. The corrosive environment can
cause a faster crack growth at a lower tension level than in dry air. Even relatively mild
corrosive atmospheres can reduce the fatigue strength of aluminum structures
considerably, down to 75% (or even 25%) of the fatigue strength in dry air (KTS, 2010).
Fatigue cracks usually nucleate from the surface (Argon, 1971). In case of metals in a
7
corrosive environment, the environment produces surface roughening (Hunsche et al.,
1988). Moreover, it is known that pitting is the most common and important form of
corrosion and it was observed that fatigue cracks initiate on surfaces roughened by pitting
corrosion (Kawahara et al., 1988).
Corrosion is among the major factors affecting durability and service life of
bridges, particularly those located in cold regions where de-icing salts and other
aggressive chemicals are used that accelerate the corrosion process. This is common in
Canada.
1.2.4.3 Factors Influencing Corrosion
Many factors could affect the corrosion rate such as the design details, material
properties, maintenance, operation, environment, and coating system. In general, the
primary influencing factors are the local environment and the protective coating system.
The following is brief description of the most important factors.
The pH value of the surrounding environmental significantly affects the corrosion
rate. Corrosion usually occurs at low pH (highly acidic conditions) or at high pH (highly
alkaline conditions). At intermediate pH, a protective oxide or hydroxide often forms.
Deposits of film-forming materials such as oil and grease, and sand and silt can also
contribute to corrosion by creating crevices and ion concentration cells.
The most important substance involved in the atmospheric deterioration of metals
is sulphur dioxide (NACE, 2011). Also, steel structures that are near the sea would be
significantly affected by the presence of chlorides. The corrosion rate of ferrous metals is
determined by two factors, namely the time of wetness and the rate of sulphur deposition.
The corrosion itself is an electrochemical process that operates in the presence of water.
8
Any dissolved air pollutant ions that may be present increase the conductivity and,
therefore, the rate of corrosion (Boden, 1989).
Corrosion of steel increases significantly when the relative humidity is greater than
60%. Corrosion is also aggravated by alternate wet and dry cycles, where it was found
that longer periods of wetness tends to increase the effect (USACE(EM), 2001).
Paint and other protective coatings are the primary preventive measures against
corrosion on hydraulic steel structures. Sharp corners, edges, crevices, weld terminations,
rivets, and bolts are often more susceptible to corrosion since they are more difficult to
coat adequately. Any variation in the paint system can cause local coating failure, which
may result in corrosion under the paint (USACE(EM), 2001).
Snow fighting has a long history. However, the first use of salt for de-icing roads
can only be traced back to the 1930s and it was not until the 1960s that the use of salt in
conjunction with blowing became widespread after winter maintenance personnel learned
of its effectiveness (KTS, 2010).
1.2.4.4 Effects of Corrosion on Structural Components
Corrosion can seriously weaken a steel structure or impair its integrity as it affects
the strength, stability, and serviceability of the structure. A study on the cost and
preventative strategies mandated by the U.S. Congress estimated the total direct cost of
metal corrosion in 26 industrial sectors to be US$276 billion per year (Koch et al., 2002).
The major degrading effects of corrosion on structural members are a loss of cross
section leading to increase in stress levels. Also, accumulation of corrosion products
(rust) at structural connections, for example between the web and the adjacent plates or
9
angles of a built-up section, would cause prying action. This is referred to as corrosion
pack out and results from expansion during the corrosion process. Also localized pitting
corrosion can form notches that may serve as fracture initiation nodes. Also, notching is
known to reduce the fatigue life of a steel member significantly.
1.3 Motivations for this Research
Due to the difficulty and limitations of fatigue life prediction tests, the author was
motivated to develop a numerical tool to predict the fatigue life for any steel assembly
(riveted or bolted) using the finite element method. Although corrosion can drastically
reduce the fatigue life of structural components, this effect has not been thoroughly
studied by researchers in the field of structural engineering and is not explicitly
quantified in almost all design codes of practice. The methodology that the author
adopted to address this shortage is by modifying the strain-life fatigue life prediction
method to account for the effects of corrosion. This can enable designers to estimate the
fatigue life of any structural component in any corrosive environment. This research will
approach the corrosion problem geometrically by modelling the thickness loss and
numerically by the newly proposed strain-life model.
On the other hand, structural engineers normally use the stress-life method along
with the constant amplitude S-N curves for fatigue life prediction. It is not feasible for
most of structural engineers to perform a detailed strain analysis of a structural
component. Moreover, most structural engineers may not have enough detailed
knowledge of the details of the fatigue phenomenon nor the concepts of fracture
mechanics. Thus, for design purposes, the developed numerical tool is intended to bridge
10
the gap between civil engineering and fracture mechanics by being able to generate
several fatigue design charts taking into account the effects of corrosion.
Most of the current structural engineering codes of practice recommend the use of
Miner‟s linear damage accumulation rule in fatigue life calculations due to its simplicity.
It will be further discussed in Chapter 3 that this approach can be sometimes detrimental
to the safety of the bridge structure. Thus, a new stress-based fatigue damage
accumulation model is proposed and verified using experimental test results from
literature.
11
1.4 Objective and Scope of Research
The objective of this research is to develop a numerical tool that is capable of
predicting the fatigue life of steel members, connections, and assemblages while
accounting for the effects of corrosion, and real complex loading patterns. In order to
reach this objective, the scope of this study is to:
1. Propose a new stress-life damage accumulation model based on the review the
existing stress-life based fatigue damage accumulation models.
2. Develop a numerical tool using the finite element method and utilising the strain-
life fatigue life prediction approach along with the Theory of Critical Distances to
predict the remaining fatigue life of different structural details.
3. Verify results obtained from the finite element tool against experimental data and
classical linear fracture mechanics.
4. Extend the numerical tool‟s capabilities to include the geometrical effect of
corrosion.
5. Propose a new strain-life corrosion-fatigue model and propose a standardized
experimental methodology for determining the newly proposed corrosion material
properties.
6. Perform several validation examples versus experimental tests for fatigue and
corrosion to verify the developed numerical tool.
7. Providing guidelines for developing S-N design charts for structural engineers for
predicting the fatigue life of various important riveted and bolted structural details
while accounting for the effect of corrosion.
12
1.5 Thesis Organization
The thesis is arranged as follows: Chapter 2 is a literature review on fatigue life
prediction approaches, available fatigue analysis software, limitations in current codes of
practice. Chapter 3 reviews different stress-life fatigue damage accumulation models and
proposes a new model. Chapter 4 discusses the implementation of a proposed numerical
tool using the strain-life methods and its technical details in addition to three verification
examples. The verification examples highlight the ability of the proposed tool to predict
crack initiation life of any plane or riveted component. Chapter 5 proposes a new strain-
life corrosion fatigue model that incorporates the mean stress effects and also proposes
new corrosion material properties along with an experimental methodology for
determining the newly proposed corrosion material properties. Chapter 5 extends the
features of the numerical tool to include the effect of corrosion in the finite element
simulation with a validation example that highlights the capabilities of the proposed
numerical tool in simulating corrosion. Different conclusions and recommendations are
drawn. Consequently, qualitative guidelines are proposed for deriving environmentally
dependant design charts for several commonly used riveted connection details under
different corrosion levels based on the stress-life approach. Finally, Chapter 6
summarizes this research, its applications and limitations, as well as the conclusions and
recommendations for future work.
13
2 CHAPTER 2
LITERATURE REVIEW
2.1 History of Fatigue
For more than 150 years, numerous research studies were conducted by
mechanical engineers, material scientists, physicists, chemists and mathematicians to
understand the fatigue phenomenon. In 1870, Wohler published his results of fatigue tests
of railway axles and constructed the S-N curves concept. In 1899, Goodman developed a
method for fatigue life calculation at different levels of cyclic stresses. Later in 1924,
Palmgren provided a simple criterion for predicting the extent of fatigue damage induced
by various stress blocks and was later formulated by Miner in 1945 (Miner, 1945), which
was thereafter known as Palmgren-Miner cumulative damage law. In 1970, Elber (1970)
first introduced the crack closure phenomenon which is still a challenging and
controversial topic until now. In 1965, studying of fatigue crack propagation was first
embraced by Paris et al. (1961) namely, the Paris law which relates the rate of advancing
of the fatigue crack width per stress cycle, da / dN , to the range of the stress intensity
factor K . More than 10 years after Paris law, the cyclic J-integral range was proposed
by Dowling and Begley (1976) characterizing the advance of crack under elastic-plastic
conditions. Since then, hundreds of researchers added to the body of knowledge. More
details about fatigue study history are described in (Schutz, 1996). One of the latest
advances in the fatigue and fracture mechanics is the development of the “Theory of
Critical Distances” (TCD) (Taylor, 2008) which sums up the pioneering work done by
Neuber (1958), Peterson (1959) and El Haddad et al. (1979).
14
2.2 Fatigue Testing Approaches
This section describes the fatigue testing as pertinent to structural engineering
applications without getting into the applications within fracture mechanics context.
Many fatigue tests that were reported in literature helped shaping the current S-N curves
(Stress range – Number of cycle‟s relationship) in various codes of practice. Many of
these experimental results were reporting tests of either welded details or riveted details.
As this thesis focuses on the fatigue behaviour of riveted details, a review of previous
tests on riveted connections only will be conducted. For a more detailed review of other
fatigue tests in literature, refer to the survey done by El-Sisi (2009). Fatigue tests reported
in the literature will be divided into two main groups; fatigue tests without studying
corrosion and fatigue tests with corrosion.
2.2.1 Fatigue Evaluation of Riveted Connections
Connections (either riveted or bolted) are the most essential components in non-
welded bridges. It is important to adequately design these connections for fatigue.
Although it is usual in contemporary design for fatigue resistance of bolted connections
to use slip-critical joints, many existing bridges are more likely to have bearing-type
joints that use either rivets or high-strength bolts that usually have staggered holes'
patterns. Fatigue fracture of tension members with bearing-type joints that use staggered
holes is observed to take place on a plane perpendicular to the axis of the member. The
s²/4g rule set by Cochrane (1922) (where s and g are the spacings between the bolt rows
in the direction of the loading and the perpendicular direction, respectively), commonly
used for static strength design of bolted tension members, is not applicable for this case.
15
Moreover, according to Josi et al. (1999) there is no clear definition for what net section
is to be used to calculate the stress range in fatigue. Literature reports many tests that
were conducted to evaluate the fatigue life of riveted connections, as will be discussed in
the next two sections.
2.2.1.1 Fatigue testing without corrosion
Reemsnyder (1975) investigated full scale specimens of gusset plate connections
from an “ore bridge” under fatigue loading. Tests showed that specimens which
originally had rivets that were replaced with high strength bolts at locations of observed
or anticipated cracking, experienced an increase in the fatigue life of about 2 to 6 times
and retarded the crack growth and prevented crack initiation. An increase in bolt
clamping force from 70% to 90% of the bolts yielding strength slightly increased the
fatigue life. Baker et al. (1985) investigated the fatigue life of riveted connections by
conducting fatigue tests on 11 riveted girders. High strength bolts were used to replace
missing rivets. Riveted connections had a fatigue life greater than the referred detail
category D in AASHTO.
Fisher et al. (1989) published a report on fatigue evaluation of riveted bridges with
an extensive review of previous fatigue data on riveted shear splices and full-scale
members. They concluded that differences between riveted details were not significant
and that category D provided a lower bound for crack detection. They also noted that the
failure of the cross section and the loss in loading capacity did not happen until category
C was reached. The results obtained from testing various riveted details were very
scattered and it would be very conservative to assign them all to category D, knowing
that these categories were first created for welded details which are different in behaviour
16
than riveted and bolted details. Mang et al. (1993) investigated methods to determine the
remaining life of riveted structures. They conducted fatigue tests on thirteen full scale
main girders from bridges taken out of service. It was noticed that the previous load
history of the tested specimens did not seem to affect the fatigue life. Tests with high
strength bolts provided positive response of the fatigue performance.
Zhou et al. (1995) investigated the effect of clamping force of rivets and the
influence of hole preparation on fatigue capacity. Tests with constant amplitude limit
were performed at stress levels under 70 MPa. Investigations concerning the cut off limit
were performed with stress ranges from 44.1 MPa to 54.4 MPa with a stress spectra
provided by the Canadian National (CN) North America Railway. A total of 20 tests were
performed, 12 at constant amplitude and 8 with a varied stress range. The result showed
that rivet holes were the most frequent origin for crack initiation and was depending on
the surface of the hole. Other factors were corrosion and welding. Girders with punched
holes provided lower fatigue endurance than drilled or sub punched and reamed. The
fatigue limit where no fatigue damage occurred was determined to be 41 MPa. The
fatigue detail category D of the American Railway Engineering Association (AREA)
(AREA, 1987) provided a lower bound for riveted girders in general and wrought iron
girders exhibited lower fatigue endurance than steel. Category C is equivalent to category
71 in Eurocode (Eurocode3, May 2003).
Al-Emrani (2002) investigated the fatigue endurance of stringers. The fatigue
threshold was investigated at a stress range of 60 MPa. The European design curve C =71
provided a lower bound estimation for the fatigue life of the tested stringers at the stress
range of 100 MPa. The stringers showed structural redundancy and a slow and steady
17
crack propagation, with a rather “ductile” fracture scenario. A total of six stringers were
used in the investigation of the fatigue threshold.
2.2.1.2 Fatigue testing with corrosion
Out et al. (1984) investigated the fatigue resistance of four riveted stringers. The
tests focused on the corroded region of flange angles and the riveted connection between
the web and the angle. Measurements on the girders that were still in service at that time
showed that 1% of the stress cycles exceeded the detail category C. The extreme life
endurance of riveted connections between web and flange were situated near the detail
category C. Fatigue damage from service was found to be negligible. The resistances of
the corroded sections lied between AASHTO detail category E and C depending on loss
of cross section. Riveted beams showed redundancy behaviour due to that stresses
distributed to nearby parts. Tests performed at reduced temperatures at periodic intervals
did not result in unstable crack growth.
Abe (1989) studied the fatigue performance of plates and stringers with a varying
state of corrosion. The fatigue investigations were conducted in tension, with a stress
range varying from zero to the yield strength of the material. The details tested were 5
small scale tests including riveted connections and the tension part of webs and flanges of
9 riveted beams. The investigation concluded that the effect of slight corrosion on riveted
connections did not shorten the fatigue life. Severe corrosion in the riveted connections
shortened the fatigue life. This was believed to be a result of reduced net area
contributing to higher stress concentrations at rivet holes. The effect of the rough surface
due to corrosion was also believed to influence the results. The previous loading history
did not affect the fatigue life of the investigation.
18
Fisher et al. (1990) conducted an extended literature survey with results from over
1200 fatigue tests. Fourteen full-scale tests on riveted girders fatigue endurance were
performed with test conditions varying from room temperature to -73 oC. Despite the low
test temperatures, crack growth or premature fracture did not seem to be affected. The
literature survey showed that tests with high stress range seemed to provide low fatigue
life endurances, because of yielding in the material. Plates with empty holes had a better
ability to endure fatigue than riveted joints. Results from the investigation on full-scale
tests were that girders without severe corrosion, developed cracks in the net section at a
rivet hole. Corrosion notch effects made the beams develop fatigue cracks at the gross
cross section. The cracks usually formed at rivet holes unless more than 20% of the gross
cross section was decreased due to corrosion.
Forsberg (1993) investigated the fatigue life of corroded steel plates. Tests were
retrieved from corroded 55 beams with varying state of corrosion. In service stage, the
stress range was between 20 to 30 MPa and the beams had endured approximately 107
cycles. Fatigue tests were done under both varied and constant stress range. Six
specimens were included in the investigation, retrieved from the tension flange. The state
of corrosion varied from light to heavy, with very rough surfaces. The effect of minor
corrosion didn‟t seem to affect the fatigue life, but a more severe state of corrosion made
the fatigue performance drop drastically, below the detail category C.
Adamson et al. (1995) investigated the fatigue behaviour of stringers retrieved
from a bridge built in 1911. It was concluded from the load history and strain
measurements that the fatigue damage was negligible. Presence of corrosion on stringers
was also believed to have a negligible effect on the fatigue performance. The
19
investigation included five full scale tests on stringers. None-bearing riveted details
showed a tendency of having fatigue resistance higher than riveted connections designed
to act in bearing. The redundancy of riveted structures added significantly to its fatigue
resistance. The results of the fatigue endurance of the stringers were covered by the detail
category D in American Association of State Highway and Transportation Officials
(AASHTO). DiBattista et al. (1995) investigated the fatigue performance of tension
members from the truss girder bridge investigated by Adamson et al. (1995). A uniform
corrosion existed on all tension members. Stress ranges in the tests were selected to
provide information near the category C and D fatigue limit American Association of
State Highway and Transportation Officials (AASHTO). A total of seven full scale tests
were performed. No accumulated fatigue damage was present due to previous load
history, based on measured strains while in service and from inspections. The tests
showed that the fatigue resistance of diagonals and their connections to the bottom chord
panel could be evaluated by detail category D American Railway Engineering
Association (AREA, 1987), depending on definition of net section area. Repair of
cracked tension members to the gusset plate with preloaded bolts extended the life of the
connection significantly.
Xiulin et al. (1996) investigated fatigue tests on plates with removed rivets. The
investigation included tests of 28 small scale samples. The plates were retrieved from
tension chords in a bridge. Results from the investigation were comparable with results
available in literature. The initiation phase of the fatigue cracking occupied the major part
of the fatigue life of the material investigated.
20
DeJong et al. (2009) recently investigated the corrosion-fatigue resistance of
coupons of MMFX microcomposite and 316 LN stainless steel. They performed several
cyclic tension tests with overloads both in air and in an aqueous 3.5% NaCL solution to
compare the fatigue resistance of these materials. They showed that corrosion fatigue
reduced the periodic overload performance of both materials although they retained their
intrinsic fatigue limit. Under constant amplitude loading, MMFX had a reduced
performance in corrosive environment while the 316 LN stainless steel did not show
much reduction (except for high loads).
2.3 Evaluation of corrosion
Corrosion and fatigue are two major factors that contribute to aging of structures.
These two factors can act separately, sequentially or simultaneously (Du, 1998). Few
approaches were done by Doyle et al. (1990) using Laser-based profilometry to locate
and measure corrosion fatigue cracking or using Laser speckle sensors to measure surface
roughness to study corrosion fatigue crack initiation. The influence of corrosive
environment on fatigue has been studied by many researchers such as Gangloff (1990)
and can be further referred to by the state-of the-art review by Duquete (1998).
2.3.1 Corrosion damage functions
The term "damage function" denotes a mathematical dose-response function
connecting material damage to the factors involved in the damaging process. Many of
these damage functions are listed by Boden (1989). The input of such damage functions
is the concentration of harmful substances, time of wetness, temperature, etc. The output
of such functions can have many forms such as weight loss, loss in thickness, etc). These
21
damage functions are widely used to describe atmospheric corrosion and atmospheric
deterioration. Moreover, the ISO standard 9223 classifies the corrosivity of an
atmosphere based on measurements of time of wetness (TOW), and pollution categories
(sulphur, airborne chlorides). The standard was not intended to be used in extreme
service atmospheres such as those within chemical or metallurgical processing facilities
or where there is direct contact with salt spray (NACE, 2011).
Only airborne chlorides and sulphur dioxide are considered in terms of classifying
the pollution, this gives good coverage of rural, urban, industrial and marine atmospheres.
Based on these measures an atmosphere is classified as being in one of five categories in
terms of its corrosivity using two types of units, i.e. short term corrosion rate (CR) of
steel as g.m-2.year-1 (one year) or m/year (twenty years):
2.3.2 Time of Wetness (TOW)
From the fundamental theory, the time of wetness (TOW) of a corroding surface is
a key parameter, directly determining the duration of the electrochemical corrosion
processes. This is a complex variable, since all the means of formation and evaporation of
the surface electrolyte solution must be considered. The TOW refers to the period of time
during which the atmospheric conditions are favourable for the formation of a surface
layer of moisture on a metal or alloy (KTS, 2010). This moisture film is extremely
important from the point of view of the chemical mechanisms of the corrosion process.
For the purposes of the standard this has been defined as the time period during which the
relative humidity is in excess of 80% and the temperature is above 0 degrees Celsius.
This measure can either be determined from weather data or measured directly through
various means.
22
TOW categories range from "Internal microclimates (T1) with climatic control" to
"Part of damp climates, unventilated sheds in humid conditions (T5)." TOW units
are hours per year when relative humidity (RH) > 80% and the temperature > 0oC.
Table 2.1 lists the different TOW categories.
Table 2.1 TOW categories according to ISO standard 9223 (1992)
Time of Wetness (TOW) Designation

TOW ≤ 10 T1
10 < TOW ≤ 250 T2
250 < TOW ≤ 2,500 T3
2,500 < TOW ≤ 5,500 T4
5,500 < TOW T5
The TOW is strongly dependent on the critical relative humidity. Apart from the
primary critical humidity (associated with clean surfaces); secondary and even tertiary
critical humidity levels may be observed where the corrosion rate increases abruptly.
Hygroscopic corrosion products and capillary condensation of moisture in corrosion
products are thought to account for these effects. Other sources of surface electrolyte
include chemical condensation (by chlorides, sulphates and carbonates), adsorbed
molecular water layers and direct moisture precipitation (ocean spray, dew, rain).
2.4 Fatigue Life Prediction Approaches
Numerical techniques for fatigue life prediction could be classified into two major
approaches: empirical correlation approach, and fracture mechanics approach (Ellyin,
(1997); and Dowling (1999)). In the empirical correlation approach, a damage parameter
„D‟ is used to present actual fatigue test results, in which fatigue life calculation is
generally performed with respect to either final fracture or crack initiation. The
23
application of fracture mechanics approach is widespread, especially for crack
propagation life (Chen et al., 2005). In order to predict the fatigue life under a specified
condition, different empirical damage parameters, D, have been proposed to correlate
with the fatigue life (number of cycles to failure, Nf ). The empirical correlation approach
is generally divided into three categories, i.e., “stress-life” method, “strain-life” method
and energy-based method, where stress, strain or energy is used as the damage parameter
D, respectively, for each method.
2.4.1 Stress-life method
The “stress-life” method uses the alternating stress amplitude to predict the number
of cycles to failure. This method is based on comparing the stress amplitude to a stress
versus fatigue life curve (S-N curve), which comprises the influence of material,
geometry and surface condition (Radaj et al., 1998). The S-N curves are based on
empirical formulas derived from experimental data. The “stress-life” method is generally
used for high cycle fatigue (HCF) where the material response is mostly elastic such as in
case of the analysis of bridges.
The “stress-life” method emphasizes nominal stresses, rather than local stresses
and strains, and it normally employs elastic stress concentration factors and empirical
modifications to account for the concentration effect of notches (Leis et al., 1973).
However, the accuracy of life prediction depends heavily on an accurate evaluation of the
nominal stress range and the stress concentrations at the fatigue detail. It has been shown
that, in some cases, “stress-life” methods predict fatigue lives that differ from test values
by more than two orders of magnitude (Everett, 1992).
24
2.4.2 Strain-life method
The “strain-life” method uses true strain to predict the number of cycles to failure.
When components are under high load and/or have critical locations (notches and/or
bolted holes), the stress-strain relationship is no longer linearly related. In such situations,
the plastic strain becomes a significant part of the deformation. Since the primary
mechanism in fatigue is plastic deformation, an elastic model is not appropriate. The
“strain-life” method has found wide applications in fatigue analysis, especially for fatigue
crack initiation life calculation. This method is currently considered as a comprehensive
method that can be used instead of “stress-life” methods (Dowling, 1999). In contrast to
the “stress-life” method, the “strain-life” method considers the plastic deformation that
may occur in localized regions where fatigue cracks initiate. The strain-based method
assumes the material in highly strained areas, such as at a notch root, behaves similarly to
material in a smooth specimen under cyclic strain controlled loading with the same strain
(Morrow et al., 1981). Thus, this method can account directly for the difference in stress
concentrations among different fatigue details through an inelastic finite element
evaluation of the strain distribution. Dowling (1982) has reported that the estimated life is
not very sensitive to the calculated strain, thus making the use of the local strain method a
useful tool for the prediction of crack initiation life.
The strain-based method uses a cyclic stress versus strain curve and a strain versus
life curve instead of the S–N curve used in “stress-life” method. The coefficients and
exponents that define these curves are treated as fatigue properties of the material. At the
early developmental stages for the technique, there were insufficient fatigue data to
quantify the fatigue properties of many engineering metals and various equations were
25
proposed to correlate the fatigue properties to the tensile properties (Morrow et al., 1981).
The widespread adoption of closed-loop mechanical testing systems and the development
of the “strain-life” method have largely eliminated the need for these empirical equations
and there is an abundance of data defining the fatigue properties of numerous engineering
metals (Rice et al., 1988). For engineering materials at room temperature (same as the
case in laboratory tests), cyclic hardening or softening is usually rapid at first and then
approaches a stable condition. The stable cyclic stress versus strain curve is often defined
using the Ramberg-Osgood equation (Dowling, 1999). Figure 2.1 shows how the stable
cyclic stress-strain passes through the tips of the hysteresis loops.
Figure 2.1 Stable cyclic stress-strain hysteresis loops using Ramberg-Osgood equation
(Dowling, 1999)
Whenever a smaller range cycle follows a bigger one, this causes a closed loop that
doesn‟t affect the main stable cyclic curve; this is what is sometimes referred to as the
material memory effect (Drapper, 2008), this phenomenon is illustrated in Figure 2.2.
26
Figure 2.2 Cycle closure and material memory effect (Drapper, 2008)
The following sections introduce several fatigue life criteria based on the strain-
based method.
2.4.2.1 Principal strain criterion:
This criterion proposes that fatigue cracks initiate on the planes experiencing the
largest principal strain amplitude. This criterion is recommended for the analysis of
brittle metals like cast iron and some very high strength steels. In general, it tends to give
unsafe life estimates for ductile metals (Drapper, 2008). The endurance using the
principal strain criterion can be obtained from equation (2-1):
1  f
'
 2 N f b   'f 2 N f c (2-1)
2 E
where 1 is principal strain range,  'f is the fatigue strength coefficient,  'f is
the fatigue ductility coefficient, E is the elastic modulus, b is the fatigue strength
27
exponent (Basquin‟s exponent) and c is the fatigue ductility exponent (Coffin-Manson
exponent).
2.4.2.2 Von Mises equivalent strain criterion:
Since the von Mises criterion provides an estimate of the onset of yielding, this
criterion has been proposed for fatigue life estimation. The major problem with the von
Mises criteria is that the effective von Mises strain  EFF is always positive. Some
approximations have been proposed to assign the sign based on hydrostatic stress or
strain which makes this criterion named signed von Mises. The endurance using the
principal strain criterion can be obtained from equation (2-2):
 EFF  f
'
 2 N f  b   'f 2 N f  c (2-2)
2 E
where

 EFF   1   2 2   2   3 2   3  1 2  0.5
(2-3)
where  is a proposed term based on an equivalent value of Poisson‟s ratio 

*
based on elastic and plastic strains and Poisson ratios.
1  E  E   P P
 where  * 
1   
*
2 T (2-4)
28
Von Mises criterion correlates poorly with test data, especially for biaxial stresses
when the two in-plane principal stresses change their orientation during the fatigue
loading (Drapper, 2008).
2.4.2.3 Smith-Watson-Topper relationship:
As an approach to include the effect of mean stresses, Smith-Watson-Topper
(1970) modified the principal strain criterion to include the mean stress as shown in
equation (2-5):

 max 
  2 N 
 'f
2
2b
  'f  'f 2 N f 
b c
(2-5)
f
2 E
This method was extended by Socie and Bannantine (1989) for brittle metals and
shown in equation (2-6):
1
 1, max 
  2 N 
 'f
2
2b
  'f  'f 2 N f 
bc
(2-6)
f
2 E
where 1 is principal strain range,  1,max is the maximum principal stress,  'f is
the fatigue strength coefficient,  'f is the fatigue ductility coefficient, E is the elastic
modulus, b is the fatigue strength exponent (Basquin‟s exponent) and c is the fatigue
ductility exponent (Coffin-Manson exponent).
2.4.2.4 Maximum shear strain criterion:
The maximum shear strain criterion proposes that the crack will initiate on planes
which experience the maximum shear strain amplitude (causing the slip bands). The
principal strain criterion would be rewritten to replace the principal strain with the
29
maximum shear strain  max . This criterion tends to give conservative life estimates for
ductile metals, but can give unsafe life estimates for brittle metals (Drapper, 2008). The
criterion is shown in equation (2-7):
 max  '
 1.3 f 2 N f   1.5 'f 2 N f 

b c
(2-7)
2 E
where  max is maximum shear strain range,  'f is the fatigue strength
coefficient,  'f is the fatigue ductility coefficient, E is the elastic modulus, b is the
fatigue strength exponent (Basquin‟s exponent) and c is the fatigue ductility exponent
(Coffin-Manson).
2.4.2.5 Brown-Miller combined strain criterion:
Same as the maximum shear strain criterion, the Brown-Miller criterion (Kandil et
al., 1982) assumes that the fatigue damage occurs on the plane which experiences the
maximum shear amplitude, but additionally this function incorporates both shear strain
and strain normal to this plane. The Brown-Miller criterion gives realistic life estimates
for ductile metals and tends to be non-conservative for brittle metals. The criterion is
shown in equation (2-8):
 max  N  '
 1.65 f 2 N f   1.75 'f 2 N f 

b c
 (2-8)
2 2 E
where  max is maximum shear strain range,  N is the normal strain range
perpendicular to the plane of maximum shear strain range,  'f is the fatigue strength
coefficient,  'f is the fatigue ductility coefficient, E is the elastic modulus, b is the
30
fatigue strength exponent (Basquin‟s exponent) and c is the fatigue ductility exponent
(Coffin-Manson exponent).
2.4.2.6 Universal Material Method:
If the cyclic properties of a steel material or an aluminum material are not present,
the Universal material method can be used to obtain cyclic properties from static
properties. Baumel and Seeger (1990) have published this method for plain carbon and
low to medium alloy steels and also for aluminum and titanium alloys. Experience shows
that this method generally gives satisfactory agreement with measured materials
properties (Drapper, 2008). The method can be summarized as shown in Table 2.2.
Table 2.2 Coefficients for deriving the cyclic coefficients using the Universal Material
Method.
Plain Carbon Steels Aluminum and Titanium alloys

 'f  1.5 u  'f  1.67 u
 'f  0.59a *  'f  0.35
b  0.087 b  0.095
c  0.58 c  0.69
n '  0.15 n '  0.11
K '  1.65 u K '  1.61 u
* a  1.0 for  u / E  0.003, otherwise a  1.375  125 u / E
31
2.4.3 Energy-based method
Experimental observations have confirmed the significant role that plastic
deformation plays in the fatigue damage process. As cyclic plastic deformation is related
to slip along crystallographic planes and dislocation movement, cyclic stress is related to
the resistance to such movement at the microscopic level and strain energy is dissipated
during such irreversible deformations (Ellyin, 1997). The energy-based method uses
energy as a damage parameter to characterize fatigue, emphasizing the interrelation
between stress, strain, and the fatigue damage process. It unifies high and low cycle
fatigue, and has the potential to bridge fatigue data obtained in different laboratories
using specimens of different geometry and size and tested under different controls. This
method is not in the scope of this research.
2.4.4 Fracture Mechanics Approach
The use of fracture mechanics in fatigue propagation life prediction became
widespread since it was first applied to fatigue crack growth about 50 years ago (Paris et
al., 1961). For many structures, the major portion of the fatigue life is expended in
propagating a crack from an existing flaw, that is, only the fatigue crack propagation life
needs to be determined.
The parameter describing the stress field around the advancing crack tip is an
important component in the fracture mechanics approach. The stress intensity factor, K, is
used in Linear-Elastic Fracture Mechanics (LEFM) when the nominal stress versus strain
response is essentially elastic. When plasticity effects are considered, various parameters
have been proposed, among which Crack Tip Opening Displacement (CTOD) and J-
integral are the most commonly used in Elastic-Plastic Fracture Mechanics (EPFM). In
32
highly ductile materials and where the crack tip plastic zone is large, EPFM may be more
appropriate. Crack propagation life calculations are carried out from a specific initial
crack size to a final crack size at failure, which may be determined from the material
fracture toughness. However, a number of parameters are difficult to determine in
practice, especially the initial crack size and shape. The problem is that concepts of
LEFM are not accurate for short cracks (usually less than 1mm). Thus the crack initiation
phase cannot be accurately captured (Taylor, 2008). The theory of critical distances
bridges the gap between the short crack regime and long crack regime that can be
predicted by LEFM.
2.4.5 Theory of Critical Distances
One of the latest advances in the fatigue and fracture mechanics is the development
of the “Theory of Critical Distances” (TCD) (Taylor, 2008) which sums up the
pioneering work done by Neuber (1958), Peterson (1959) and El Haddad et al. (1979).
The theory applies for short cracks and assumes that there is an intrinsic characteristic
critical length “L” for each material, where if the average stresses at a distance L/2 away
from the surface crack or notch exceeds a material constant stress which is usually near
its ultimate stress, the crack would propagate, otherwise it can be considered as a non-
propagating crack. Moreover, if the surface crack length is less than this critical distance
“L”, LEFM would not accurately apply. The critical distance “L” can be expressed as
follows:
2
1K 
L   C  (2-9)
  o 
33
where K C is the critical stress intensity factor of the material in case of static
loading and  o is the critical stress which is usually equal to the strength of the material
as measured on unnotched samples. The same equation can be extended to fatigue
loading as follows:
2
1  K th 
L   
   o  (2-10)
where K th which is the fatigue threshold intensity factor in case of fatigue
loading.  o is the critical stress range which is usually equal to the fatigue limit. This is
true for metal fatigue and for static fracture of brittle ceramics (Taylor, 2008). The
current availability of robust FEA solvers, makes this method a promising method for
fatigue crack initiation and propagation analysis.
2.5 Cycles Counting
Counting the number of cycles in a given loading history is very important for
fatigue life calculations. Because the stress pattern obtained from traffic data does not
have a constant amplitude, a systematic manner of counting stress cycles is needed.
Various cycle-counting algorithms have been developed for the purpose of reducing
complex histories into a finite number of variable amplitude cycles. Among the earliest of
these are the level-crossing counting, peak counting, and simple range counting
techniques and the rain flow analysis. The following subsections explain these counting
methods.
34
2.5.1 Level-Crossing Counting
Level-crossing counting involves dividing the stress axis into an arbitrary number
of equal increments. A reference stress is initially chosen, and each time a positively
sloped portion of the stress record crosses an increment above the reference stress a count
is recorded for that particular increment value. Likewise, each time a negatively sloped
portion of the stress history crosses an increment value below the reference stress, a count
is recorded. Then, the counts are combined to form full cycles. Usually this is done by
combining the counts to form the largest possible cycle, and the remaining counts are
combined to form the next largest possible cycle. When all counts have been assigned to
a cycle, the process is completed.
2.5.2 Peak Counting
Peak counting also involves dividing the stress axis into increments and choosing a
reference stress. All local maxima above the reference stress and all local minima below
the reference stress are recorded. Then, these counts are combined by sequentially
grouping the greatest maxima with the least minima to form complete cycles.
2.5.3 Simple Range Counting
Simple range counting involves recording the range between successive stress
reversals and counting each range as one half-cycle.
2.5.4 Rainflow Cycle Counting
There are certain limitations to the cycle-counting techniques described above.
Mainly, each of these methods disregards the actual sequence of applied stress cycles.
35
Consequently, in the late 1960s, a new type of cycle counting algorithm was introduced.
It attempted to identify closed hysteresis loops in the stress-strain response of a material
under cyclic-loading. The term rainflow counting has been applied to the general family
of such algorithms. Each stress record from the field testing must be processed through a
rainflow cycle-counting algorithm.
The current rainflow cycle counting algorithm used in engineering is originally
attributed to Downing et al. (1982). Some pre-processing of the field stress history data is
required before this algorithm can be applied. As with all rainflow counting algorithms, it
is first necessary to re-arrange the stress record so that it begins and ends with the stress
value of the greatest magnitude. This ensures that no half-cycles are counted. This
method is the most commonly used method in all fatigue signal processing software.
2.6 Corrosion fatigue Life Prediction
Corrosion-fatigue behaviour of a given material-environment system refers to the
characteristics of the material under fluctuation loads in the presence of a particular
environment. Corrosion-fatigue damage occurs more rapidly than would be expected
from the algebraic sum of the individual effects of fatigue, corrosion, or stress corrosion
cracking. Fatigue crack initiation threshold K th in a non-corrosive environment can be
related to the yield strength of some materials (Barsom et al., 1999). In corrosive
environments, the fatigue threshold varies and can be related to the yield strength too, but
not all experimental evidences on all tested materials support this observation. Novak
(1983) has investigated the corrosion fatigue crack initiation (CFCI) behaviour for four
types of steel (the A36, A588-A, A517-F, and V-150 steels). His main observation was
36
that there is no clear corrosion fatigue crack initiation threshold K th for the four types
of steel after corrosion, in contrast to fatigue crack initiation thresholds clearly obtained
in non-corrosive environment.
Further research is needed to establish the relationships, if any, between the
properties of the material and the corrosion-fatigue threshold at different corrosive
environments. It is important to mention that the corrosion fatigue threshold K th varies
with the applied stress ratio R, loading frequency, for each environment-material system
which makes the corrosion fatigue phenomenon more complex and needs further
experimental investigations.
2.7 Review on cumulative fatigue damage models
Fatigue damage increases with applied load cycles in a complex cumulative
manner. Cumulative fatigue damage analysis plays a key role in the fatigue life prediction
of components and structures subjected to field load histories such as bridges. Due to the
complexity of the fatigue phenomenon and in order to assist designers in considering the
several factors involved in the fatigue process such variable loading histories, mean stress
effects, effects of multi-axial loading, many fatigue damage model were proposed
throughout the past few decades. Since the introduction of damage accumulation concept
by Palmgren (Palmgren, 1924) about 70 years ago and „linear damage rule‟ by Miner
(Miner, 1945) about 50 years ago, the treatment of cumulative fatigue damage has
received increasingly more attention. Consequently, many damage models have been
developed. A more detailed and comprehensive review will be presented in Chapter 3. A
new damage model will be proposed and compared with other approaches.
37
Due to the previously mentioned difficulties of the variable amplitude fatigue
testing, results of constant amplitude fatigue testing are to be extrapolated towards
variable amplitude loading using cumulative damage laws. For further state-of-the-art
reviews on different fatigue damage models, refer to the comprehensive work done by
Fatemi and Yang (1998), and Hwang and Han (1986).
2.8 Review of Current Standards
The American Association of State Highway and Transportation Officials
(AASHTO) produced a guide for fatigue evaluation of existing steel bridges based upon
an extensive report by Moses et al. (1987). It was intended that the guide eventually be
included in the AASHTO Manual for Maintenance Inspection of Bridges (AASHTO,
1989). The guide focused on highway bridges and traffic loadings, and it provides
guidelines for calculating the remaining fatigue life as either the remaining mean fatigue
life or as the remaining safe fatigue life. The remaining mean life has a 50% probability
of being exceeded, and it is considered the best estimate of the remaining fatigue life. It is
generally used for cost comparisons and estimates.
The final section of the AASHTO guide (AASHTO, 1989) provides alternative
solutions if the calculated remaining fatigue life is inadequate. The solutions provided
can be either to restrict the traffic that uses the bridge, to modify the bridge to eliminate
or extend the life of the critical detail, or to institute inspections of the critical details so
as to enable timely discovery of any crack growth. The AASHTO guide recommends
using the fatigue resistance category D for the base metal at the net section of a riveted
connection. Fisher et al. (1987) recommended changes to the 1983 version of the
AASHTO Manual for Maintenance Inspection of Bridges to the effect that riveted steel
38
members that resist net section tensile stress by three or more components (e.g., one web
and two flange angles) can be checked according to category C for fatigue resistance.
This is based upon their conclusions that category D identifies fatigue crack development
and category C characterizes fatigue resistance, defined as the ability of a member to
carry load with one or two cracked components.
Fatigue evaluation of existing railway bridges is covered in the AREA (American
Railway Engineering Association) standard (AREA, 1987) which is now recognized as
the American Railway Engineering and Maintenance-of-Way Association (AREMA).
AREMA (AREMA, 2011) recommends the use of category D for members with riveted
or bolted connections with low slip resistance. However, this may be increased to
category C if the engineer can verify that the rivets have developed normal clamping
force. No guidance is provided for making such verification. The reasoning behind this
benefit is that category D was defined with riveted connections that had low clamping
force, and that rivets with high clamping force are better designed by category C. Hence a
higher fatigue resistance is permissible for rivets with normal clamping force.
More stringent fatigue category requirements are specified for fracture critical
members. i.e. members whose failure would make the bridge unable to fulfill its intended
service. Riveted members are not considered to be this type of member because of their
internal cross-sectional redundancy. However, if the riveted members do not satisfy the
fatigue category requirements, the requirements may be waived if it can be deemed that
the members have adequate structural redundancy to redundancy the load when one of
the components cracks. Inspections must be frequent enough to discover the local failure
and to perform repairs. There is no specific category for the riveted members or the
39
riveted connections. category D can be used as a lower bound for the evaluation of
riveted members. There are no S-N curves for fatigue in corrosive environments in the
aforementioned codes of practice.
The Canadian Highway Bridge Design Code (CHBDC S6, 2006) also uses detail
category D for riveted connections, but there is a big scatter in the results that were used
for choosing this detail category as shown in Figure 2.3.
Figure 2.3 Fatigue life of several riveted fatigue details from literature (adopted from
Kulak (2005))
40
2.9 Remaining Fatigue Life
This section discusses how the remaining fatigue life is calculated in years from
the fatigue life in load cycles. The first step in calculating the remaining fatigue life is to
ascertain the traffic over the bridge. The average daily traffic (ADT) and the traffic
growth rate on the bridge can be obtained from the reports published by ministries of
transportation. The ADT can be determined by using equation (2-11).
ADT (Y )  G  g r .Y (2-11)
where G is the current predicted ADT, gr is the growth rate and Y is the number of
years starting from the year of construction. The percent truck traffic of the traffic should
be obtained. Average daily truck traffic (ADTT) for the main lane is determined using
equation (2-12):
ADTT (Y )  ADT (Y )  FT  FL (2-12)
where FT is the percentage of truck traffic and FL is the percentage of trucks of
main lane in case of roadway road. To determine the number of load cycles to failure, the
following relationship is used:
RL
N f  365  CL  ADTT (Y ).dY

A
(2-13)
where Nf is the number of load cycles to failure (remaining fatigue life in cycles),
CL is the load cycles per truck, RL is the remaining life of the detail, and A is the current
age of the structure. The remaining life can be found by integrating and solving for RL.
41
2.10 Review on available finite element fatigue software
2.10.1 Introduction
In the past two decades, the computer advancement revolutionized engineering
analysis methodologies, and gave better confidence in accomplishing lengthy numerical
analyses that were previously difficult to carry out and needed several simplifications
with many assumptions. The prior availability of the Finite Element Method (FEM) in
mechanical and structural engineering assisted such numerical tools; it helped in
performing complex stress analysis on complex structural and mechanical components.
One of the fields that were greatly affected by the advancement in FEM and numerical
tools is the field of fracture mechanics. This contributed significantly to the advancement
of aerospace technology. Also, it enabled researchers in the field of fracture mechanics to
verify their analytical closed-form solutions for different fracture mechanics problems.
Advancement in computer technology and parallel processing helped in simulating
difficult phenomena such as crack propagation problems, flow of plasticity and even
micro-structural inter-granular interactions and sliding.
The main challenge for design of a structural steel element for fatigue is to
incorporate all parameters that would affect the fatigue life computation. Such
requirement is not currently implemented in the available commercial fatigue load
postprocessors. Several approaches were made in the literature to develop finite element
programs or adapt existing software for fatigue evaluation purposes. More details on
these approaches can be found in Hanq et al. (2000), Engelstad et al. (2001) and Cojocaru
et al. (2008).
42
2.10.2 Object Oriented Programming
Since this research involves the development of a numerical tool, a quick review
for the Object-oriented programming (OOP) methodology used is explained. OOP
concepts began developing since the 1960s but it developed as the dominant
programming methodology in the early and mid-1990s when programming languages
supporting the OOP techniques became widely available. Before that, the procedural
programming was the dominant methodology where procedures can be represented as a
network of routines which call one another, i.e., "call tree". Whereas in the OOP
approach, there is a collection of discrete classes (objects) that incorporate data structures
and combined with it, the procedures which apply to that data structure. An object
oriented program is composed of objects, each with a number of attributes that define the
state of the object and methods (functions) that define the behaviour of an object for
changing or returning the state of the object.
Since the scope of this study focuses on developing a numerical tool for fatigue
life prediction utilizing the FEM, several commercially fatigue postprocessors that are
available at the time of writing this thesis are reviewed in the following sections. Detailed
review of different fatigue software can be found in Papuga (2005).
2.10.3 Fe-Safe™ (2011)
Fe-SafeTM is developed by Safe Technology in Sheffield, England (Fe-Safe, 2011).
Fe-Safe is distributed via the ABAQUS distribution network; this is besides it being a
stand-alone product as well. Fe-SafeTM incorporates another product of Safe Technology,
which is the Safe4FatigueTM aimed at signal processing and damage computation without
43
FE-data. They added new modules, namely, “Rotate” for solution of axis-symmetric
components and “TMF” for thermo-mechanical fatigue solution including creep effects.
The multi-axial fatigue solution incorporates both the stress based (S-N) and the
strain based (e-N) approaches. A representative of combined criteria Brown-Miller
Brown et al. (1973) with normal and shear strains is implemented too. The program is not
capable of representing corrosion simulation.
2.10.3.1 MSC.Fatigue™ (2011)
The program is highly modular. It is fully integrated into the Nastran™
environment, which is an interesting feature. The multi-axial fatigue module is separated
as well as modules Vibration, Welds, Fracture and Utilities.
The Weld module has its limitations in specific requirements on modelling of weld
area, which considerably complicate work. Thus, e.g. the fillet weld of two perpendicular
components has to be modelled with a row of elements connecting both sheets under 45°.
The Fracture module is the only module of commercial products described, which
computes the crack growth phase as well. This is in accordance with the Nastran‟s focus
towards aviation industry. The program offers nearly all solutions except for the thermo-
mechanical fatigue with creep. The only thermal effect incorporated is the change of S-N
curves under specific high temperature. The program does not have corrosion simulation
features.
2.10.3.2 FemFat™ (2011)
The program is highly modular. FemFat™ (FemFat, 2011) is entirely based on the
S-N solution. Even the multi-axial method is based on S-N curves, which are evaluated
44
over specific planes. FemFat™ does not offer complete clarification of its approach and
leaves it as its know-how (black box).
Another advantage of FemFat™ is the existence of the Heat module with
embedded thermal fatigue and creep model. The Visualiser module enables graphic
inspection of results on FE-mesh, but it has only very basic functions.
2.10.3.3 Ansys™ nCode Design Life (2011)
ANSYS™ created a Fatigue Powerful Module in collaboration with HBM nCode
(2011) engine incorporating both stress-life and strain-life analyses with a variety of
mean stress correction methods, including Morrow, Smith-Watson-Topper (SWT) and
many other criteria, the ANSYS™ Fatigue Module provides contour plots of fatigue life,
damage, factor of safety and bi-axial stresses. Additional results include rainflow matrix,
damage matrix, fatigue sensitivity and hysteresis. The drawback of this powerful program
is that it does not alter the mesh to simulate corrosion penetration or run any crack
propagation analysis. It is worthy to mention that this program has significantly improved
from 2010 and added many features compared to its state at the beginning of this thesis.
45
2.10.3.4 Comparison and Proposed tool
From the previous review of the available known software packages, it can be
found that most of these programs:
 Do not allow adding custom damage laws (for research purpose), they are
like a black box with fixed features.
 Do not allow altering the mesh progressively and simulating corrosion
effects on the mesh (which is a contribution of this research).
 User can not add custom interaction laws that correlates input parameters
together such as (corrosion with temperature, corrosion with fatigue ...etc).
One of the objectives of this research is to develop a numerical tool that helps in
overcoming the previously mentioned drawbacks of available fatigue software. The
numerical tool will be called CorrFLP (CORRosion Fatigue Life Predictor). Table 2.3
shows a quick comparison between existing software and the proposed numerical tool.
Table 2.3 A comparison of features offered by commercial fatigue postprocessors and the
proposed tool CorrFLP. (Maximum is marked with five stars, whereas No stars
corresponds to no implementation).
Ansys. Proposed
Criterion ZenCrack Fe-Safe MSC.Fatigue FemFat
Fatigue CorrFLP
Corrosion Simulation (Geometric) *****
Corrosion-Fatigue (Strain-life model) ***
Thermal fatigue * ***** ** * ***** *
Creep **** ****
Multiaxial solution ***** ***** **** ***** **** ***
Post Processing FE results ** ***** ***** ***** ***** ****
Transient solution (sequence of FE-calc.) **** **** **** ***** ***** ****
Load history operation **** ***** **** ** ** ***
Crack growth ***** ***** **
Internal visualiser ***** ***** **** ****
Adding Custom Damage Rules *** * ** * **
Adding Interaction Rules ***
Seam welds * ** *** ****
Spot welds * ***** *****
46
From Table 2.3 it can be seen that CorrFLP is the only tool that incorporates the
simulation of corrosion. CorrFLP also incorporates the concept of programmable
interaction rules, which enables any researcher as a user to test any theory and apply any
damage model that interacts with all input factors. On the other hand, it should be
mentioned that CorrFLP is not powerful in thermal fatigue and fatigue in welds, which
could be a point for future research. Also, Zencrack and MSC.Fatigue are powerful in
crack propagation analysis, while CorrFLP has limited crack propagation features (out of
the scope of this research). Next chapters will illustrate the methodology used in this
research.
47
3 CHAPTER 3
DEVELOPMENT OF A FATIGUE ACCUMULATIVE-DAMAGE

MODEL
3.1 Introduction
Most of the available data regarding fatigue life is based on experimental work
conducted on specimens subjected to constant amplitude cyclic loading (stress or strain).
As a matter of fact, during the service life of any structure, it is expected that the structure
would be subjected to various random loads that could have a constant- or random-
loading pattern. The current practice for quantifying the accumulated damage at each
stress (or strain) level, and consequently the fatigue life of the structural element, is to use
these available constant amplitude experimental data. Unfortunately, experiments using
variable amplitude loading are expensive, time-consuming and having much uncertainties
about the in-service loading spectrum (Paepegem et al., 2002). This is the reason that
studying multi-stress level fatigue has always been a challenge for researchers and
engineers and much more difficult to formulate than constant amplitude stress fatigue.
Many damage models were proposed in the literature to explain the multi-stress
level fatigue phenomena and predict its life by cumulative damage model approaches.
Cumulative damage model approach could be classified as damage models defined by
number of fatigue cycles, material characteristic variables, the applied stress levels,
temperature, frequency of loading, moisture content and the geometric shape of the
specimens (Du, 1998). In general, cumulative damage function can have either trend A,
B or C as shown in Figure 3.1.
48
1
Damage
B
0
Cycle ratio (n/N) 1
Figure 3.1 Possible trends in fatigue damage models (Hwang et al., 1986)
3.2 Review on Cumulative Damage Models
In this section, a more comprehensive and technical review on several available
damage models is conducted. The purpose of this review is to highlight the points of
strength and weakness in these damage models. Moreover, a quick inspection of the
common phenomena traced by these models will give a clearer picture of fatigue damage
and paves the way to propose a new fatigue damage model. In the following subsections,
a detailed review on some of the most important fatigue damage models is presented.
3.2.1 Miner’s Linear damage Rule (LDR)
In 1945, Miner (Miner, 1945) expressed the concept of damage accumulation in a
very simple linear trend (trend B in Figure 3.1). Miner expressed this damage law (also
known as Palmgren-Miner‟s Law) in the following mathematical form:
49
ni
D (3-1)
N fi
Where D denotes the damage, ni and Nfi are the applied cycles and the total
number of cycles to failure under the ith constant-amplitude loading level, respectively.
This rule is still the most popular rule used due to its simplicity. Unfortunately, this
damage model has many deficiencies. This rule fails to predict the effect of load history
sequence. Experimental data indicate that the sequence in which various stress levels are
applied has significant influence on the fatigue behaviour of materials. It is also widely
known that Miner‟s damage sum to failure is greater than unity for low-to-high tests and
less than unity for high-to-low tests (Hwang et al., 1986). Another major limitation of
Miner‟s rule is that it is stress level independent, in that its damage trend follows only
one function (which is linear) for any stress level.
3.2.2 Marco-Starkey theory
In 1954, and as a remedy to the deficiencies of Miner‟s LDR, Marco and Starkey
(1954) proposed the first non-linear load-dependent damage theory. This damage model
is based on (D-r) curves, where „r‟ represents different stress levels. This model has the
form:
C
 n 
D i  (3-2)
N 
 fi 
50
where „C‟ depends on the stress level and has a value greater than unity, which
means it always follows the trend „C‟ in Figure 3.1. Although this rule is stress
dependent, but its main disadvantage is the possibility of having the damage sum greater
or less than unity depending on the load sequence as shown in Figure 3.2.
Figure 3.2 Load sequence effect in Marco-Starkey theory (1954)
3.2.3 Damage theories based on endurance limit reduction
One of the deficiencies of the previous damage models that they always correlate
to the Number of cycles to failure Nf of a virgin material under constant amplitude
loading. Many approaches were done to represent the accumulated damage as a reduction
in the fatigue endurance limit of the material. Examples of these are theories proposed by
Henry (1955), Gatts (1961) and Bluhm (1962). All these damage models are nonlinear
and are able to account for the load sequence effects. However, the mathematical form of
these models is complicated. Moreover, they do not take into account the load interaction
effects.
51
3.2.3.1 Corten and Dolan theory
Corten et al. (1956) performed hundreds of experimental tests on cold-drawn steel
wires by applying different loading blocks with different stress levels. Their experimental
test program had several series of two-stress level loading sequences repeated until
failure. They described damage in a phenomenological hypothesis that represents the
nucleation of microscopic voids which develop into cracks. The damage is given in terms
of number of damaged nuclei and the rate of damage propagation as a power function of
the number of cycles as shown in Figure 3.3.
1 > 2 > 3 > 4 > 5

Damage %
1 2 3 4 5

Number of Cycles
Nf1 Nf2 Nf3 Nf4 Nf5
Figure 3.3 Corten‟s hypothesis of the progress of fatigue damage (Corten et al., 1956)
Corten et al. (1956) formulated the expected life in a two stress level loading block
history in a simplified equation as follows:
N
N 1
(3-3)
R 1/ a
1   
52
where N is the total accumulated life, N 1 is the number of cycles to failure under
constant amplitude loading under the high stress level,  is the ratio of the number of
high stress level cycles to the low stress level cycles in one loading block, R1 / a is a
factor that can be related to the stress ratio.
6.57
S 
R 1/ a
  1  (3-4)
 S2 
where S 1 is the high stress level and S 2 is the low stress level. Results using this
equation had a good agreement with their tests. Equation (3-3) is limited to a two-level
stress block sequence loading history. The constant 6.57 in equation (3-4) was
determined by fitting experimental results.
3.2.3.2 Manson’s theories
With over 20 years of testing, Manson and his co-workers proposed several
conceptually-different damage models, and continuously updated them to correlate with
their considerable test result data. Manson et al. (1967) first introduced the concept of
Double Linear Damage Rule (DLDR). After which, their work was extended to develop
the concept of Damage Curve Approach (DCA) and later the Double Damage Curve
Approach (DDCA) (Manson et al., 1981). These models were in good agreement with
experimental work in low cycle fatigue under two loading levels. These damage models
are all load-level dependent, but do not account for the load interaction effect and small
amplitude cycle damage. Also determining the Knee point (which is a new concept
introduced in Manson‟s work) is a tedious procedure that could be inconvenient.
53
3.2.3.3 S-N Curve modification theories
Many researchers agreed that the errors in the LDR life predictions are not
necessarily due to the linear summation of damage but to the assumption of damage-rate
independence of loading levels (Fatemi et al., 1998). Based on experimental results,
Subramanyan (1976) introduced a modified damage model of several damage iso-lines
having a knee point or pivot near the fatigue limit of the material. This approach has
problems at stresses near the fatigue limit due to the singularity at the knee point. Later,
Hashin et al. (1978) presented a discussion of the S-N curve rotation approaches and
performed analytical calculations along with experimental work. The predicted results
were in good agreement with test data. Later, Leipholz (1986) revived the concept of a
modified S-N curve as shown in Figure 3.4, named S-N' curve, which accounts for load
interaction effects. Leipholz‟s model is represented as:
N 1   i N i'  (3-5)
where N is the total accumulated life, and  i is the frequency of cycles and N i' is
the modified life with loading level  i .
S
Virgin S-N curve
Modified S-N curve
S1
N1,m N1,o N
Figure 3.4 Modified S-N curve approach
54
3.3 Factors Affecting Damage Models
Based on available models, it could be said that, in general, damage state indicator
of the material can be expressed as follows:
D  f ( N , s, r, f , t , c, Do ,.....) (3-6)
where,
D = material damage state (0 for undamaged, 1 for completely damaged)
N = number of fatigue cycles
s = previously applied stress level
r = applied stress level
f = frequency
t = temperature
c = corrosion rate
Do = Previous material damage state
In the following sections, a brief discussion on the influence of different
parameters and their respective effects that need to be considered when formulating a
comprehensive, yet simple, fatigue damage model.
3.3.1 Mean Stress Effects
Fatigue tests have shown that applying a tensile mean stress results in a shorter life
than those test specimens that are subjected to a zero mean stress. Many equations were
developed to take this into account, such as Goodman (1899), Morrow et al. (1981) and
Smith-Watson-Topper (Kandil et al., 1982). The latter two equations showed the best
agreement with many test data on different materials (Drapper, 2008).
55
3.3.2 Effect of Frequency
Developing a unified damage law that takes into account all factors might not be
computationally effective; hence, some effects could be omitted from a damage model
due to their insignificant influence. For example, many researchers reported that the
frequency of loading does not significantly affect the fatigue behaviour if it is within the
range of 1-200 Hz (Frost et al., 1974).
3.3.3 Effect of temperature
At elevated temperatures, mean stress effects become extremely complex to
simulate because of the interactions among creep, fatigue and environment. Moreover,
the linear elastic stress intensity factor has more limitations at elevated temperatures
because of the appreciable plasticity (Cui, 2002). Fortunately, regular fatigue-sensitive
structures are not usually designed for elevated temperatures. Thus, these effects could be
neglected. It is important to mention that temperature fluctuation in snowy regions
(which can reach -80 oC in some places) can have a minor effect on the ductility of metal
and the mean stress. This could be easily taken into account when studying fatigue of a
structural component.
3.3.4 Effect of corrosion
Fatigue in a corrosive environment is a very complex phenomenon known as
corrosion-fatigue. It is important to differentiate between corrosion-fatigue and stress-
corrosion-cracking which is the environmentally assisted fracture under static loading.
Due to the experimental difficulty of simulating corrosion and measuring its effect on the
fatigue behaviour, this topic needs more future investigation. Although the damage rule
56
that will be proposed later in this chapter does not take into account the effect of
corrosion, it has the potential of being extended to account for the effect of corrosion
provided that there is a huge amount of long-term experimental test results under several
corrosion rates available.
3.3.5 Other effects from previous damage models
In order to have a comprehensive fatigue damage model, it has to take into
account:
 Load interaction & load sequence effects.
 Stress-level dependency.
 Simplicity in its formulation and use by designers.
3.4 Proposed Damage Model
In this section, a new damage model is proposed. This damage model will aim to
alleviate some of the shortcomings in the previously mentioned available damage models.
The damage model will incorporate the effects of stress-level dependency and load
sequence effects while still maintaining a simple form (which is the main reason
designers still use the Palmgren-Miner rule).
3.4.1 Discussion
In this research, a phenomenological approach for estimating the level of fatigue
damage will be used. Material damage can be described as the loss in expected life by
accumulation of loading cycles under certain stress levels. A new concept is proposed
which will be called the “Virtual Target Life Curves” (VTLC). In this approach, it will be
57
assumed that any material has a virtual (theoretically infinite) expected life that is, by
definition, greater than the real failure life under constant amplitude loading. As constant
amplitude fatigue loading proceeds, this expected life reduces in magnitude depending on
the number of cycles, where the rate of deterioration keeps changing depending on the
stress level. In cases of variable amplitude fatigue loading, some sudden jumps of
damage (and consequently corresponding loss in expected life) will occur due to
overloading effects resulting from the increase from low stress levels to high stress levels
(and vice versa). The magnitudes of these jumps depend on the ratio of the two stress
levels. Figure 3.5 shows that the concept of virtual target life curves –in case of constant
amplitude loading– will lead to the regular S-N curve (which represents the number of
constant amplitude cycles that leads to failure of a laboratory tested element).
Focal Point
Log ()
Original S-N curve

Virtual Target Life Curve
Nexpected
i
b bv
Reduction in Expected Life

due to accumulated damage
n=1 Log ( N )
Figure 3.5 The proposed Virtual Target Life Concept
58
It is important to note the difference between the proposed approach and other
approaches in the literature that are based on modifying the S-N curve. Previous
approaches such as Manson‟s early work and Subramanyan (1976) were based on
shifting the so-called virgin material curve as a starting point to account for cumulative
damage in cases of variable amplitude loading. The author doesn‟t agree with the term
virgin material curve to be used for the S-N curve as this curve represents in reality the
failure of the specimen, whereas the term virgin material implies a totally unloaded
material. In this research, a different approach is adopted; the material is supposed to
endure until it reaches a virtual target life. The virtual life decreases as damage
accumulates. The rate of decrease of this virtual life changes based on the loading history
(including both: stress levels and loading sequences) and depends on the material. Figure
3.6 illustrates the difference between Manson‟s approach, Subramanyan‟s approach and
the VTLC approach.
Figure 3.6 Comparison between the proposed VTLC and previous S-N curve
modification approaches
59
The VTLC approach aims to give regular S-N curves (for constant amplitude
excitations) a time- or loading- history dependency by making them dynamically moving
according to the damage state. Moreover, it would be useful to have the S-N Curve as a
way of monitoring the history of deterioration of a certain component in a structure as
illustrated in Figure 3.7.
Focal Point
Log ()
Virtual Target Life Curve of a

component at construction date
20 199 196 192 189

10 0 0 0 0
Year Log ( N )
2010
1890
Accumulated
Number of Cycles
Figure 3.7 VTLC as a measure of deterioration of structures
3.4.2 Derivation
The concept of fatigue damage has always been challenging as it is not easy to set
a clear definition for damage or provide a physical correlation to it. The VTLC approach
doesn‟t use physical damage as a failure criterion. The VTLC approach gives a different
perspective for the concept of damage. It views damage rather than being an
accumulative sum of cycles up to failure, to being a concept of accumulated loss in the
expected life. Unlike the regularly used criterion for failure when damage reaches unity,
60
the VTLC failure criterion is reached when the accumulated number of cycles reaches the
current target life at a certain stress level.
Using S-N curves, at certain stress amplitude  a , the number of cycles to failure
could be determined from the following equation:
1 1
  b
  'f  b
N f   a'    (3-7)
   
 f   a 
where b is the fatigue strength exponent and  'f is the fatigue strength
coefficient of the material. Equation (3-7) can be re-written as :
 
log  a  b logN f   log  'f   (3-8)
To construct the virtual target life (VTL) curve from an existing S-N curve, we
will assume another curve of slope bv, which has to be less than b. All (VTL) curves are
assumed to meet at the same intercept at N = 1 cycle which means the equation will have
the same constant term: log  'f . Thus, the equation for any Virtual Target Life Curve
would be:
 
log  a  bv logNVTL   log  'f   (3-9)
Therefore, the virtual target life (VTL) of a component at a certain stress level i
can be expressed by the following equation:
1 1
  bvi
  'f  bvi
NVTLi   i'    (3-10)

   
 f   i 
The rate by which the VTLC approaches the original S-N curve is important.
Corten et al. (1956) proposed a stress dependency trend that schematically depends on the
61
stress level following a power law as shown in Figure 3.3. In order to simplify the
damage calculation process, the trend in this research was selected to be linear. However,
a new factor for excess overloading damage is introduced. Generally, the total change in
slope bi of VTL curve from the original initial slope bv at stress level i, can be
expressed as:
bi  ri N ia (3-11)
where a is the power chosen for the trend, N i is the total equivalent number of
accumulated cycles under stress level i and r can be determined from:
a
ri  bvo N fi (3-12)
where bvo is the total expected change in slope of the VTL curve in case of
constant amplitude loading and N fi is the number of cycles needed for failure under
constant amplitude loading at stress level i.
 N 
N i  N i-1  fi  (3-13)
 N f i-1 
where N i , N i 1 are the total equivalent accumulated number of cycles under the
current and previous stress levels i and i-1, respectively. N fi , N f i 1 are the number of
cycles to failure under the current and previous stress levels i and i-1, respectively.
62
Thus it can be easily proved that by adding ni cycles:
a
b  N fi 

bi  ri N i  ni 
a
 voa  N i 1  ni  (3-14)
N fi  N f i 1 

To make the trend linear, taking a = 1 in equation (3-14)
bvo  N fi 
bi  N i 1  ni  (3-15)
N fi  N f i 1 

This equation can be formulated in terms of stress ratio
  S 
1
b 
bvo   f i 1
 
bi   N i 1  ni  (3-16)
N fi   S  
  fi  
By using a = 1, the trend will become linear as shown in Figure 3.8.
bv
 1 > 2 >  3 > 4 > 5
bvo 1 2 3 4 5
Number of Cycles
Nf1 Nf2 Nf3 Nf4 Nf5

Figure 3.8 Change in slope in multi-stress level loading using a linear trend
In the VTLC method, the change in slope can be interpreted as being the loss in
expected life. Applying a constant amplitude loading at a certain stress level should
decrease the life by a well-known value N expected corresponding to a change in slope of
63
bvo such that the moving target life curve coincides with the original S-N curve as
shown in Figure 3.9.
Focal Point
Log ()
Original S-N curve
Virtual Target Life Curve
Nexpected
i
b bv
Reduction in Expected Life

due to accumulated damage
n=1 Log ( N )
Figure 3.9 VTLC as a measure of deterioration of structures
Since both the original and VTL curves are linear on a log-log curve, then the
expected loss in life curve should also be linear as shown in Figure 3.10.
Log ()
For Constant
Amplitude Conditions
'
log( Nexpected )
Figure 3.10 Schematic curve for the expected loss in life at each stress level
64
The total change in slope bvo will be taken equal to the original S-N curve slope.
This assumption is verified by the experimental results and validation examples that will
be discussed later in the following sections.
3.4.3 Overloading effect
The effect of load sequence and interaction can be dealt with using two approaches
(Li et al., 2001). The first approach is to account for the effect by evaluating an additional
term of fatigue damage due to the effect of load interaction. The second approach is to
account for the effect by obtaining an effective reduced fatigue life. The method adopted
in this paper combines both approaches indirectly. It is important to take into account the
overloading damage that occurs due to shifting from low stress level to high stress level
(L-H). This damaging effect depends on the ratio between the two stress levels and the
current damage level. A new term bDj is introduced, which is the jth additional jump in
slope due to changing the stress level. This overloading term must be introduced once the
stress level changes from low stress level S2 to high stress level S1. bDj depends on the
ratio of stress amplitudes between consecutive blocks rb = S2/S1 as in equation (3-17).
bDj 
bvo
100

1  rb
p
 (3-17)
where p could be a material constant that can be determined from experimental
results and is a function in the ratio rb.
65
3.4.4 Proposed procedure
Assuming that the element is subjected to a loading history with multiple stress
levels, the fatigue life prediction procedure using the VTLC approach would be as
follows:
1. For each cycle block, get the total change in slope bi due to the additional
number of cycles ni from equation (3-16).
2. If the loading block i has a higher stress level, add the overstress term bDj to
bi from equation (3-17).
3. Calculate the new virtual target life at stress level i at loading block j from the
following equation:
1
  bv  b j
NVTLij   i'  (3-18)
 
 f 
4. Store the real total accumulated number of cycles.
5. Check that the equivalent total accumulated number of cycles Ni < NVTLij.
Repeat from steps 1 to 5 until the condition in step 5 is violated. Then the total life
would be the last value calculated from step 4.
If the total accumulated number of cycles is still less than the last calculated NVTLij ,
then the tested specimen still has some fatigue life remaining. Next section will illustrate
how to calculate the remaining fatigue life using the VTLC method.
66
3.5 Remaining life calculations
Using the VTLC method, the life curve is being reduced as the number of cycles
keeps accumulating and increasing. This makes the prediction of the fatigue life an
iterative problem. Using any programming script or a spreadsheet program could easily
calculate the remaining life. Despite the iterative nature of the problem, it could still be
solved manually by repeating steps 3 to 5 in the previous sub-section using small loading
blocks until reaching failure. Counting these extra blocks and adding them up simply
gives the expected remaining life. A sample object oriented C# computer code is
provided in Appendix A
3.6 Validation
3.6.1 Introduction
In order to validate the proposed methodology, results from an intensive
experimental investigation performed by Corten et al. (1956) were used. Over 700 steel
wire specimens were tested in that investigation under several stress levels and different
loading patterns. Corten et al. (1956) studied the specimens on three phases and noted
them as Series A, B, and C. Figure 3.11 shows the loading pattern used in the tests.
n(1-) n
Stress
S2 S1
Cycles
n = 10,000 cycle
Figure 3.11 Testing Loading Pattern (Corten et al., 1956)
67
Results of the proposed damage model were compared to the commonly used
Miner‟s Linear Damage rule (LDR) and Corten and Dolan‟s rule mentioned in section
3.2.4. The chosen value of p for equation (3-17) for the steel material used in Corten et al.
tests was found by curve fitting to be as follows:
2
p  100rb  120rb  36 (3-19)
Figure 3.12 and Figure 3.13 show a comparison between the results obtained using
Miner‟s LDR, Corten et al.‟s rule and the proposed VTLC method. It is convenient to
combine the uncertainty in measurements by forming the ratio N/Nf1, which in terms of
logarithm is equal to log(N)-log(Nf1). Each set of results represent 17 tested specimens.
The round circles for each set in the figures represent the mean value, while the error
range bars represent the 95 per cent confidence limits of the experimental results. By
applying the VTLC approach procedure described in section 3.4.3, very good predictions
of the experimental results were obtained.
68
100
Analytical Predictions
Group 2 Group 1
VTLC 96,000-76,000
Corten 96,000-76,000
High to Low cycles ratio per block  Miner 96,000-76,000
VTLC 96,000-66,000
Corten 96,000-66,000
Miner 96,000-66,000
10
Group 1 Group 2
0.0
0.0 0.2 0.4 0.6 0.8 1.0
log (N)-log(Nf1)
Figure 3.12 Analytical predictions of fluctuating stress amplitude experiments with high
stress of 96,000 psi using three methods: VTLC, Corten, and Miner.
(├─O─┤Experimental results, Corten et al., 1956)
Analytical Predictions
Group 3 Group 2 Group 1
100 VTLC 86,000-76,000

Corten 86,000-76,000
Miner 86,000-76,000
VTLC 86,000-66,000
Corten 86,000-66,000
Miner 86,000-66,000
High to Low cycles ratio per block 
VTLC 86,000-56,000
Corten 86,000-56,000
Miner 86,000-56,000
10
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
log (N)-log(Nf1)
Figure 3.13 Analytical predictions of fluctuating stress amplitude experiments with high
stress of 86,000 psi using three methods: VTLC, Corten, and Miner.
(├─O─┤Experimental results, Corten et al., 1956)
69
3.6.2 Observations and Conclusions
From the above results it could be seen that the VTLC approach captures different
ratios of low to high (L-H) loading patterns with a relatively small error compared to the
non-conservative results obtained using Miner‟s LDR rule. Miner‟s LDR rule has good
predictions for loading patterns with block stress ratio rb = S2/S1 that are very close to
constant amplitude loading (close to unity). Although the VTLC method takes a linear
(thus a more simple) trend in describing damage, it has very good agreement with life
predictions that uses Corten‟s rule. This could be attributed to the fact that VTLC
accounts for damage jumps that are caused by overloading through the term bDi . As
such, it could be said that the VTLC approach is a linear damage rule for measuring the
loss in life that incorporates nonlinear effects arising from overloading in case of a
variable fatigue loading pattern. It can be observed that for high values of the ratio  , the
predicted life using either the Miner rule, Corten‟s rule or the VTLC method is usually
more conservative. This might be attributed to the crack closure phenomenon reported by
Elber (1970), which can be observed when the amount of high stress cycles becomes
significantly more than the low stress cycles. This complex phenomenon could be further
investigated and incorporated through the term bDi in future work. Also, it should be
noted that it is interesting to explore the model for case of both variable stress and
variable frequency patterns, but there was no available experimental data in the literature
to correlate the model with.
70
3.7 Extending the methodology
One of the interesting features of this method is that it could be extended to include
the effect of corrosion-fatigue. As corrosion of a component increases, the damage
increases, thus the slope of the VTL curve will automatically decrease. Finding a relation
between the rate of corrosion and the change in slope is the challenging part as there is
very limited experimental data available in literature to calibrate such model. In order to
achieve this target, many experiments on several environmental-material systems should
be done, to study the effect of such system on the rate of change in slope of the VTL
curves. Moreover, an experimental program should be conducted to study the effect of
overloads in several corrosive environments. Such a program should also evolve different
frequencies of loading to account for different loading frequencies with time.
3.8 Case Study
As an example for the practical application of the developed damage model, the
remaining fatigue life of a main member of a riveted single-track railway bridge built in
Croatia in 1895 was investigated. This bridge was chosen because it was the only bridge
(available to the authors during the course of this thesis) that had detailed data for its
loading over about 100 years. Detailed bridge data and loading was acquired from the
report provided by Kühn et al. (2008). No climate data or corrosion data have been
reported.
71
3.8.1 Geometry
The bridge is composed of two equal trusses. The structural system of each truss is
shown in Figure 3.14 . In year 2000, after over 100 years of service, there was a
motivation for assessment since the bridge has reached the end of its design working life.
Therefore, an assessment was carried out to determine the residual service life of the
bridge. The actual traffic at that time has reached the maximum capability of the railway
line. Construction of a new railroad is expected in the future (double track).
2 11 4 12 6 13 8 14 10 15 12 16 14 17 16 18 18 19 20 20 22
3.75
32 33 34 35 36 37 38 39 40 41
21 22 23 24 25 26 27 28 29 30 31
1 21
1 3 2 5 3 7 4 9 5 11 6 13 7 15 8 17 9 19 10
3.5 3.5
10x3.5 =35m
Figure 3.14 Structural System of the truss bridge showing the investigated member D-36
Concerning the number of locomotives and wagons involved, a detailed
explanation is provided in the report prepared by Kühn et al. (2008). Since the bridge has
been in use for 105 years, there is a higher probability of failure due to fatigue than due to
static overloading. Data about traffic load and about materials were obtained from the
railways archives. The structural system of the main truss is shown in Figure 3.14 . It
should be noted that each specific structural element has a different fatigue life. The
diagonal element in the middle of the span (D-36) was estimated as the most critical in
fatigue. The highest number of cycles and the highest tensile stress range are expected to
occur in this element (ordinates of the shear influence line are both positive and negative,
and at their highest value). It was confirmed by reports of measurements that the highest
72
stress range was at this element (Kühn et al., 2008). Member (D-36) is 5.13m long and
has a cross-section of 4L 80x10 with a gross area of 60.4 cm² and a net area of 52.0 cm².
It is connected at its ends by a riveted connection using rivets 16mm in diameter.
Detailed data on traffic such as number of trains each year, the type of trains and tons
transported yearly was collected from the railways archives and available to do a Miner‟s
sum. The analysed element was built using rivets, which corresponds to the detail
category D in AASHTO (2007).
3.8.2 Formation of the stress range spectrum
On the basis of the data reported by Kühn et al. (2008), velocity measurements
were taken along with corresponding stress ranges to calculate the dynamic amplification
factor which was calculated to be about 1.17. The total secondary stresses that it could
amount to were practically chosen to be 16% of the primary stresses. In that report, the
calculations were based on the net cross-sectional area of the member D-36. The stress
range spectrum for the time period from 1895 until 1980 was provided and corresponding
fatigue damage was calculated using Miner‟s rule as shown in Table 3.1(a). Another
stress range spectrum was done for each year from 1980 to 2000 and the corresponding
number of crossings as shown in Table 3.1(b). Kühn et al. (2008) considered the studied
detail having a fatigue category D according to AASHTO. Calculations showed that there
is no remaining design fatigue life at 1985. The member chosen for the analysis cannot be
considered safe anymore after 90 years of service, although the bridge was reported to
still be in service for more than 105 years.
73
Table 3.1 Tables reported by Kühn, et al. (2008) for estimated damage from (a) 1895-
1980, and (b) 1980-2000
m : represents the slope of the S-N curve

ni : represents the actual number of cycles
Ni : represents the number of cycles to failure
The data from Kühn et al.‟s report has been thoroughly investigated and rearranged
chronologically year by year as shown in Table 3.2.
74
Table 3.2 Chronologically ordered data from Kühn et al. (2008) report
Train Passing  .  Dsk .   Number of cycles
Cargo wagon 23.6 90000

1895-1913 S 123 Locomotive 60.8 140000
Passenger wagon 20.4 173000
S 32 82.8 369000
1914-1946 Cargo wagon 34.7 345000
S 38 92.6 235000
1947-1959 Cargo wagon 34.7 136000
E 362 70.8 318000
2E 362 64.3 18000
1960-1980 Cargo wagon 45.9 880000
E 320 37.6 63000
E 362 70.8 20000
1980-Present
Cargo Wagon 45.9 65000
(Yearly)
75
3.8.3 Fatigue Life Prediction using VTLC
By using the same detail category of the connection (category D) as reported by
Kühn et al. (2008), the VTL curve for this member was constructed using an initial slope
equal to double the original slope 3. By performing the procedure mentioned in section 0
several VTL curves were plotted showing the loss in fatigue life of the studied member
(which has a detail category D) throughout the years of service as shown in Figure 3.15 .
In order to compare the effect of considering the detail as category C on the predicted
fatigue life, several VTL curves showing the loss in fatigue life were plotted as shown in
Figure 3.16.
Category D
1000
Stress Range (MPa)
100
10
0.0001 0.01 1 100 10000 1000000 100000000
Number of cycles (Nf) Millions
Figure 3.15 VTL curves for the investigated member D-36 showing the deterioration in
fatigue life with years (assuming the detail is category D)
76
Category C
1000
Category D
Stress Range (MPa)
100
C
D
10
0.0001 0.01 1 100 10000 1000000 100000000 1E+10
Number of cycles (Nf) Millions
Figure 3.16 VTL curves for the investigated member D-36 showing the deterioration in
fatigue life with years (assuming the detail is category C)
3.8.4 Discussion and Conclusions
The proposed VTLC method provides a simple and a visually illustrative method
for predicting the fatigue life of a certain structural detail under variable amplitude
loading. It has been shown that this method could be calibrated to take into account the
effect of stress overloading. On the other hand, similar to all damage accumulation
models, they need calibration with a wide range of real experimental data to become
more accurate and reliable. The VTLC predictions showed that there is no remaining
design fatigue life after year 1982, which is about 87 years of service. This estimation is
more conservative than the 90 years predicted by Kühn et al. (2008) using Miner‟s
cumulative damage rule.
77
It is important to mention that the bridge was reported to still be in service for
more than 105 years, while the VTLC estimations where about 87 years. This difference
could be attributed to the choice of detail category D for all riveted connections, which is
−as mentioned before in the literature review− a lower bound for all riveted connections
that could go up to detail category C or slightly more in some cases. If the same detail
was to be considered as category C, the predicted life using the VTLC would have
extended to the year 2000, which corresponds to 104 years of service life. This shows
that there is a need to categorize riveted connections and further classify them into more
categories. This point will be highlighted in a case study in Chapter 4.
78
4 CHAPTER 4
NUMERICAL FORMULATION OF THE FATIGUE LIFE

PREDICTION TOOL
4.1 Introduction
Finite element method (FEM) is a vital tool for the analysis of various structural
elements. The continuous rapid advancement in computing technology led to the
development of several finite element software packages that facilitate solving large
structural models in a relatively shorter time. This latter aspect (speed of executing the
analysis) encouraged many researchers to simulate physical tests using FEM. Within the
context of this thesis, the FEM has facilitated performing stress analyses along with
implementing the concepts of linear fracture mechanics in order to predict the fatigue life
of structural steel components. This approach provides a better alternative to
experimental fatigue tests which are time consuming and costly. Moreover, a thorough
literature review showed that only few fatigue tests were based on variable amplitude
loading which represents the real in-service loading scenario. On the other hand, fatigue
testing results using constant amplitude are affected by the frequency of loading, which,
in most cases, is higher than reality (in order to complete the test in a relatively shorter
duration).
An add-on program named CorrFLP (Corrosion Fatigue Life Predictor) is
developed using the C# language. CorrFLP (about 40,000 lines of object oriented code)
acts as an add-on that could be associated with several FE software packages, where it
enables the fatigue analysis capabilities by providing many fatigue damage models.
CorrFLP is capable of analyzing the input mesh and adding fatigue and corrosion
79
parameters to the regular FEM analysis. CorrFLP can simulate the crack propagation
process along with the geometric simulation of corrosion by simulating the loss in
thickness. CorrFLP can present the output in the form of contours of damage, stress
concentrations or expected remaining life. The fatigue assessment process implemented
in CorrFLP involves three main phases; pre-processing phase, analysis phase and post-
processing phase. Figure 4.1 illustrates the whole process of the fatigue analysis as being
modeled in CorrFLP. Sections 4.2, 4.3, and 4.4 elaborate on the details and features of
the three main phases of CorrFLP, namely, pre-processing phase, analysis phase and
post-processing phase, respectively.
Pre-processing Module (Section 4.2)
 Modeling (loading, boundary conditions, meshing .. etc.)

 Input Undamaged material properties
 Input Cyclic Fatigue material properties
 Input Load, corrosion rates and temperature history blocks
 Apply any possibly assumed imperfections
 Perform an initial elastic simulation to determine stress intensity factors and
locations of stress concentrations.
Analysis Module (Section 4.3)
Step 1
Finite element simulation for current loading cycle :

 Analyze loading spectrum using Rainflow analysis.
 Calculate damage based on selected damage hypothesis.
 Geometrically apply corresponding thickness reduction due to corrosion .
 Reduce stiffness properties for next analysis based on current damage state.
 Perform checks for determining the end of crack initiation.
Step 2
 Evaluation of local cycle jump for each Gauss-point. Jump
 Statistically determine the global cycle jump.
Step 3
 Crack initiation calculations.

 Calculate stress intensity factors for crack propagation phase.
Post Processing Module (Section 4.4)
 Display Contours for estimated remaining life for the studied component.
 Damage growth and damages distribution across the finite element mesh
Figure 4.1 Illustration for the three main modules of CorrFLP and the interaction
between applying damage and cycle jumps
80
4.2 Pre-Processing Module
4.2.1 Model Preparation Section
ABAQUSTM 6.10 (Dessault-Systems, 2010) was chosen as the finite element
software package to be used for modeling and analyzing all the FE models in this
research. In this research, CorrFLP acts as a fatigue analysis add-on for ABAQUSTM; it
uses the finite element models created in ABAQUSTM. To simulate real-time
environmental and loading effects on the studied component, CorrFLP accepts the
following input:
1. Customized loading patterns (Load Cycle Block), to simulate the daily traffic
loading intensity on the studied component per hour.
2. Corrosion rates expressed by the rate of loss of thickness in mm/year.
3. Monthly temperature change.
4. The damage model chosen for fatigue life prediction.
5. The user can also predefine any pre-existing damage in the model to simulate
imperfections.
The user then specifies if the simulation is to proceed up until failure or if it is
required for a certain time interval. The program also gives the option to choose the
method of cycle jumps to reduce the total execution time. The concept of cycle jumps is
covered in Section 4.3.3. The input data file is stored in a Fatigue and Corrosion Data file
(.FCD).
81
4.2.2 Simulation Section
After the preparation phase is complete, CorrFLP prepares the ABAQUSTM input
file (.INP) and exports it to the ABAQUSTM solver. CorrFLP runs each analysis step and
finally extracts its results from the ABAQUSTM results database file. The extracted
analysis results are then processed by CorrFLP which starts predicting the damage caused
by this loading cycle to all the elements in the mesh using the user-chosen fatigue
damage model (from a library of built-in strain-based damage models, as explained in
Chapter 2). CorrFLP consequently changes the material properties of the damaged
elements, prepares the model for the next loading cycle, exports it to ABAQUSTM which
runs this cycle, and in turn returns it back to CorrFLP and so on until the process is
interrupted by the user, or reached its pre-defined duration limit, or the studied
component fails. The communication process between CorrFLP and ABAQUSTM is
schematically explained in Figure 4.2.
82
Preparation Phase
ABAQUS Modeling
Read ABAQUS (.inp) Input File
Module
Input Loading, Corrosion and

Temperature History Blocks
CorrFLP Engine
Create CorrFLP (.FCD) Data File
ABAQUS
ABAQUS Solver Create Updated ABAQUS Input File

Simulation Phase
(Analysis) for Analysis
Read Results
Apply Fatigue Damage
Apply Corrosion Damage
Keep Records for current state of

damage
Figure 4.2 Schematic drawing for the CorrFLP and ABAQUSTM communication process
4.2.3 CorrFLP’s User Interface
CorrFLP is designed to have a user friendly interface which is totally independent
of the ABAQUSTM interface; in that it is not necessary to have the ABAQUSTM user
interface open while using CorrFLP. CorrFLP can display the models along with all
results within its interface. CorrFLP provides a detailed log window for all processes
working in the background during the analysis.
Figure 4.3 shows a screenshot of CorrFLP‟s user interface during an analysis
showing the real-time logging of the calculations being done in the background.
83
Figure 4.3 CorrFLP‟s user interface snapshot
84
4.2.4 Material Editor
As previously mentioned, CorrFLP acts as an add-on to FEM packages, thus it
adds extra fatigue features to the regular FE input. CorrFLP accepts the input of many
fatigue cyclic properties such as the cyclic and hysteresis fatigue curve constants.
CorrFLP has a big library of fatigue properties for many materials (over 70 different steel
and Aluminum types) gathered from available published literature. . A list of these
materials is provided in Appendix B. If the user does not have any fatigue data available,
CorrFLP can derive fatigue properties from elastic and plastic material properties using
the “Uniform Material method” of Baumel et al. (1990). Figure 4.4 shows the material
editor interface. The cyclic properties provided in the material editor are used to construct
the static stress strain curve that is used in the FE analysis. The material editor can be
used to input material properties required for crack propagation analysis too.
Figure 4.4 CorrFLP‟s material editor (showing the list of built-in materials library)
85
4.2.5 Spectrum Editor
The spectrum editor enables the user to input load spectra, corrosion rate spectra
and temperature spectra. Figure 4.5 shows the spectrum editor interface.
(a)
(b)
Figure 4.5 CorrFLP‟s spectrum editor (a) Input window, and (b) History graph window
86
4.3 Analysis Section
4.3.1 CorrFLP’s Object Oriented Structure
CorrFLP was programmed using an extendible object oriented structure and
having a rich 2D and 3D finite element library of elements. When the ABAQUS TM finite
element model is imported into CorrFLP along with the results, CorrFLP updates its
originally generated mesh with that updated information. The previous mesh elements
automatically communicate together to modify their sizes and orientations in space. The
mesh can be automatically refined without affecting the boundaries such that the finite
elements optimally capture the stress gradients at locations of stress concentration. Next
sub-sections will illustrate the object oriented hierarchy of the classes (objects) within the
“Main Mesh Object”.
4.3.2 Main Mesh Object
CorrFLP has one “Main Mesh Object” that stores all data. It has a structure made
up of several managerial objects (programming classes) that monitor the input and output
of the program and make decisions based on their (each managerial object)
specialization. Figure 4.6 shows the main structure of the “Main Mesh Object” and its
most important manager objects. The following sub-sections give detailed description for
each manager object.
87
Figure 4.6 Schematic drawing for the mesh managers in CorrFLP
4.3.2.1 Fatigue Management Object
The “Fatigue Management Object” is responsible for communicating with the
material class object of each element to extract its fatigue cyclic properties and its current
damage state in order to pursue the fatigue damage calculations. The “Fatigue
Management Object” calculates the expected endurances at each integration point.
Moreover, it calculates the stress intensity factors at the stress raiser locations for crack
propagation analysis. It can also specify the expected locations for crack initiation to
monitor the crack initiation length.
88
4.3.2.2 Corrosion Management Object
The “Corrosion Management Object” is responsible for applying uniform
corrosion (reduction in shell element thickness) on the mesh based on values gathered
from the input corrosion spectrum. It is also responsible for providing corrosion data
modifications corresponding to the current temperature and the material properties. The
user specifies the expected corrosion type (i.e. galvanic corrosion, crevice corrosion,
fretting corrosion, etc.) such that the “Corrosion Management Object” defines the
appropriate corrosion rates. As the analysis proceeds, the “Corrosion Management
Object” assigns corrosion states to each external surface of the studied component
individually based on its real level of exposure.
4.3.2.3 Temperature Management Object
The “Temperature Management Object” is responsible for applying uniform
temperature on the mesh based on values gathered from the input temperature spectrum.
The “Temperature Management Object” can also assign individual surface temperatures
for each external surface of the studied component individually based on its real
temperature. It can also modify the material‟s stress strain curves based on the specified
normal temperature.
89
4.3.2.4 Interaction Management Object
The “Interaction Management Object” is responsible for synchronizing results and
data obtained from the fatigue, corrosion and temperature managers and applies user‟s
selected interaction rules to couple the effects of corrosion and temperature with fatigue.
This feature is added for applying any coupling effects between the input factors in future
based on the advancements in the body of knowledge available.
4.3.2.5 Refinement Management Object
The “Refinement Management Object” is responsible for all mesh geometrical
editing processes. It is capable of subdividing elements, splitting or resizing elements
without affecting the connectivity of the elements. The “Refinement Management
Object” communicates with the “Fatigue Management Object” to get the locations of
stress concentrations and crack initiation. For crack propagation analysis, the
“Refinement Management Object” is capable of modeling the crack tip and inserting the
appropriate singular crack-tip elements. Although crack propagation analysis is beyond
the scope of this thesis, this feature is implemented for future extensions. The
“Refinement Management Object” communicates with the “Corrosion Management
Object” to geometrically apply thickness reduction or element removal based on the
specified corrosion rate for each element. For example, in order to simulate corrosion for
a given corrosion thickness loss history, the 2D elements change their thicknesses with
time, while the 3D solid elements adaptively move their faces inwards to simulate the
effect of loss of thickness over time. This happens by having all elements communicate
together to know the outer faces, and decide which faces will be moved. More details on
this technique will be demonstrated in Chapter 5.
90
4.3.2.6 Element Damage Management Object
The “Element Damage Management Object” is responsible for storing all damage
and damage rates information for each integration point in the element. The rules for
strain based damage models are covered in Chapter 2. The “Element Damage
Management Object” manages the averaging of damage per element. Also, it detects the
stress gradient per element and communicates with the “Refinement Management
Object” to refine the element to capture this gradient.
4.3.2.7 Spectrum Management Object
The “Spectrum Management Object” is responsible for managing all input time-
history spectra (load spectra, corrosion rates spectra and temperature spectra). It gathers
data from all spectra at the current simulation time and then applies the appropriate
values corresponding to the current time. The “Spectrum Management Object” interacts
with the “Fatigue Management Object” to get a reasonable value for the cycle jumps (as
will be discussed in section 4.3.3).
4.3.2.8 Input/Output Management Object
The “Input Output Management Object” is responsible for importing data files
from different finite element software packages into CorrFLP. It is also responsible for
exporting CorrFLP data files to other finite element software input file formats. This
manager is responsible for generating animation frames for the gradual damage
happening to the studied model. Figure 4.7 shows the file formats supported by CorrFLP.
91
Figure 4.7 Supported formats for the Input Output Management Object
4.3.2.9 FE Components Object
The “FE Components Management Object” acts as a container for all CorrFLP‟s
finite element objects such as Elements, Nodes, Sections, and Surfaces, etc. It is
responsible for adding, modifying or removing those components and synchronizing
them. Figure 4.8 shows the components managed by the “FE Components Management
Object”.
Figure 4.8 Components managed by the FE Components Management Object
92
4.3.2.10 FE Containers Object
The “FE Containers Management Object” acts as a container for container-types.
Container-types are groups that contain several objects. For example, an “Element Set” is
a container-type that refers to a group of “Elements”. Another example for container-
types is a “Part” which physically represents a component. A “Part” could include many
elements, nodes, surfaces, sections, etc. An “Assembly” is a collection of “Parts” that
make up the whole component. The program uses the same terms and conventions used
in most finite element packages. Figure 4.9 shows the containers managed by the “FE
Containers Management Object”.
Figure 4.9 Containers managed by the FE Containers Management Object
4.3.2.11 Searching Management Object
The “Searching Management Object” is responsible for tracking and searching all
components in the smart mesh. It could fetch any component in the mesh based on
various searching criteria. For example, the “Searching Management Object” can fetch an
“Element Set” of “Elements” that lie within a certain radius in the vicinity of a certain
93
point in space. Moreover, the “Searching Management Object” can fetch “Elements” that
are damaged beyond a certain level.
4.3.2.12 Statistics Management Object
The “Statistics Management Object” is responsible for statistically populating the
damaged elements and the proposed cycle jumps at each integration point. This manager
helps other managers of the smart mesh in taking decisions.
4.3.2.13 Mesh Info Management Object
The “Mesh info Management Object” is responsible for storing all project
information such as project name, current cycle number, current cycle jump, current time,
etc.
4.3.3 Cycle Jumps
A loading cycle has two peak amplitudes; one in tension and one in compression,
with respect to a certain mean stress. To simulate a cycle with minimal computation
effort, one needs to perform at least two analyses; one for tension loading and another for
compression. One way for having a time efficient analysis is to jump or skip some cycles
by considering a “super-cycle” that is equivalent to a certain amount of cycles and having
its same effect. The concept of cycle jumps was adopted by many researchers using
different cycle jump criteria, for further details, refer to Paepegem et al. (2001) and
Cojocaru (2006).
94
4.3.3.1 The cycle jump concept
CorrFLP automatically chooses the size of the “cycle jump” depending on the
expected life of the component based on how much its mesh is damaged and within the
limits specified by the user. Figure 4.10 illustrates the concept of cycle jump. The cycle
jump technique adopted in CorrFLP is the same one used by Paepegem et al. (2001).
Since each fatigue loading cycle represents a physical amount of time, then the cycle
jump for all elements of the mesh must be the same, but it has to be carefully chosen such
that it captures the flow of damage in the critically stresses parts of the mesh without
sacrificing the sensitivity of the analysis. Mesh zones having relatively low stress levels
and low damage rates would be rather safe to jump relatively larger cycle jumps without
leading to errors in the damage distribution. On the other hand, other zones with high
stress levels and high stress gradients or having high rates of damage flow, must have
small cycle jumps that are small enough to capture the stress redistribution to the
neighbouring zones and consequently accurately model the flow of damage in the mesh
elements.
First Cycle Second Cycle Third Cycle

Jump Jump Jump
Simulated Extrapolated
Cycle Cycle
dD dD dD dD
dN 1
dN 2 dN 3 dN 4
N=1 N=3 N=5 N=8
Figure 4.10 The super cycle and cycle jump concept Paepegem et al. (2001)
95
4.3.3.2 Calculation of Local Cycle jump
CorrFLP automatically calculates the proposed cycle jump for each element in the
mesh at each of its Gauss points based on its current damage state and the previous rate
of damage. Damage can then be extrapolated based on the chosen cycle jump at each
Gauss point using the simple Euler explicit integration formula:
dD
DN N  DN  N Local (4-1)
dN N
Where DN and DN  N are the damage states before and after the cycle jump;
respectively. dD is the previous rate of damage at the studied integration point. In order
dN N
to accurately capture the damage flow, the extrapolated damage DN N can be limited by
a certain increment; for example DN  D where D is the damage tolerance that can be
specified by the user. Based on this damage tolerance, a maximum local value (at each
Gauss point) for the cycle jump N Local can be determined. Figure 4.11 illustrates the
concept of limiting the number of cycles based on a certain limited extrapolated damage.
Damage







 N Local

 Number of Cycles
Figure 4.11 Damage extrapolation and corresponding cycle jump
96
4.3.3.3 Choice of the Global Cycle jump
The global cycle jump N Global for all mesh elements must be the same. The
simplest way for choosing N Global is by taking the minimum N Local . Although this
approach would be the most accurate method, but it is not recommended as it will
increase the processing time. Alternatively, N Global could also be taken as the number of
cycles NTime corresponding to a certain time interval specified by the user. The approach
adopted in CorrFLP would be by taking N Global as a percentage of the cumulative
relative frequency distribution of N Local without exceeding NTime . Figure 4.12(a)
shows the relative frequency distribution of the local cycle jumps N Local calculated at
each element‟s Gauss points. Figure 4.12(b) shows the cumulative relative frequency
distribution of the local cycle jumps N Local showing the choice of N Global as a chosen
percentage of the cumulative relative frequency, this percentage is recommended by
(Paepegem et al., 2001) to be less than 15%. The “Statistics Object” is responsible for all
these calculations.
Cumulative Relative Frequency



Relative Frequency
 
 
 
 



 

N Local 
N Local
(a) (b)
Figure 4.12 Choice of global cycle jump based on frequency distribution (Paepegem et
al., 2001)
97
4.3.4 CorrFLP’s Fatigue calculations:
4.3.4.1 Fatigue Life Prediction
Fatigue analysis in CorrFLP is based on the strain-based method. CorrFLP
implements all the endurance prediction rules listed in section 2.4.2. It also implements
the Rainflow counting algorithm for simplifying complex loading histories. Cycle
counting in CorrFLP takes into account the material memory effects by following the
material‟s hysteresis backbone curve as discussed in section 2.4.2. After each analysis
step, the material input for the next super-cycle is affected by the current damage state of
each element, for instance, Young‟s modulus would be degraded to Eo (1  D) where
Eo is the initial value for Young‟s modulus and D is the damage, which varies from 1 to
0 (1 for completely damaged, 0 for undamaged virgin material).
CorrFLP predicts the estimated number of cycles for crack initiation using the
theory of critical distances mentioned in section 2.4.5. Any element with high damage
rates and high stress gradients is further meshed (as a parent element) into smaller
elements (children elements) that are automatically chosen within 0.01mm to 0.05mm or
can be specified by the user to capture stress gradients at the critical distance inside the
material below the mesh outer surface. The sub-meshing process of parent elements into
children elements can take various techniques based on the type of element used and the
existing meshing, a sample sub-meshing technique is shown in Figure 4.13.
98
Critical Surface
Element in the mesh
Critical Element (Meshed)

(Children Elements)
Critical Element
(Parent Element)
Sub-Meshing
Figure 4.13 Sub-meshing of critical (parent) elements into fine (children) elements
Decisions are taken by the parent elements based on the total stress and strain
gradients within their children elements. Volumetric calculations within the parent
element are performed to calculate several factors such as the stress intensity factor range
K I . If the amount of damaged children elements passed the critical distance “L” that
can be calculated from (2-10), the number of cycles predicted NP is stored as an average
estimate for the fatigue crack initiation life. K th can be specified by the user based on
values published in literature. For a wide variety of values of K th for different materials,
the reader can refer to rich resource for fatigue thresholds in the compendium of fatigue
thresholds written by Taylor (1985). In air environment, the fatigue threshold K th in
various steels tested at a stress ratio of R=0.1 is equal to about 6.162 MPa m . The
critical stress range  o is equal to the fatigue limit the number of cycles elapsed
(Taylor, 2008). In order to cover some of the uncertainties involved in the calculations
99
and the error in the provided material fatigue properties. A ±15% in the calculated critical
length is proposed to give two fatigue life estimates for a specimen. The range of ±15%
was arbitrarily chosen, and it proved to result in very good predictions of fatigue life, as
will be shown later. Hence, CorrFLP provides an upper and lower bound for the
estimated crack initiation life corresponding to 0.85L and 1.15L. Where NP,Lower and
NP,Upper are the estimated crack initiation lives corresponding to the damage propagating
along critical distances of 0.85L and 1.15L; respectively. This range is a preliminary
estimate proposed in this research and can be modified by the user. Figure 4.14 shows the
propagation of damage in the (children) elements of a critical surface element.
Critical Distance "L"
Damaged Elements
Figure 4.14 Propagation of Damage in (children) elements
100
4.3.4.2 Analysis Procedure
The main analysis procedure is schematically described in Figure 4.15 in the form
of a flowchart showing the function of each manager in the process.
Main Analysis Loop
FCD IO Management Object Read FCD File Create Input File (.FCD)
Read starting time, finishing time and Load Spectrum

Spectrum Management Object
current cycle number
Temperature Spectrum
Corrosion Rate Spectrum
Spectrum Management Object Current time Fatigue Material Properties

< Finish time
Damage Rule
YES
Spectrum Management Object Read current Load Block
For Each Peak in

current Loading Block
Spectrum Management Object Change Load Amplitude
Abaqus IO Management Object Write Analysis File

Processed By :
Abaqus IO Management Object Run ABAQUS Solver
Abaqus IO Management Object Read Results to Mesh
Statistics Management Object Check if next loading needs Analysis
Proceed to next peak
Fatigue Management Object Apply Fatigue Damage
Corrosion Management Object Simulate Corrosion
Temperature Management Object Simulate Temperature
Apply Interaction rules for Fatigue,

Interaction Management Object
Corrosion and Temperature
Statistics Management Object Propose Next Time-Cycle Jump
Refinement Management Object Refine Mesh
FCD IO Management Object Keep Records of current cycle
Increment Time
by Time Jump
End of Analysis
Figure 4.15 Schematic flow chart for the analysis process showing the “Management
Object” responsible for each task
101
4.4 Post-Processing Phase
The program displays the mesh at studied time intervals by:
1. Plotting damage contours at each analysis step for each element in the mesh.
2. Plotting contours of the expected remaining life for each element.
3. Plotting contours of principal stresses and strains and their ranges for each element at
any time interval.
4. Plotting contours for stress concentration factors (with respect to a remote load in an
elastic analysis).
5. Animating the change in mesh with time displaying the input change at each cycle,
the time elapsed and the remaining life of the whole component.
6. Generating detailed reports on crack initiation locations and damage values and rates
at the integration points of each element.
4.5 Verification with experimental results
Using CorrFLP and the analytical techniques and procedures described earlier in
this chapter, fatigue analyses of several details from literature are carried out to predict
their fatigue lives. The following subsections present a description of the test setup,
results, description of the finite element models and analysis details.
102
4.5.1 Case Study 1: Plate with a circular hole
4.5.1.1 Test Description
In this case study, CorrFLP is verified using the experimental work reported by
Sehitoglu (1983). The researcher performed fatigue tests on hot rolled ASTM A36 steel
plates with a circular hole subjected to uniform tension. This detail is chosen because the
test data and details are well listed. Moreover, both the stress concentration factor and the
stress intensity factor for this detail are well established in literature.
4.5.1.2 Specimen Dimensions and Material Properties
The test specimens consisted of 280 x 50 x 5.7 mm with a 7.6 mm diameter hole at
its centre. The effective part of the plate is its middle 100 mm part; as there are about 90
mm of grip lengths at each end as illustrated in Figure 4.16. The specimens were tested
under load control conditions.
Modelled Zone
100 mm
D=7.6 mm 50 mm
Gripping Area Gripping Area
25 mm
50 mm
140 mm
5.7 mm
Figure 4.16 Test specimens from Sehitoglu (1983)
103
A fully reversed cyclic loading was applied. The cyclic stress vs. strain and fatigue
properties of the tested material have been reported by Sehitoglu (1983); E = 200,000
MPa, K' = 1336 MPa, n' =0.226,  'f = 1118 MPa, b = -0.11,  'f = 0.242 and c = -0.48.
Therefore, based on the cyclic parameters provided by Sehitoglu (1983), the cyclic
isotropic hardening material model can be expressed by:
1 / 0.226
  / 2   / 2 
   (4-2)
2 200000  1336 
where  and  are the strain range and stress range, respectively.
4.5.1.3 Finite Element Model
A 3D finite element model has been modeled. The reduced integration solid
continuum element C3DR from the ABAQUS FE library was used. The stress strain
curve is defined as a series of stress and strain pairs based on equation (4-2). Figure 4.17
shows the finite element model showing the loading, boundary conditions and meshing.
Figure 4.17 The finite element model (Loading on the left, meshing on the right) using
CorrFLP
104
Static nonlinear finite element analyses of the specimens were carried out at
several load levels. Each load level represents certain nominal stress amplitude on the net
area, which varied from 118 MPa to 271 MPa.
Several analyses have been performed for each stress level using different strain-
life prediction models such as the Smith-Watson-Topper (SWT) method, the Brown-
Miller and the regular strain endurance method. Figure 4.18 shows a snapshot of the
damage contours at crack initiation. Crack initiation life is determined as the micro-
cracks propagate to a certain critical length. The critical length crack initiation length in
CorrFLP was taken as an average of 0.077mm to cover the range of 0.066mm to
0.088mm reported by Sehitoglu (1983) so that the reported crack initiation lives could be
accurately compared.
Figure 4.18 Damage during the crack initiation life using CorrFLP
105
As can be seen in Figure 4.19, there is a very good agreement between the crack
initiation lives reported by Sehitoglu (1983) and the results obtained by CorrFLP using
different strain-life prediction approaches. This very good correlation can be attributed to
the simplicity of the model (simple plate with a hole) and the accuracy of the meshing.
The reduced integration elements C3DR were used as the fully integrated elements can
produce high gradients at locations of stress concentration which numerically gives
unrealistic fatigue life predictions. Moreover, this element is less computationally
expensive too. The results obtained by the SWT method were usually better than the
other approaches; this observation is in agreement with many other experimental
observations for metals (Drapper, 2008).
1.E+07
SWT
Experimentally Observed Number of cycles
Strain Life
1.E+06 Brown-Miller
(Sehitoglu, 1983)
Conservative
1.E+05
Non-conservative
1.E+04
1.E+03
1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
Predicted Number of Cycles using CorrFLP
Figure 4.19 Predicted lives using CorrFLP vs. experimentally observed by Sehitoglu
(1983) using three different criteria
106
4.5.2 Case Study 2: Bolted plates with staggered circular holes
4.5.2.1 Test description and geometry
In this case study, CorrFLP is verified using the experimental work done by Josi et
al. (1999). The researchers performed fatigue tests on hot rolled CSA G40.21 300W steel
double splice plates with circular holes in order to investigate the effect of bolt hole
stagger on fatigue resistance of bearing-type connections. The four series correspond to
different hole staggers, s, namely (S0) having s = 0 mm, (S1) having s = 25.4 mm, (S2)
having s = 50.8 mm and (S3) having s = 76.2 mm. The holes configurations and the
dimensions of the specimens are shown in Figure 4.20. The plate width = 95 mm. The
thickness of the main plates is taken as 25 mm, while the splice plates were chosen to
have a thickness of 9.5 mm to be the critical elements in the testing specimens. The 20
mm diameter bolt holes were match drilled for 19 mm bolts, which were installed to a
snug tight condition.
4.5.2.2 Materials and Properties
Three tension coupons were tested from the splice plate materials (Josi et al.,
1999). It was reported that the average elastic modulus is 212000 MPa; static yield
strength is 420 MPa. The cyclic fatigue properties were not reported, thus the material
properties presented by Sehitoglu (1983) for A36 steel are used for the finite element
analysis and the fatigue life predictions.
107
273.0
34.9 60.3 82.6 60.3 34.9
25.4
60.3 111.1
25.4
S0-Series (No stagger)
304.8
34.9 25.4 25.4 25.4 82.6 25.4 25.4 25.4 34.9
25.4
44.5 95.3
25.4
S1-Series (1 in. stagger)
457.2
34.9 50.8 50.8 50.8 82.6 50.8 50.8 50.8 34.9
25.4
44.5 95.3
25.4
609.6
34.9 76.2 76.2 76.2 82.6 76.2 76.2 76.2 34.9
25.4
44.5 95.3
25.4
Figure 4.20 Test specimens of Josi et al. (1999)
108
The plates were modeled using solid C3DR reduced integration elements. Only
one half was modeled due to symmetry. The bolts were assumed to be rigid relative to the
plates as the failure in all tests was in the plates, thus they were modeled as analytical
surfaces (a feature in ABAQUSTM that will be discussed in the next section.) to reduce
the calculations. Figure 4.21 shows the boundary conditions, loading and meshing of the
FE model for (S0) series. The top of the bolt heads are modelled rounded for simplicity;
this would not have an effect of the fatigue life prediction of the tested specimens.
Figure 4.21 The Finite Element model for series (S0) from Josi et al. (1999) (Meshing on
the left and boundary conditions on the right) using CorrFLP
4.5.2.4 Interaction surfaces
The analytical surfaces in ABAQUSTM can be used to model the bearing of bolts
on the plates by specifying appropriate interaction surfaces. To accurately model the
interaction of the bolt or rivet head with the plates, several interaction surfaces have to be
defined in the model for each bolt. The interaction surfaces will detect the bearing of any
surface on another and will smoothly transfer the loads accordingly. The formulation of
109
the interaction surfaces was assumed to be frictionless as the bolts usually become loose
with time and they do not provide significant pretensioning that can clamp the plates
together and transmit forces by friction. The surface-to-surface contact discretization with
small sliding formulation is chosen as it gives more accurate results. At each iteration
step in the analysis the constraints between the contacting surfaces have to be enforced
using one of several methods provided by the solver. The augmented Lagrange method
was chosen as it uses augmentation iterations to improve the accuracy of the
approximation. The way the solver works is by finding a converged solution with the
penalty method. If a slave node (on one surface) penetrates the master surface by more
than a specified penetration tolerance (usually one-tenth of a percent of the characteristic
interface length), the contact pressure is “augmented” and another series of iterations is
executed until convergence is once again achieved. ABAQUSTM solver continues to
augment the contact pressure and find the corresponding converged solution until the
actual penetration is less than the penetration tolerance.
The default penalty stiffness for the augmented Lagrange method is 1000 times the
representative underlying element stiffness. Lagrange multipliers are used for the
augmented Lagrange method if the penalty stiffness exceeds 1000 times the
representative underlying element stiffness computed by ABAQUSTM; otherwise, no
Lagrange multipliers are used. Therefore, Lagrange multipliers are not used for the
augmented Lagrange method with the default penalty stiffness.
Figure 4.22 illustrates the locations of the required interaction surfaces to
accurately simulate the interaction between the bolts and the plates. It is assumed that
there is no clamping force in the bolts such that the bolt is only bearing on the plates
110
without holding them tightly together. In bearing type connections where clamping forces
can be neglected, it can be assumed that the same modelling technique can be applied for
riveted and bolted connections.
Figure 4.22 Locations of interaction surfaces in the Finite element model
Since the specimens were not tested under a fully reversed condition, finite
element analysis was conducted at both the maximum load and the load amplitude used
in the tests to get the exact stress range and mean stresses as the experiment. Several
analyses have been performed for each stress level studied by Josi et al. (1999) using the
SWT strain-life prediction model. Figure 4.23 shows section cut through one line of bolts
in the analyzed model to show the relative movement between the plates and the stress
concentration at the edges of the bolt holes near the bolt head. Figure 4.24 shows section
cut through one line of bolts in the analyzed model to show the distribution of damage in
the specimen calculated by CorrFLP.
111
Figure 4.23 Section cut in the CorrFLP FE model showing the stress distribution and the
relative sliding between interaction surfaces (Scaled deformed shape for illustration)
Figure 4.24 Damage during the crack initiation life using CorrFLP surfaces (Scaled
deformed shape for illustration)
112
The results obtained by Josi et al. (1999) have a noticeable scatter within the three
different replicas of the each specimen; this is quite common in fatigue tests as the
number of cycles to failure could vary in an order of a magnitude, or even two, in some
cases. As the crack initiation lengths were not reported, so CorrFLP‟s default method
discussed in section 4.3.4.1 was implemented giving a range for the expected crack
initiation life. Table 4.1 shows the results obtained by Josi (1999) versus the range of
expected lives from CorrFLP. Figure 4.25 shows a plot of the results in Table 4.1 such
that the minimum expected result from CorrFLP is compared to the minimum
observation obtained from the experimental work and vice versa.
113
Josi (1999) CorrFLP
Load Observed Number
Specimen NP,Lower NP,Upper
Range Of Cycles
S0 a 190 2167000
S0 b1 190 5720000 1500000 2100000
S0 b2 190 767000
S0 c1 240 355000
S0 c2 240 620000 400000 460000
S0 d 240 578000
S1 a1 200 255000
S1 a2 200 303000 225000 350000
S1 b 200 1558000
S1 c1 160 2827000
S1 c2 160 2900000 2750000 5000000
S1 d 160 8520000
S2 a1 200 1062000
S2 a2 200 307000 300000 450000
S2 b 200 649000
S2 c1 160 3801000
S2 c2 160 1497000 1500000 2500000
S2 d 160 2650000
S3 a1 200 675000
S3 a2 200 695000 450000 600000
S3 b 200 515000
S3 c1 130 3562000
S3 c2 130 13673000 3300000 6000000
S3 d 130 Runout
Table 4.1 Observed Results from Josi et al. (1999) vs. minimum and maximum expected
cycles to crack initiation from CorrFLP
114
log (Nf )
8.0
7.0
Numerical Experimental
(CorrFLP) (Josi 1999)
6.0
5.0
(a,b) (c,d) (a,b) (c,d) (a,b) (c,d) (a,b) (c,d) Specimen

S0 S1 S2 S3 Group
Figure 4.25 Predicted lives using CorrFLP vs. experimentally observed by Josi et al.
(1999) using three different criteria
4.5.2.6 Discussion of results
It had been reported that most of the shear splices showed single cracks at the time
of failure which was always the case in CorrFLP simulations due to the stress
concentrations near the holes. The fatigue life of the test specimens varied from 3x105 to
13x106 cycles depending on the geometry and the applied load level of the bolted
connection. This considerable scatter in the experimental results in a relatively simple
connection can be attributed to several factors; such as the amount of clamping forces in
the bolts that can help in increasing the fatigue life of the specimen. The determination of
the exact clamping force is one of the main problems of the fatigue assessment of the
existing bolted connections because it depends on how the bolt was initially hammered
and its pretensioning loosens with time. The modeled specimens in CorrFLP had no
115
clamping forces assumed to get the most conservative condition. Moreover, the existence
of any micro-cracks or defects can cause premature failure in the specimens. The results
had a wide scatter that shows the probabilistic nature of fatigue problems.
It is worth mentioning that Josi et al. (1999) reported that the regression analysis of
their test results indicates that the slope of the fatigue curve for the bearing-type shear
splices tested has a slope, m, equal to 7. This is significantly different from the slope
defined in the modern codes for other fatigue (welded) details that all have a slope of
m=3. This conclusion should not be generalized on all bolted connections, but it points
out that categorizing all bolted details under a single fatigue category is not accurate in
most of the design codes and needs future investigation. This emphasises on the
importance of carrying out further experimental and/or numerical simulations on bolted
structural details to correctly classify them.
4.5.3 Case Study 3: Built-up beams
4.5.3.1 Introduction
In order to calibrate and verify the fatigue life prediction of CorrFLP for more
complex assemblies such as built-up riveted girders, another case study will be simulated
and verified against some of the experimental work conducted by the NCHRP (National
Cooperative Highway Research Program, NCHRP (1988). The NCHRP has published
several reports to determine the fatigue categories and behaviour of several steel details.
These reports helped shaping the current design standards for fatigue.
116
4.5.3.2 Experimental Tests
Fourteen full-scale riveted girders, all removed from riveted steel bridges, were
tested by Fisher et al. (1988) in the NCHRP report 302 to evaluate their fatigue and
fracture resistance. The test girders were obtained from three different sources Girders 1
to 8 were obtained from the Santa Fe Railroad. Girders 9 to 12 were supplied from an
Ocean County, New Jersey, highway bridge that was dismantled. Girders 13 and 14 were
removed from the Minsi Trail Bridge in Bethlehem, Pennsylvania, at the time it was
dismantled. Except for Ocean County girders, which were heavily corroded, the other
girders were generally in good condition. The stress ranges of the cycles applied before
dismantling the girders were mostly below fatigue limit as it has been reported by Fisher
et al. (1988) that no fatigue cracks were detected in any of the test girders before the
laboratory tests. Thus it could be said that the loading history prior to laboratory testing
can be ignored. Figure 4.26 and Figure 4.27 show the geometry and profiles of the test
girders.
117
Figure 4.26 Riveted Girders tested by Fisher et al. (1988)
Figure 4.27 Cross-Section of tested girders Fisher et al. (1988)
118
Only three out of the fourteen bridges will be selected and used for this
verification; only the tests conducted in the room temperature without any severe
corrosion will be used for verification. Therefore, Minsi trail bridge girders and the
Ocean Country girders will be excluded from this study. Unfortunately, no corrosion
rates or any data on the corrosivity of the environment were reported for the Ocean
country bridges which were the only bridges that were severely corroded. Table 4.2 lists
the tested specimens, their sources and the minimum, maximum flexural stresses applied
during testing.
Table 4.2 Selected Specimens in this study from NCHRP Report 302 (1988)
Minimum Maximum Stress
Source Specimen ID Specimen ID
Stress Stress Range
Bridge (Current Study) (NCHRP report 302)
(MPa) (MPa) (MPa)
Santa Fe SF-1 1 14 117 103
Santa Fe SF-2 2 14 117 103

Santa Fe SF-3 4 14 97 83
4.5.3.3 Materials
Tensile tests were conducted for each of the different types of girders. The results
show that all of the test girders were fabricated from mild steel with yield strength
between 200 MPa and 290 MPa. Table 4.3 lists the static material properties of the girder
parts. No fatigue properties for the strain based method have been reported as the tests
were targeting the stress based method.
Table 4.3 Material properties of NCHRP report 302 (1988)

Source Yield Stress (MPa) Ultimate Stress (MPa)
Bridge Angles and Webs Cover Plates Angles and Webs Cover Plates
Santa Fe 198 218 371 399
119
4.5.3.4 Experimental Test Procedure
The girders were tested under four-point bending. Two 490-kN Amsler jacks were
used to load each test girder. A 1.52 m constant moment length between the jacks
permitted a significant length of beam to be subjected to the same stress range. In order to
provide lateral stability of the girder compression flange, two lateral bracing bars were
attached to the compression flanges. Two lateral bracing bars were connected to the
bottom flange to simulate the restraint of the bracing system in the bridges and to
minimize the lateral movement of the tension flange. Cyclic loading was applied so
that the bending stresses at the outer fibres of the lower cover plates would conform with
the values listed in Table 4.2.
The girders were modeled in ABAQUS to be simulated for fatigue analysis in
CorrFLP. All the girders where modelled using the C3DR reduced integration elements.
Surface interaction between rivet surface and the attached plates are modelled using the
same techniques discussed in section 4.5.2.4. To reduce the computational time of the
model, only one quarter of the beam cross section is modeled. It had been reported that
all failures were in the zone of constant moment, thus a small segment of the beam length
at the constant moment zone will be considered as shown in Figure 4.28. A denser mesh
is provided around the holes susceptible to crack initiation, to accurately monitor the
damage propagation.
120
Figure 4.28 Extraction of modeled part in CorrFLP analysis
4.5.3.6 Boundary Conditions and Loading
Symmetry boundary conditions are applied along the center plane of the web. A
fixation boundary condition is applied to the surfaces along the girder center line. All
rivets where restrained from rotation about their axis. A gradient stress is applied such
that the stress at the girder horizontal centerline (at zero strain) and the stress at the
extreme fibres is taken from the values listed in Table 4.2. Figure 4.29 shows the loading,
boundary conditions and meshing of the model representing Santa Fe girders.
121
Figure 4.29 Finite Element Model for Santa Fe Girders (Loading and boundary
conditions on the left, meshing on the right)
4.5.3.7 Modeling material properties
The materials were considered as elastic perfectly plastic. Modulus of elasticity
was taken as 200,000 MPa. As no fatigue strain based properties were reported, cyclic
fatigue properties were assumed using the Unified Material Method mentioned in section
2.4.2.6. Since the static properties E = 200,000 MPa,  u = 400 MPa for Santa Fe and
Ocean Country Girders are same, thus the calculated cyclic stress strain and fatigue
properties for both girders are; K' = 660 MPa, n' =0.15,  'f = 600 MPa, b = -0.087,  'f
= 0.59 and c = -0.58.
The same procedure discussed in section 4.3.4.1 was adopted. Table 4.4 lists the
number of cycles to detection of fatigue cracks reported by Fisher et al. (1988) versus the
minimum and maximum estimated number of cycles to crack initiation obtained from
122
CorrFLP. Figure 4.30 plots the number of cycles to detection of fatigue cracks reported
by Fisher et al. (1988) versus each of the minimum and maximum estimated number of
cycles to crack initiation obtained from CorrFLP.
Table 4.4 Fatigue Crack initiation lives for Santa Fe Girders from the NCHRP report 302
(1988) and from CorrFLP simulation
NCHRP report 302 CorrFLP
Stress Cycles to Detection
Specimen NP,Lower NP,Upper
Range of fatigue cracks
SF-1 103 588000 500000 750000
SF-2 103 1094000 500000 750000
SF-3 83 2630000 1900000 2800000
log (Nf )
7.0
6.0 Numerical Experimental

(CorrFLP) (Fisher 1988)
5.0
Specimen
SF-1, SF-2 SF-3
Figure 4.30 Plot for Fatigue crack initiation lives for Santa Fe Girders from the NCHRP
report 302 (1988) vs. those from CorrFLP simulation
123
All of the cracks in Santa Fe girders were reported to initiate from the rivet holes.
This was clear from the analysis of damage propagation patterns obtained from CorrFLP.
Moreover, there is a very good correlation between the life estimations obtained from
CorrFLP and the observed number of cycles reported by Fisher et al. (1988). Figure 4.31
shows, as an example, the crack initiation and propagation patterns reported by Fisher et
al. (1988) versus the crack initiation patterns obtained from CorrFLP.
Figure 4.31 Reported crack initiation and propagation (Fisher et al., 1988) vs. CorrFLP
damage initiation for specimen SF-2
124
4.5.3.9 Summary
The built-up riveted girders used in this verification example are widely used in
many existing bridges. It is very important to be able to simulate them with an acceptable
level of confidence. Results obtained from Santa Fe Girders are of very good agreement
with the experimental results obtained. Thus, the modeling technique using the FEM
adopted in this validation example is convenient and can be extended to any built-up
girder configuration. It is important to note that this technique is sensitive to the boundary
conditions and the interactions applied, thus it is important to calibrate the model using
experimental tests or field measurements ensure that the finite element model is
representative of the real girders. Moreover, the fatigue crack initiation lives predicted
using CorrFLP are usually on the conservative side, due to the assumptions of neglecting
the friction between plates and neglecting the clamping forces in the rivets. In fatigue
simulations, it is common that results within an order of magnitude or two are considered
to be acceptable. Thus, the level of accuracy of the results obtained by CorrFLP, which
were usually within one order of magnitude, can be considered as a good agreement with
the experimental results. Moreover, it is worth noting that the fatigue lives estimated by
CorrFLP are usually on the safe side.
4.6 Discussion
Three case studies were conducted to verify the accuracy of CorrFLP. The first is a
plate with a hole to be used as a benchmark test for CorrFLP. The numerical predictions
showed very good agreement with experimental results; this gives confidence in the
numerical tool in performing fatigue analysis for simple models. The study was then
125
extended to bearing-type shear splice plates connected together using bolts subjected to
shear. The model is more complicated and involves interactions between surfaces. The
analysis was performed ignoring the effect of clamping forces to give the most critical
(conservative) fatigue life estimate. Numerical predictions from this study had a good
agreement with the experimental results. This is a good indicator of the capability of the
developed program to test more complex plate assemblies in fatigue. A third case study
was conducted on built-up steel riveted girders which are commonly used in existing old
bridges. The fatigue life predictions from CorrFLP were of good agreement with the
experimental results. This gives a high level of confidence in the results obtained by the
developed tool. Next chapter will discuss the procedure for extending this tool to simulate
the fatigue behaviour of corroding elements.
126
5 CHAPTER 5
EFFECT OF CORROSION ON FATIGUE LIFE
5.1 Introduction
Corrosion is a process of degradation of a metal by an electrochemical reaction
with its environment. This means a loss of electrons of metals reacting with water and
oxygen. The weakening of iron due to oxidation of the iron atoms is a well-known
example of electrochemical corrosion which is commonly known as rust (Trethewey et
al., 1988). Much of the corrosion damage in structures within the last 40 years might have
been caused by a change in the chemical composition of the air due to increased
industrial activity (Boden, 1989). Atmospheric deterioration is influenced by the
following natural factors: moisture, temperature (mean value and variations), sunlight, air
movement (wind speed and direction), sea salt, and fog. Some experiments in literature
(Vernon, 1935) indicate that corrosion at relative humidity below 50% is minimal. As the
relative humidity increases from 60% to 80% or even greater, the protective oxidized
layer on the metal surface breaks down and allows corrosion. Moisture in the form of rain
may, however, reduce atmospheric corrosion by washing away dangerous pollutants.
As previously discussed in Chapter 2, there is not so much experimental data for
fatigue tests on corroded structural components. Moreover, there are almost no tests
available for full scale specimens that are cyclically loaded along with the environmental
effect of corrosion being simulated simultaneously. This shows the importance of the
numerical tool developed in this thesis to simulate the simultaneous effects of corrosion
along with cyclic loading.
127
5.2 Implementing the effect of corrosion
There are two proposed approaches for simulating the effect of corrosion; the first
one is by geometrically simulating the thickness loss and pitting. The other approach is to
model the deterioration in material properties due to corrosion by implementing a newly
proposed strain-life fatigue model that simulates the degrading effect of corrosion on the
fatigue life. The next subsections will cover the aforementioned approaches.
5.2.1 Geometrical representation of corrosion
CorrFLP can simulate the thickness loss due to corrosion; moreover, it can
approximately simulate the effect of pitting corrosion by applying a stress concentration
factor based on the pit dimensions. CorrFLP‟s “Corrosion Management Object” is
responsible for communicating with the “Spectrum Management Object” to know the
current time and current corrosion rate and thus selects the appropriate corrosion
thickness loss (penetration). The “Corrosion Management Object” also communicates
with the “Main Mesh Object” and locates the points of high stress concentration, to be
points of initiation of stress corrosion, given any model that could be developed for it in
future. The user can also specify any node in the finite element model and identify it as
an expected location of pitting corrosion initiation. The user can specify the initial pit
aspect ratio to model an existing corrosion pit. CorrFLP can calculate the stress
concentration factor for the stresses at this node using the numerical method provided by
Cerit et al. (2009) as shown in Figure 5.1. This feature is included for future work, as the
topic of pitting corrosion is so complicated and is out of the scope of this research.
128
Figure 5.1 Stress concentration factor due to pitting corrosion depending on pit shape
(Cerit et al., 2009)
CorrFLP can also incorporate many available corrosion damage functions to
account for different atmospheric conditions and different concentrations of sulphur
dioxide and oxidants (when this data is available for the user). Geometrical simulation of
the corrosion thickness loss is done by communication between the “Corrosion
Management Object” and the “Refinement Management Object” where the latter
geometrically moves the external surfaces or faces of the corroding elements by the
amounts decided by the “Corrosion Management Object”. The amount of corrosion
penetration t p applied to any surface can be calculated by multiplying the corrosion rate
at the current time t p (t ) by the time corresponding the cycle jumps applied as shown in
equation (5-1).
t p  t p (t )  time jump (5-1)
129
Each surface can have a different corrosion rate history attached to it to
realistically simulate the non-uniform corrosion rates of different surfaces in some
structural assemblies. Figure 5.2 shows how specific surfaces can be selected and
assigned a specific corrosion rate on a sample FE half-model of a built-up I-beam. Figure
5.3 shows how CorrFLP simulates the corrosion thickness loss in the FE model shown in
Figure 5.2.
Figure 5.2 Assigning corrosion rate factors for mesh surfaces in CorrFLP
Figure 5.3 Simulation of corrosion thickness loss on selected critical surfaces

(Left: No Corrosion, Right: Corrosion (not shown to-scale))
130
The user can assign different corrosion histories and different rates to different
surfaces on the model to simulate the real conditions. For example, while modeling a
built-up I-beam, there are several zones of corrosion severity, these zones could be
specified by the user. A useful resource (Kulicki et al., 1990) for the various locations
susceptible to have more corrosion damage can be used as a guideline for the user.
CorrFLP also provides several environmental corrosion damage functions provided by
Boden (1989).
5.2.2 Proposed corrosion fatigue strain-life model
It is very important to note that due to the complexity of the corrosion-fatigue
phenomenon and due to the very long time its experiments take, there are very limited
experimental data, especially in the field of civil engineering on structural steels.
In this research, several approaches for modeling the corrosion-fatigue were
investigated. One of the approaches was to use a modified corrosion-fatigue stress
intensity threshold K th for the studied material and use it for determining the critical
length L for the studied specimen in a corrosive environment. This approach is difficult
and unpractical as many researchers such as Novak (1983) and Taylor (1985) reported
that they could not get a clear value for K th in corrosive environments. Moreover, the
fatigue limit is not usually clear in corrosive environments.
Since the critical distance “L” depends on both the K th and the fatigue limit
which are not clear in corrosive environments, thus this approach will not be useful until
further detailed research is done in this particular point. Alternatively, the hot-spot
method –regularly used in fatigue prediction software– will be used in the developed
131
numerical tool with a newly proposed strain-life model. This method uses the maximum
strain values at the surfaces of the elements.
Before proposing the model, a brief introduction to some of the strain-life
approach basics should be done. As it is already known, the endurance of any material to
cyclic loading can be determined using any of the strain-life methods mentioned in
section 2.4.2 by using an elastic part and a plastic part. The elastic part is usually defined
by the fatigue strength exponent, b, also known as Basquin‟s exponent which is a
material property. The plastic part is defined by the fatigue ductility coefficient, c, which
is a material property known as Coffin-Manson exponent. Combining both the elastic and
plastic parts from the total endurance life as shown in Figure 5.4.

2
'
'
E
b
elastic
plastic
Endurance (2Nf )
Figure 5.4 Relationship between total strain amplitude and endurance in a non-corrosive
environment
132
In this research, it is proposed to experimentally obtain modified values for b and c
for a certain corrosive environment and they will be denoted as b' and c' respectively.

2
'
c'
Corrosive
Environement
'
E Non-Corrosive
b' Environement
c
Endurance (2Nf )
Figure 5.5 Relationship between total strain amplitude and endurance in a highly
corrosive environment
Four new factors will be proposed, namely  b ,  c ,  corr ,   . The factor  b will be
the ratio of b' / b where b' is measured in a standard NaCL 3.5% highly corrosive
environment. Similarly,  c will be the ratio of c' / c where c' is to be measured in a
standard NaCL 3.5% highly corrosive environment as a way of standardizing the
procedure. The factor  corr will be the environmental corrosivity intensity factor varying
from 0 to 1, where 0 corresponds to a totally non-corrosive environment and 1
corresponds to a highly corrosive environment equivalent to the NaCL 3.5% solution
simulation environment usually used in accelerated corrosion tests. The factor   will be
introduced as a correction factor for the mean stress effects. Once we have our
experimental data plotted, b' and c' corresponding to  corr  1 can be easily calculated,
133
then both  b and  c can be obtained and saved as material constants for a harsh
environment corresponding to  corr  1 .
For any other corrosive environment corresponding to  corr  0  1 , both b' and c'
can be easily calculated using the following equations:
b'  b (1   corr  b ) (5-2)
c'  c (1   corr   c ) (5-3)
From surveying the literature and from the results obtained in the case studies
conducted in the previous chapter, it has been observed that the SWT model mentioned in
section 2.4.2.3 predicts the fatigue endurance of metals with a relatively high accuracy as
it takes into account the mean stress effects. In this research, a proposed modification is
introduced to the SWT strain-life method to accommodate for the different environment
systems will be by replacing b by b' and replacing c by c'. Thus the modified SWT
model format becomes:

 max 
 'f   2 N 
2
2b '
  'f  'f 2 N f 
b ' c '
f (5-4)
2 E
Where the factors b' and c' incorporate the factor  corr , which varies from 0 to 1
depending on the corrosivity of the environment. The factor   is a correction factor that
depends on the maximum applied stress  max .
134
The assumptions made in this proposed strain based corrosion model are logically
acceptable and the constants proposed for each material can be derived from
experimental data. Yet, it would be difficult to accurately verify the exact values for both
 b and  c for a certain material (in the current time) as there is very limited corrosion
fatigue data in literature. Thus, there would be a need for testing enough structural
materials in order to have a comprehensive database of their proposed environmental
properties. The proposed environmental corrosivity intensity factor  corr could be
correlated to the categories provided by the ISO-9223 (1992) and the ISO-9224 (1992) as
shown in Table 5.1.
5.2.3 Calibration of the proposed model
In this research, a relation is proposed to correlate the environmental corrosivity
intensity factor  corr with the average annual penetration in (m/year). This relationship
was deduced by calibrating with the work done by Albrecht et al. (1994) which will be
further discussed in the next section. Equation (5-5) shows the proposed relation between
the logarithm of the penetration versus the proposed corrosion factor  corr .
 corr  0.0761 log( pa )2  0.2109 log( pa )  0.221 (5-5)
where pa is the average annual penetration in (m/year). Figure 5.6 shows the plot
used to deduce the proposed equation.
135
1.2
0.8
 corr
0.6
0.4
y = 0.0776x2 + 0.2121x + 0.2179
R² = 0.9806
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
log (penetration) m/year

Figure 5.6 Plot for the proposed relation between the log of the penetration versus the
proposed corrosion factor  corr .
Table 5.1 Correlation between and the ISO-9224 categories

ISO-9224 category
 corr
(ra = average corrosion rate)
Very low corrosivity C1 ( ra  0.1m / year ) 00.09

Low corrosivity C2 ( 0.1  ra  2m / year ) 0.090.29
Medium corrosivity C3 ( 2  ra  8m / year ) 0.290.47
High corrosivity C4 ( 8  ra  15m / year ) 0.470.57
Very High corrosivity C5 ( 15  ra  80m / year ) 0.570.9
Due to the limited experimental corrosion fatigue life data available for plain
specimens made of structural steel presented in the form of strain-life data points, it is
necessary to be able to convert any available S-N data points for a corrosive environment
and to convert them to strain-life data points. In this research, the author was able to
obtain stress-life (S-N) corrosion fatigue life data for plain coupons of weathering steel
(which is usually used in bridges) tested by Kunihiro et al. (1972). Kunihiro et al. (1972)
136
tested plain specimens fabricated from SMA weathering steel commonly used in Japan,
which is also similar to the A588 weathering steel commonly used in North America
(Albrecht, 1983). All specimens were 19 by 50 by 459 mm. Figure 5.7 shows a typical set
of Kunihiro‟s data showing the 0 years (unweathered) coupons versus the 2 years
weathered coupons.
Figure 5.7 Fatigue strength of two year weathered plain rolled SMA steel specimens
(Kunihiro et al., 1972)
The data was then digitized and transformed into strain-endurance life data points
by using the cyclic fatigue properties of this steel which are K' = 1400 MPa, n' =0.15
using equation (5-6). The S-N data was adjusted by using an average notch factor of 1.3
reported by Kunihiro et al. (1972).
1 / 0.15
  / 2   / 2 
   (5-6)
2 200000  1400 
137
The properties of SMA weathering steel is of an average of 580 MPa tensile
strength, 355 MPa yield point and E = 200 000 MPa. Using the universal material method
of Baumel et al. (1990) we obtain K' = 1400 MPa, n' =0.15,  'f = 870 MPa, b = -0.09,
 'f = 0.59 and c = -0.58. By processing the data, we can obtain e-N data points for the
steel as shown in Figure 5.8. The curve fitted with no corrosion had a best fit with the
fatigue exponents b = -0.11 and c = -0.59 which are very reasonable values for steels. The
curve fitted with 2-year corrosion had a best fit using b' = -0.13 and c' = -0.61. Thus, the
proposed corrosion material constants to be  b = 0.182 and  c = 0.034 for this material.
2.50E-03
e-N Data (2 Years Corrosion)
e-N fit (2 Years Corrosion)

2.00E-03
e-N Data (No Corrosion)
e-N fit ( No Corrosion)
1.50E-03
/2
1.00E-03
5.00E-04
0.00E+00
4.00E+04 4.00E+05 4.00E+06 4.00E+07
Endurance (2Nf)
Figure 5.8 Kunihiro et al. (1972) data points converted into strain-life data points and
fitted
Kunihiro et al. (1972) tests were performed in an open exposed environment, the
severity of this environment was not reported, thus it was not possible to correlate the
proposed environmental corrosivity factor  corr to environment. In this case, the
properties of SMA steel can be reported for a regular service environment as the product
138
 corr    b = 0.182 and the product  corr    c = 0.034. If the environmental data is
available,  corr can be interpolated based on the average penetration per side as an
indicator of the corrosivity of the environment. Figure 5.9 shows the corrosion
penetration of several types of steel as reported by Albrecht (1983). Further research is
recommended to further correlate different environments to the proposed factor  corr .
Figure 5.9 Corrosion of weathering steel compared with copper-bearing and mild steels
(Albrecht, 1983)
5.2.4 Implementation in CorrFLP
The proposed technique is adopted in CorrFLP, where the user should be able to
provide the material-environment properties  b and  c of the tested material. Based on
the environmental data available about the surrounding environment, the user can specify
the level of corrosivity using the factor  corr . The fatigue life prediction cannot be
performed the same way as described in section 4.3.4.1 due to the inaccuracy of the
139
fatigue threshold in corrosive environments. Thus, the theory of critical distances cannot
be applied. In this study, the hot-spot approach will be used (which uses the values of
strain on the surface of the specimen). In order to model the probabilistic nature of
fatigue, a range of values of  corr corresponding to  corr  0.05 will be used to give a
predicted life range. Moreover, the user can also provide several corrosion thickness loss
histories and apply them to different surfaces on the studied element to accurately
simulate the real conditions.
5.3 CorrFLP (Corrosion Module) Verification
5.3.1 Introduction
In order to calibrate and verify the fatigue life prediction of CorrFLP in corrosive
environments, a real case study will be simulated and verified against published test
results. The chosen experimental investigation was done by Albrecht et al. (1994) to
investigate the fatigue behaviour of 25 corroded rolled beams made of A588 steel. The
beams were weathered under two conditions. One set was boldly exposed to the
environment for 62 months and stress cycling in a moist freshwater environment. The
other set was exposed for 67 months under a metal deck that simulated the shelter
provided by highway bridge decks and was lightly sprayed with a salt solution during the
winter months to simulate the use of de-icing salts for snow removal. A spreader beam
distributed the load to two points spaced 914 mm (3 ft) apart as shown in Figure 5.10.
The loading frequency was 0.75 Hz. The period of 1/0.75 = 1.33 s corresponds to
the time needed for a truck to cross a 30 m long highway bridge at a speed of
80 km/h (50 mph).
140
Figure 5.10 Configurations of tested beams (Albrecht et al., 1994)
The corrosion rates were monitored and reported by Albrecht et al. (1994) as
shown in Figure 5.11. These values were used to generate suitable corrosion rate histories
for each face (surface) of the finite element model representing the studied beam.
Figure 5.11 Test matrix and corrosion rates table adapted from Albrecht et al. (1994)
141
5.3.2 Material
The beams were made of A588 grade B weathering steel ("A588 Standard" 1992)
with average 577 MPa tensile strength, 408 MPa yield point (regularly 345 MPa) and
E = 200 000 MPa. Using the universal material method Baumel, et al. (1990) we obtain
K' = 1400 MPa, n' = 0.15,  'f = 870 MPa, b = -0.09,  'f = 0.59 and c = -0.58. Since it was
reported by Albrecht (1983) that the properties of SMA steel are similar to those of A588
steel, the proposed environmental constants will be taken for the boldly exposed
specimens as concluded in section 5.2.2 as  corr    b = 0.182 and  corr    c = 0.034.
By comparing the values of corrosion penetration rates reported by Albrecht et al. (1994)
and the values listed in Table 5.1, the factor  corr = 0.45 was selected as an average for
the specimens boldly tested in air. Also, the factor  corr = 0.75 was selected as an average
value for the specimens tested in moist freshwater environment. For the sheltered
specimens tested in moist saltwater environment, which is a very harsh environment, the
values from the ISO-9224 were extrapolated and the factor  corr = 1.1 was used as an
average value to represent this environment. As the critical length theory will not be used
while including the effect of corrosion, the approach of using percentages of the critical
length “L” to get an estimated range for the expected fatigue life cannot be applied. In
order to get a realistic estimated range for the expected life for each specimen, the
expected lives corresponding to  corr  0.05 will be used.
142
5.3.3 Finite Element Model
A 3D finite element model has been modeled. The reduced integration solid
continuum element C3DR from the ABAQUS FE library was used. Quarter of the beam
is modelled using the symmetry about the section centerline and the beam centerline. One
half of each concentrated load is applied to the top of the flange. Figure 5.12 shows the
finite element model showing the loading, boundary conditions and meshing.
Figure 5.12 Finite element model for the girder (loading and boundary conditions in the
top, meshing and cross-section in the bottom)
Twenty four beams were modeled; the loading was based on stress ranges for the
unweathered sections. Fatigue calculations were based on the final weathered sections.
Corrosion rates for each part of the specimens were reported by Albrecht et al. (1994) and
all the geometric and loading input values are listed in Table 5.2.
143
Table 5.2 Input data for the finite element model
Non-Weathered Section Weathered Section

Stress Minimum Maximum Stress Max Weathering Corrosion Rate Corrosion Corrosion Rate Corrosion Corrosion Corrosion Number of
Beam Range Stress Stress
Pmin Pmax Maximum
Range Stress
Strain
Time Bottom Flange Penetration 
Top Flange Penetration Rate Web Penetration Cycles to cor
(Newton) (Newton) Strain Range Range
(MPa) (MPa) (MPa) (MPa) (MPa) (Months) (mm/year) Bottom (mm) (mm/year) Top (mm) (mm/year) Web (mm) Failure
Group 1: Boldly Exposed Beams Tested in Air
B3 248 7 255 2633.13 95921.27 0.001240 250.40 257.47 0.001252 59.00 0.0100 0.0492 0.0058 0.0285 0.0100 0.0492 1750000
B4 248 7 255 2633.13 95921.27 0.001240 250.41 257.47 0.001252 59.00 0.0100 0.0492 0.0058 0.0285 0.0100 0.0492 4042600
B5 248 7 255 2633.13 95921.27 0.001240 250.41 257.47 0.001252 59.00 0.0100 0.0492 0.0058 0.0285 0.0100 0.0492 4424600
B6 248 7 255 2633.13 95921.27 0.001240 250.41 257.47 0.001252 59.00 0.0100 0.0492 0.0058 0.0285 0.0100 0.0492 1734200
Group 2: Boldly Exposed Beams tested in moist saltwater environment
B11 165 7 172 2633.13 64699.84 0.000825 167.71 174.83 0.000839 66.00 0.0170 0.0935 0.0110 0.0605 0.0091 0.0501 1595500
B12 165 7 172 2633.13 64699.84 0.000825 167.71 174.83 0.000839 66.00 0.0170 0.0935 0.0110 0.0605 0.0091 0.0501 1576600
B8 214 7 221 2633.13 83131.77 0.001070 216.40 223.48 0.001082 63.00 0.0120 0.0630 0.0120 0.0630 0.0054 0.0284 1036000
B10 214 7 221 2633.13 83131.77 0.001070 216.40 223.48 0.001082 63.00 0.0120 0.0630 0.0120 0.0630 0.0054 0.0284 940500
B1 276 7 283 2633.13 106453.80 0.001380 279.09 286.17 0.001396 63.00 0.0120 0.0630 0.0120 0.0630 0.0054 0.0284 386400
B2 276 7 283 2633.13 106453.80 0.001380 279.09 286.17 0.001396 63.00 0.0120 0.0630 0.0120 0.0630 0.0054 0.0284 456700
B9 276 7 283 2633.13 106453.80 0.001380 279.09 286.17 0.001396 63.00 0.0120 0.0630 0.0120 0.0630 0.0054 0.0284 435600
Group 3: Sheltered Beams tested in moist saltwater environment
B15S 47 7 54 2633.13 20312.74 0.000235 66.56 76.48 0.000333 70.00 0.2670 1.5575 0.0410 0.2392 0.2310 1.3475 6943200
B7S 69 7 76 2633.13 28588.30 0.000345 91.43 100.70 0.000457 65.00 0.2260 1.2242 0.0560 0.3033 0.2440 1.3217 4952100
B8S 69 7 76 2633.13 28588.30 0.000345 99.99 110.13 0.000500 66.00 0.2840 1.5620 0.0760 0.4180 0.2840 1.5620 1884600
B9S 69 7 76 2633.13 28588.30 0.000345 99.99 110.13 0.000500 66.00 0.2840 1.5620 0.0760 0.4180 0.2840 1.5620 1551300
B10S 110 7 117 2633.13 44010.93 0.000550 169.20 179.97 0.000846 71.00 0.2840 1.6803 0.0530 0.3136 0.3280 1.9407 468900
B11S 110 7 117 2633.13 44010.93 0.000550 169.20 179.97 0.000846 71.00 0.2840 1.6803 0.0530 0.3136 0.3280 1.9407 497400
B13S 110 7 117 2633.13 44010.93 0.000550 169.20 179.97 0.000846 71.00 0.2840 1.6803 0.0530 0.3136 0.3280 1.9407 420000
B5S 147 7 154 2633.13 57928.92 0.000735 194.78 204.06 0.000974 65.00 0.2260 1.2242 0.0560 0.3033 0.2440 1.3217 132900
B4S 152 7 159 2633.13 59809.73 0.000760 201.41 210.68 0.001007 65.00 0.2260 1.2242 0.0560 0.3033 0.2440 1.3217 201800
B6S 152 7 159 2633.13 59809.73 0.000760 201.41 210.68 0.001007 65.00 0.2260 1.2242 0.0560 0.3033 0.2440 1.3217 105500
B1S 208 7 215 2633.13 80874.79 0.001040 275.61 284.89 0.001378 65.00 0.2260 1.2242 0.0560 0.3033 0.2440 1.3217 154400
B2S 214 7 221 2633.13 83131.77 0.001070 283.56 292.84 0.001418 65.00 0.2260 1.2242 0.0560 0.3033 0.2440 1.3217 112400
B3S 214 7 221 2633.13 83131.77 0.001070 283.56 292.84 0.001418 65.00 0.2260 1.2242 0.0560 0.3033 0.2440 1.3217 16200
144
5.3.4 Fatigue Life Prediction
In order to include the mean stress effects and relate it to the material properties,
values of  corr   were plotted against the values of the maximum stress normalized to
the ultimate stress  max /  u as shown in Figure 5.13.
Trend
5
Group 1
Group 2
4
Group 3
 corr  
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
 max /  u
Figure 5.13 Plot of  corr   values versus  max /  u .
By curve fitting, an equation (5-7) was derived as follows:
  max 
 corr   5.77   5.35 (5-7)
 u 
By having the values of  corr defined for each environment, values of  corr   are
derived and thus the fatigue life for each specimen based on the modified SWT model
proposed in equation (5-4) can be determined. Values of all factors and predicted lives
are listed in Table 5.3.
145
Table 5.3 Derived Corrosion factors and estimated number of cycles to failure
Non-Weathered Section Weathered Section

Stress Minimum Maximum Number of Estimated number Estimated number Estimated number
Beam Range Stress Stress
Maximum Max Stress Strain
Cycles to of cycles of cycles of cycles
 corr    corr      max /  u  corr  corr
corr 
Strain Range (MPa) Range
(MPa) (MPa) (MPa) Failure (SWT) (Lower Bound) (Upper Bound) (Min) (Max) (Min) (Max)
Boldly exposed Beams Tested in Air
B3 248 7 255 0.001240 257.47 0.001252 1750000 19000000 1350000 2350000 1.11 1.39 2.775 0.45 0.4 0.5
B4 248 7 255 0.001240 257.47 0.001252 4042600 19000000 1350000 2350000 1.11 1.39 2.775 0.45 0.4 0.5
B5 248 7 255 0.001240 257.47 0.001252 4424600 19000000 1350000 2350000 1.11 1.39 2.775 0.45 0.4 0.5
B6 248 7 255 0.001240 257.47 0.001252 1734200 19000000 1350000 2350000 1.11 1.39 2.775 0.45 0.4 0.5
Boldly exposed Beams tested in moist saltwater environment
B11 165 7 172 0.000825 174.68 0.000838 1595500 82000000 1700000 3300000 2.52 2.88 3.603 0.30 0.7 0.8
B12 165 7 172 0.000825 174.68 0.000838 1576600 82000000 1700000 3300000 2.52 2.88 3.603 0.30 0.7 0.8
B8 214 7 221 0.001070 223.48 0.001082 1036000 58000000 650000 1000000 2.18 2.49 3.115 0.39 0.7 0.8
B10 214 7 221 0.001070 223.48 0.001082 940500 58000000 650000 1000000 2.18 2.49 3.115 0.39 0.7 0.8
B1 276 7 283 0.001380 286.17 0.001396 386400 6300000 250000 370000 1.74 1.99 2.488 0.50 0.7 0.8
B2 276 7 283 0.001380 286.17 0.001396 456700 6300000 250000 370000 1.74 1.99 2.488 0.50 0.7 0.8
B9 276 7 283 0.001380 286.17 0.001396 435600 6300000 250000 370000 1.74 1.99 2.488 0.50 0.7 0.8
Sheltered beams tested in moist saltwater environment
B15S 47 7 54 0.000235 76.48 0.000333 6943200 53000000 9300000 19000000 4.81 5.27 4.59 0.13 1.05 1.15
B7S 69 7 76 0.000345 100.70 0.000457 4952100 254000000 2650000 5000000 4.56 4.99 4.34 0.17 1.05 1.15
B8S 69 7 76 0.000345 110.13 0.000500 1884600 230000000 1800000 3300000 4.46 4.89 4.25 0.19 1.05 1.15
B9S 69 7 76 0.000345 110.13 0.000500 1551300 230000000 1800000 3300000 4.46 4.89 4.25 0.19 1.05 1.15
B10S 110 7 117 0.000550 179.97 0.000846 468900 77500000 305000 480000 3.73 4.08 3.55 0.31 1.05 1.15
B11S 110 7 117 0.000550 179.97 0.000846 497400 77500000 305000 480000 3.73 4.08 3.55 0.31 1.05 1.15
B13S 110 7 117 0.000550 179.97 0.000846 420000 77500000 305000 480000 3.73 4.08 3.55 0.31 1.05 1.15
B5S 147 7 154 0.000735 204.06 0.000974 132900 115500000 200000 300000 3.47 3.81 3.31 0.35 1.05 1.15
B4S 152 7 159 0.000760 210.68 0.001007 201800 93000000 190000 270000 3.41 3.73 3.24 0.37 1.05 1.15
B6S 152 7 159 0.000760 210.68 0.001007 105500 93000000 190000 270000 3.41 3.73 3.24 0.37 1.05 1.15
B1S 208 7 215 0.001040 284.89 0.001378 154400 6800000 90000 120000 2.63 2.88 2.50 0.49 1.05 1.15
B2S 214 7 221 0.001070 292.84 0.001418 112400 5150000 90000 110000 2.54 2.78 2.42 0.51 1.05 1.15
B3S 214 7 221 0.001070 292.84 0.001418 16200 5150000 90000 110000 2.54 2.78 2.42 0.51 1.05 1.15
146
Table 5.3 lists the predicted number of cycles predicted by the original SWT
model without considering corrosion. The introduction of the factor   was useful in
calibrating the model as it was clear that the results were stress level dependant. By trial
and error, the curve fitting in Figure 5.13 was adjusted to be upper bound to most of the
sample points in order to give conservative fatigue life for most of the test specimens. As
the ratio  max /  u gets less, the sensitivity of the factor  corr   gets higher. Several
trends were tested, the simple linear trend proposed in equation (5-7) was chosen. Having
more experimental data points would help in further calibration of this proposed
equation.
5.3.5 Discussion of results
The results obtained for the expected life for each specimen are presented as a
range corresponding to  corr  0.05 for each testing environment. The range of ±0.05 was
arbitrarily chosen, and it proved to result in very good predictions of fatigue life, as will
be shown later. The ranges of values of the predicted number of cycles to failure are
plotted versus the experimentally determined values from Albrecht et al. (1994) in
Figure 5.14. Despite the valuable reported data and results regarding the tested twenty
four beams that were used to calibrate the numerical model, it is worth mentioning that
the experimental data is a bit scattered with different loadings and environmental
conditions and that there is still a need for more available data points to further calibrate
the model.
147
log (Nf )
7.0
6.0
Numerical Experimental
(CorrFLP) (Albrecht 1994)
5.0
4.0
B10S,
B3, B4,
B5, B6
B11,
B12
B8,
B10
B1, B2,
B9
B15S B7S
B8S,
B9S
B11S,
B4S,
B6S
B5S
B2S,
B3S
B1S Specimen
B13S
Group
Figure 5.14 Predicted lives using CorrFLP vs. experimentally observed by Albrecht et al. (1994)
148
5.3.5.1 Discussion of group 1 results
Beams in this group have the same values for stress ranges and maximum applied
stresses. There is already a considerable scatter in the results of the identical tested beams
in Group 1, with a standard deviation of about 1.45 million cycles which is 70% of the
minimum experimentally obtained results and 28% of the maximum experimental result.
Numerical simulations of the predicted fatigue life of this group could be visually
compared by the range bars plotted in Figure 5.14. From Figure 5.14 it can be seen that
the upper bound predicted value is close to the mean of the experimental bar, whereas the
lower bound expected value is less than (i.e. conservative than) the minimum
experimental result by 22%. From the previous observations, it could be concluded that
the predicted fatigue lives using the proposed model (in term of ranges) are in good
agreement with the available experimental results and tend to be on the conservative side.
Beams in this group have different values for stress ranges and maximum applied
stresses. Beams B8 and B10 had the closest mean predicted fatigue life as compared to
the mean experimental value, where the difference between the means of the predicted
and the experimental values was about 12%. The maximum predicted life is almost the
same as the mean of the experimental result, while the minimum predicted life was 31%
less than the minimum experimental result. Beams B1, B2 and B9 had the most
conservative life predictions as all the numerical results lied below the experimental with
a difference of 4% between the maximum expected result and the minimum experimental
value obtained. Beams B11 and B12 had the most non-conservative life predictions as all
149
the predicted results lied above the experimental with a difference of 8% between the
minimum predicted result and the maximum experimental value obtained.
Beams in this group have a very wide spectrum of different values for stress ranges
and maximum applied stresses. All the predicted lives in this group were on the
conservative side except for beams B15S and B5S. This highlights the need for more
experimental tests in order to further calibrate the proposed  corr   function. Specimens
B2S and B3S are important to mention as they have a relatively high stress range and the
highest ratio of  max /  u being equal to 0.51. There is a very big scatter in the
experimental results between these two identical specimens which indicates that
specimen B3S might have experienced premature failure. This is highly probable,
especially given that the predicted analytical range of fatigue life is so close to that of
B2S. In general, it could be concluded that the numerical predictions of fatigue life using
the proposed model correlates very well with the measured experimental fatigue life of
the twenty four tested beams.
5.3.5.4 Conclusions
As can be seen from Figure 5.14 and from the previous observations and taking
into consideration the complexity of the corrosion-fatigue phenomenon, there is a
relatively good agreement between the results obtained using the proposed model and the
values reported. The proposed model generally gives conservative results. Moreover,
there is a big scatter in the results of some of the experimental specimens, this shows the
efficiency of providing the results in ranges rather than providing a single definite value.
150
It can be noticed from the numerical results that the predicted range of fatigue
cycles increases as the stress level decreases, and vice versa. The effect of the maximum
applied stress was found to be of significant importance in this study and was taken into
account by the factor  corr   which is a function in the ratio  max /  u . The relation
between the ratio  max /  u and the factor  corr   was found to be inversely
proportional with a linear trend. Further experimental work is needed to calibrate the
proposed function and to check if the observed behaviour would be the same for other
types of steel.
Moreover, by comparing to experimental results available which are already in a
corroded state (reduced sections), it can be said that it is not sufficient to model corrosion
just by geometrically reducing the section, as this would not provide a numerically
accurate estimation for the fatigue life as is clear from the predicted fatigue life
predictions using the original SWT model listed in Table 5.3. Hence, it is important to
implement the newly proposed strain-life model to account for the effect of corrosion and
the effect of maximum applied stress.
151
5.3.6 Standardizing a procedure for determining new material constants
More future experimental data is needed to further calibrate the proposed model.
For this purpose, the author proposes a standardized experimental methodology for
obtaining the proposed corrosion constants for any material.
1. Perform several strain-controlled axially loaded fatigue endurance tests on
standard specimens according to the ASTM E606 - 04e1 Standard Practice for
Strain-Controlled Fatigue Testing in a 3.5% NaCL solution which will be
assumed to correspond to  corr  1 .
2. By curve fitting the obtained results with a confidence level of 95% (as a lower
bound), values for b' and c' can be graphically obtained.
3. Having the target values of b' and c', and using the value of   from equation
(5-7), the values of  b ,  c can be obtained and used as material constants.
By having these factors for each material, the fatigue life of different materials in
corrosive environment can be predicted using the proposed modified SWT model.
5.3.7 Summary
A new strain-life model has been proposed as a modification to the Smith Watson
Topper model. The new model takes into account the effect of the corrosivity of the
environment. The newly proposed model uses several proposed constants that depend on
the material and the corrosivity of the environment. The proposed model is validated with
available experimental work. The results obtained for the expected life for each specimen
are presented as a range corresponding to  corr  0.05 for each testing environment to
152
give a range for the expected fatigue life instead of providing one value for the expected
life. Results obtained from the newly proposed model were of good agreement with the
experimental work available. The results tend to be more conservative than the
experimental values for most of the specimens. A standardized method to obtain the
proposed material parameters has been proposed. It is worthy to mention that all the
available experimental work for corrosion in literature is done on already corroded
members. Thus, the derived material properties represent the most severe case which is
conservative for design purposes.
5.4 Guidelines for developing corrosion-fatigue design charts
5.4.1 Introduction
One of the benefits of CorrFLP is the ability to perform a real-time simulation for a
certain detail using real loads along with environmental effects such as corrosion.
CorrFLP uses the strain-based fatigue analysis which is more accurate than the stress
based method commonly used in developing the S-N curves in several codes of practice.
By performing the analysis on a given detail, the expected fatigue life can be predicted
under any realistic loading history (not limited to constant amplitude loading). This could
facilitate the generation of series of S-N curves for several structural details for design
purpose. By simulating several stress ranges, corresponding number of cycles to failure
can be determined and plotted to create an S-N curve for a given detail. In this section,
several S-N curves for different structural details will be created under different corrosive
environments.
153
This section is not intended to provide any quantitative measures for the fatigue
life of the demonstrated details, as the results are material and detail dependant.
Moreover, in order to be able to confidently assign a fatigue category to a certain detail,
several parameters still have to be varied experimentally, which is beyond the scope of
this study. This section provides qualitative guidelines on how to use the developed
numerical tool to generate detail-specific S-N corrosion-fatigue curves. Appendix C
provides details on how the S-N curves can be statistically constructed.
5.4.2 Method for generating environmental S-N curves for various riveted details
In the following sections, several S-N curves for three commonly used structural
steel riveted details will be qualitatively constructed for different levels of corrosive
environments. This section is intended to demonstrate the capabilities of such a tool for
classifying any connection using the finite element method existing in different corrosive
environments. It provides systematic procedures for carrying out parametric study that
can be conducted for several commonly used details that would lead to the development
of environmentally dependant S-N curves to be implemented in code provisions.
Different corrosion rates for different metals could be obtained from Albrecht et al.
(1989). Charts can be generated for a wide spectrum of values for the factor  corr . The
designer can interpolate to get other intermediate values. Figure 5.15 shows a qualitative
example of the expected S-N curves to be generated for a sample riveted detail in
different environmental corrosivity values corresponding to values for the factor
 corr = 0.0, 0.5 and 1.0.
154
Figure 5.15 Sample S-N curves for a studied detail for several corrosion rates
5.4.3 Splice Plates
In this section, the connection studied in section 4.5.2 will be used. This detail is
chosen because it is commonly used as splices in the flanges of bridge girders and in truss
bridge members. The same FE model used in section 4.5.2 can be used here. Three rivet
diameters can be chosen; for example 16mm, 20mm, and 25mm to represent different
bearing ratios (ratio of bearing stress in the bolt to the average tensile stress in the plate).
Three thicknesses for the outer plates (more critical) can be used; for example 12mm,
16mm and 20mm. Figure 5.16 shows a schematic drawing for the parametric FE model.
It is important to note that this model is not a generalization for all configurations of
similar connections, but acts as an example to show the methodology that can be adopted
to generate ensemble of S-N curves for a given detail.
155
Figure 5.16 The finite element model for “Splice Plate” series
The stress applied loads are chosen so that the stress ranges calculated at the net
section (passing through the rivets) could have the values ranging from 0.2 of the yield
stress to 0.8 of the yield stress. The stress ratio R (ratio of low stress to high stress) can be
included as a factor, but it will drastically increase the required test samples, thus the
stress ratio R is recommended to be 0.2, which is a practical value used as a lower bound
for many fatigue tests. Figure 5.17 shows a schematic drawing illustrating the
dimensional variables chosen for this study. The study can be repeated for several types
of commonly used steels as their behaviour in corrosive environment could be
significantly different.

t out
t mid
t out
Figure 5.17 Schematic drawing for the test variables
156
5.4.4 Angles connected to Gusset Plates
This detail is chosen because it is commonly used in the flanges of bridge girders
and as splices in truss members. Three rivet diameters can be used; for example 16mm,
20mm, and 25mm to represent different bearing ratios (ratio of bearing stress in the bolt
to the average tensile stress in the plate). Three thicknesses for the outer plates (more
critical) can be used; 12mm, 16mm and 20mm. Figure 5.18 shows a schematic for the
parametric FE model.
Figure 5.18 The finite element model for the “Angles connected to Gusset Plates” series
Figure 5.19 shows a schematic drawing illustrating the dimensional variables
chosen for this study. The stress applied loads are chosen so that the stress ranges
calculated at the net section (passing through the rivets) could have the values ranging
from 0.2 of the yield stress to 0.8 of the yield stress. The stress ratio R (ratio of low stress
to high stress) can be included as a factor, but it will drastically increase the required test
samples, thus the stress ratio R is recommended to be 0.2, which is a practical value used
as a lower bound in many fatigue tests.
157
t gusset / 2
2 3 3 2 10 
L gusset
L angle

t angle
L angle
5.4.5 Angles in built-up Plate Girders
In this section, the connection studied in section 4.5.3 is used for the generation of
S-N curves. The same FE modeling technique used in section 4.5.3 can be used here.
Figure 5.20 shows a schematic for the parametric FE model.
Figure 5.20 The finite element model for the “Built-up Beam” series
158
Figure 5.21 shows a schematic drawing illustrating the dimensional variables
chosen for this study. The stress applied loads are chosen so that the stress ranges
calculated at the net section (passing through the rivets) could have the values ranging
from 0.2 of the yield stress to 0.8 of the yield stress. The stress ratio R (ratio of low stress
to high stress) can be included as a factor, but it will drastically increase the required test
samples, thus the stress ratio R is recommended to be 0.2, which is a practical value used
as a lower bound for many fatigue tests. Practical values for the stress gradient ratio
(Smin/Smax) range from 0.7 to 0.9 depending on the overall height of the girder, thus an
average stress gradient ratio equal to 0.8 can be used. The plotted values of stresses
should be the flexural stresses calculated on the net section passing through the rivets.
t gusset / 2
2 3 3 2
Smin Smin
L gusset
L angle

t angle
Smax Smax
 t lower pl.
L angle
159
5.5 Future work
The proposed strain-life model is based on the hypothesis that the corrosion affects
both the elastic and plastic behaviour of the material, and that the proposed material
properties  b and  c are constant. Further research should be done on monitoring and
quantifying this behaviour for different materials. The author recommends future work
for accurately correlating the environmental corrosivity intensity factor  corr could be
correlated to the categories provided by the ISO-9223 (1992) and the ISO-9224 (1992) to
standardize this factor. The author recommends following the proposed qualitative
guidelines in order to create corrosion fatigue design charts for several riveted connection
details.
160
6 CHAPTER 6
CONTRIBUTIONS, CONCLUSIONS AND RECOMMENDATIONS
6.1 Summary and Contributions of this research
The main objective of this study is to overcome the current limitations in
numerical modeling tools used for predicting the fatigue life of structural steel
components by developing a tool that takes into account the effect of corrosion, which is
an influential parameter on the fatigue life of railway or highway steel bridges. In
general, structural engineers could use either the stress-life approach or the strain-life
approach for the fatigue life prediction. This study contributes to both approaches.
As for the stress-life approach, a new damage accumulation model called the
VTLC (for Virtual Target Life Curves) is proposed in Chapter 3. This model can take
into account the effect of the overloads and can be calibrated to the material mechanical
characteristics using a proposed material constant. The proposed method is validated
using experimental work results from literature (which includes results of over 700 steel
wire specimens). A case study on assessing the fatigue life of a bridge member was
conducted to demonstrate the practical application of the proposed method.
Regarding the strain-life approach, a numerical tool is developed in Chapter 4 to
predict the fatigue life of any riveted or bolted component. The numerical tool (CorrFLP)
utilizes the finite element method and can be used as an add-on to many well-known
commercial FEM packages to add fatigue features to them. CorrFLP uses most of the
classical strain-life models in literature and can predict the fatigue life using the hot-spot
161
approach. To model the probabilistic nature of fatigue, a new approach is introduced to
make use of the theory of critical distances in order to obtain a range for the predicted
fatigue life instead of obtaining a single deterministic value as per the approach of other
current available tools. Three case studies were modeled to calibrate the numerical tool
and to validate the proposed method.
As an extension to the work done for fatigue life prediction, a new strain-life
model based on the SWT (Smith-Watson-Topper method of uniaxial fatigue damage
calculation) model that includes the effect of corrosion on the fatigue life prediction is
proposed in Chapter 5. The newly proposed model uses several proposed constants that
depend on the material and the corrosivity of the environment. The proposed model was
validated with available experimental work. As it is common to have a scatter in the
results of some of the fatigue tested specimens, it is only practical to provide the results
of the analytical simulations in ranges (as has been introduced in this thesis) rather than
providing a single definite value for the predicted fatigue life time. A standardized
method to obtain the proposed material parameters is also provided.
One of the practical benefits of this study is the possibility of generating
environmentally dependant S-N curves that could be adopted by design codes concerned
with the fatigue assessment of riveted steel railway and highway bridges. The last section
of Chapter 5 provides a systematic methodology for developing such charts following a
more comprehensive prospective parametric study.
162
6.2 Features of the developed numerical tool:
6.2.1 CorrFLP Applications:
CorrFLP was developed to serve several purposes and overcome many common
limitations in available fatigue simulation programs. The numerical tool has several
usages:
1. Fatigue life prediction of new structural elements to be designed; that is by
simulating the expected loading, corrosion and temperature histories along with
applying appropriate damage models.
2. Fatigue life prediction of existing structural elements including the effect of
corrosion under the current loading.
3. Evaluation and rating of structural elements using different future loading scenarios.
This will be achieved by creating a model simulating the past loading history
including the accumulated effect of corrosion, the model can be simulated from this
state using different loading scenarios and predict the fatigue life for each one.
4. Development of S-N curves for a given detail by running several analyses at
different stress range levels including the environmental effects.
6.2.2 CorrFLP Advantages:
1. Quickly simulates models with complex loading patterns that cannot be easily
simulated experimentally.
2. Includes the effects of corrosion which is difficult, time consuming and expensive to
prepare experimentally.
163
3. Has the smart meshing feature to capture high stress gradients.
4. Acts as an add-on that integrates with different finite element packages.
5. Can be extended to implement the combined effect of temperature and fatigue (not in
the scope of this study).
6.2.3 CorrFLP Current Limitations:
Although the numerical approaches, in general, are appealing -as they are time and
cost effective compared to experimental work- they have several limitations that need to
be considered. The following are the current limitations of CorrFLP:
1. Similar to other numerical tools, use of the finite element method for fatigue life
prediction, requires a good knowledge of finite element modeling. The accuracy of
the output highly depends on the accuracy of the finite element model, the input
loads and boundary conditions.
2. Does not support creep and thermal fatigue (not in the scope of this study).
3. Does not support fretting fatigue (not in the scope of this study).
4. Does not support energy methods in fatigue life prediction.
5. Does not support crack propagation analysis (not in the scope of this study).
6. Damage models are not general to all materials or even all kinds of steels, so they
need to be when chosen based on references. The program gives some guidelines for
the user, but it is recommended to have a strong background about the used material
fatigue properties.
7. Some of the damage models in literature need the user to know some parameters that
have to be calibrated with experimental work.
164
6.3 Significance of this Research
In this research, a numerical tool that predicts the fatigue life of deteriorating
corroded steel members or connections using the finite element method is developed and
coded. The advantage of the developed numerical tool is its effectiveness in the
evaluation of existing structural details and predicting their fatigue lives, simulating the
effect of corrosion and the possibility of applying the actual complex loading history.
This tool is a significant addition to the scientific resources in the area of fatigue;
where researchers will be able to use it to examine the influence of different reported
fatigue damage models on various structural details and investigate the interaction with
other factors such as corrosivity of the environment and the temperature (which is not
within the scope of this research). Moreover, the researcher can add user-defined rules for
fatigue-corrosion-temperature interaction and assess their applicability. The author
implemented several strain-life fatigue damage models available in the literature. The
tool includes several corrosion damage functions and corrosion rate histories based on
surveys from literature. Moreover, in this research, a new strain-life corrosion-fatigue life
prediction model is proposed and implemented in the program where it showed good
predictions of experimental test results.
This tool is seen to be practical for engineers in industry; where it could be used
for rating existing bridges under current loading and under different expected loading
scenarios in the future, so that the designers could propose different rehabilitation
scenarios and get an estimation of the expected corresponding fatigue life and the current
damage level. Moreover, this tool is programmed using a powerful object oriented design
165
using the C# language to facilitate extensions and adding of new models and rules for
interaction between input parameters.
6.4 Conclusions
1. The proposed VTLC method captures different ratios of low to high (L-H) loading
patterns with a relatively small error, and showed better predictions than the non-
conservative results obtained using Miner‟s LDR rule.
2. Miner‟s LDR rule has good predictions for loading patterns with block stress ratio
rb = S2/S1 which is very close to constant amplitude loading (close to unity). [Where
S2 is the block‟s low stress level and S1 is the block‟s high stress level, respectively]
3. Although the VTLC method takes a linear (thus a more simple) trend in modeling
fatigue damage, it proved to result in very good fatigue life predictions (comparable
to those obtained by Corten‟s rule) just by taking into account the nonlinear effects
of overloading.
4. Analysing the experimental results of fatigue lives of different riveted connections
details showed that they have scattered fatigue lives at the same stress levels.
Therefore, riveted steel connections need to be further classified into subcategories
as classifying them into category D is conservative in many cases, as was the case of
the analysed bridge in Croatia.
5. Using the theory of critical distances along with a ±15% variation in the calculated
critical length proved to be a good approach to get a range for the expected fatigue
life. This assumption is compatible with the probabilistic nature of fatigue, proven by
the scatter of the test results of replica girders under the same test conditions.
166
6. As CorrFLP uses the concept of physical propagation of damage to a certain portion
of the critical length L, thus in order to have relatively accurate fatigue life estimates,
the critical surface elements should be meshed to have elements of sizes close to 0.01
mm. This is automatically done in CorrFLP using the Parent-Child element sub-
meshing.
7. The reduced integration elements C3DR of ABAQUS give more numerically
realistic fatigue life predictions than fully integrated solid elements C3D.
8. The augmented Lagrange method is recommended for modeling the interaction
surfaces between steel components. Modeling the surfaces with no friction taken into
consideration is preferred as this gives a more conservative life estimate.
9. The newly proposed strain-life corrosion fatigue model along with using  corr  0.05
has very good agreement with reported experimental results of twenty-four beams
tested under various fatigue and weathering conditions.
10. The effect of the maximum applied stress was found to be of significant importance
in this study and was taken into account by the factor  corr   which is a function
in the ratio  max /  u . The relation between the ratio  max /  u and the factor
 corr   was found to be inversely proportional with a linear trend.
11. Analysing the experimental results available for fatigue lives of corroded beams, it
was found that modeling corrosion just by geometrically reducing the steel section is
not reliable. It is important to implement the newly proposed strain-life model to
account for the effect of corrosion and different levels of maximum applied stresses.
167
6.5 Recommendations for future work
1. The proposed overloading factor in the proposed stress-life VTLC model needs to be
investigated for other types of steel. It can also be extended to account for the crack
closure phenomenon.
2. An experimental research program should be carried out to further investigate the
proposed standardized procedure for determining the corrosion material constants for
several types of steel.
3. In this study, the proposed material-environment properties  b and  c where assumed
to be constant. Provided enough experimental data becomes available, a sensitivity
analysis is required to determine how these two factors may vary based on the
material‟s elastic and plastic properties.
4. The environmental corrosivity intensity factor  corr should be correlated to the
categories provided by the ISO-9223 (1992) and the ISO-9224 (1992) to standardize
this factor.
5. Further experimental work is needed to calibrate the proposed function for  corr   to
check if the behaviour interpreted in the current research would be applicable to other
types of steel.
6. The effect of pitting on the acceleration of the process of fatigue life deterioration
should be further investigated.
7. The guidelines provided for generation of environmentally dependant S-N curves
should be implemented to create more precise categories for different riveted
connections and assemblies.
168
7 REFERENCES
AASHTO AASHTO LRFD BRIDGE DESIGN SPECIFICATIONS (SI). Washington,

D.C. : American Association of State Highway and Transportation Officials, 2007.
AASHTO Manual for Maintenance Inspection of Bridges (1983, Interim 1989)
Washington, D.C. : American Association of State Highway and Transportation
Officials, 1989.
Abe H. Fatigue Strength of Corroded Steel Plates from Old Railway Bridges. IABSE
Symposium. Lisbon, 1989: 205-210.
Adamson D.E. and Kulak G.L. Fatigue Test of Riveted Bridge Girder. Structural
Engineering Report No. 210. Department of Civil Engineering, University of
Alberta, 1995.
Albrecht P. et al. Guidelines for the use of weathing steel in bridges.Washington :
National Cooperative Highway Research Program (NCHRP - Report 314), 1989.
Albrecht P. and ShabShab C. 1994. Fatigue Strength of Weathered Rolled Beam Made
of A588 Steel. Journal of Materials in Civil Engineering: Vol. 6, Issue 3.
Albrecht P. Fatigue Design Stresses for weathering steel structures. Corrosion Fatigue:
Mechanics, Mettallurgy, Electrochemistry, and Engineering (Book auth. Crooker
T., W., Leis B., N. and Eds.) ASTM STP 801, 1983.
Al-Emrani M. Fatigue in Riveted Railway Bridges. Göteborg, Sweden : Steel and
Timber Structures, Chalmers University of Technology, 2002. Doctoral Thesis.
Ansys 2011.
http://www.ansys.com/Products/Simulation+Technology/Structural+Mechanics/AN
SYS+nCode+DesignLife. (accessed 2011)
AREA "Steel Structures," Manual for Railway Engineering, Chapter 15. American
Railway Engineering Association, 1987.
AREMA Manual for Railway Engineering. American Railway Engineering and
Maintenance-of-Way Association, 2011.
Argon A. S. Effect of Surface on Fatigue Crack initiation. Connecticut : Corrosion
Fatigue: Chemistry, Mechanics and Microstructure, NACE and AIME, 1971: 176-
210.
ASCE Infrastructure Report Card 2005. ASCE, 2005.
http://www.asce.org/reportcard/2005/page.cfm?id=22. (accessed 2011)
Baker K.A. and Kulak G.L. 1985. Fatigue of Riveted Connections. Canadian Journals
of Civil Engineering: 184-191.
Bannantine J. A. and Socie D. F. A variable Amplitude multiaxial fatigue life
prediction method. Fatigue under biaxial and multiaxial loading, Proc. Third
International Conference on Biaxial/Multiaxial Fatigue. Stuttgart, 1989.
169
Barsom J.M. and Rolfe S.T. MNL41-3RD : Fracture and Fatigue Control in Structures:
Applications of Fracture Mechanics, Third Edition,1999.
Baumel A. Jr. and Seeger T. Materials data for cyclic loading, suppliment 1. El Sevier
Science Publisher BV, 1990.
Bluhm J. I. A note on fatigue damage. Materials Research and Standards, 1962.
Boden Harald Approaches in Modeling the Impact of Air Pollution-Induced Material
Degradation,1989.
Brown M. W. and Miller K. J. 1973. A theory of fatigue under multiaxial strain
conditions. Proc. Inst. Mech. Eng.: Vol. 187: 745-755.
Brühwiler E., Smith I. F. C. and Hirt M. A. 1990. Fatigue and Fracture of Riveted
Bridge Members. Journal of Structural Engineering: Vol. 116, Issue 1: ISSN 0733-
9445/90/0001-0198.
Cerit M., Genel K. and Eksi S. 2009. Numerical investigation on stress concentration
of corrosion pit. Engineering Failure Analysis: Vol. 16, Issue 7: 2467-2472.
CHBDC CAN/CSA-S6-06 - Canadian Highway Bridge Design Code, 2006.
Chen Huating, Grondin Gilbert Y. and Driver Robert G. Fatigue Resistance of High
Performance Steel. Structural Engineering Report No. 258. University of Alberta,
2005.
Cochrane V. H. Rules for rivet hole deductions in tention members. Engineering News-
Record, 1922. - Vol. 89. 847-848.
Cojocaru D. and Karlsson A.M. 2008. An object-oriented approach for modeling and
simulation of crack growth in cyclically loaded structures. Advances in Engineering
Software: Vol. 39: 995–1009.
Corten H. T. and Dolan T. J. 1956. Cumulative Fatigue Damage. International
Conference on Fatigue of Metals, Institute of Mechanical Engineering. London:
235-246.
Cui Weicheng 2002. A state-of-the-art review on fatigue life prediction methods for
metal structures. Jounal of Marine Science and technology: Vol. 7: 43-56.
D. Cojocaru A.M. Karlsson 2006. A simple numerical method of cycle jumps for
cyclically loaded structures. International Journal of fatigue: Vol. 28: 1677-1689.
DeJong S. J. and Heffernan P. J., MacDougall, C. 2009. Periodic Overload Corrosion
Fatigue of MMFX and Stainless Reinforcing Steels. J. Mater. Civ. Eng.: Vol. 21,
Issue 1.
Dessault-Systems Abaqus 6.10 User Documentation, Dessault Systems, 2010.
DiBattista J.D. and Kulak G.L. Fatigue of Riveted Tension Members. Structural
Engineering report No. 211. University of Alberta, Department of Civil
Engineering, 1995.
170
Dowling N. E. A Discussion of Methods for Estimating Fatigue Life. Proceedings of the
SAE Fatigue Conference, Dearborn, Mich., Society of Automotive Engineers. -
Warrendale, PA, 1999: 161-174.
Dowling N. E. and Begley J. A. 1976. Fatigue Crack Growth during Gross plasticity and
the J-integral. Mechanics of Crack Growth, ASTM: Vol. STP 590: 82-103.
Downing S. D. and Socie D. F. 1982. Simplified Rainflow Cycle Counting Algorithms.
Int. J. Fatigue: Vol. 4, Issue 1: 31-40.
Doyle J. L., Wood G. R. and Bondurant D. P. 1990. Using Laser-based profilometry to
locate and measure Corrosion fatiguecracking in boiler tubes. Materials Evaluation:
556-560.
Drapper John Modern Metal Fatigue Analysis, 2008.
Du Ming Liang Study of fatigue Proces in Air and Corrosive Environments using
Optical Methods, Ph.D. Dissertation, New York, 1998.
Duquette D. 1998. Corrosion Fatigue Crack Initialtion Processes: A State-of-the-art
Review. Environment-Induced Cracking of metals: 45-54.
ECCS Recommendations for the Fatigue Design of Steel Structures. Brussels, Belgium :
European Convention for Constructional Steelwork, ECCS Technical Committee 6,
1985.
El Haddad M. H., Smith K. N. and Topper T. H. 1979. Fatigue crack propagation of
short cracks. Journal of Engineering Materials and Technology (Trans. ASME):
Vol. 101: 42-46.
Elber W. 1970. Fatigue Crack Closure under Cyclic Tension. Engineering Fracture
Mechanics: Vol. 2: 37-45.
Ellyin F. Fatigue Damage, Crack Growth and Life Prediction. New York : Chapman &
Hall, 1997.
El-Sisi A.A. Behaviour of Steel connections under fatigue loading, Ph.D. Thesis, 2009.
Engelstad Stephen P. and Haj-Ali Rami M. Cracked Hole Finite Element
Modeling. Georgia Institute of technology, Atlanta, 2001.
Eurocode3 prEN 1993-1-9: Eurocode 3 : Design of steel structures, Part 1.9 : Fatigue,
May 2003.
Everett R. A. 1992. Comparison of Fatigue Life Prediction Methodologies for
Rotorcraft. Journal of the American Helicopter Society: Vol. 37, Issue 2: 54-60.
Fatemi A. and Yang L. 1998. Cumulative fatigue damage and life prediction theories: a
survey of the state of the art for homogeneous materials. Int. J Fatigue: Vol. 20: 9-
34.
FemFat 2011.
http://www.femfat.com (accessed 2011)
171
Fe-Safe - Safe Technology, 2011.
http://www.safetechnology.com. (accessed 2011)
Fisher J. W. et al. Fatigue and Fracture Evaluation for rating riveted
bridges. Pennsylvania : NCHRP (National Cooperative Highway Research
Program), 1988.
Fisher J. W. et al. Fatigue and Fracture Evaluation for Rating Riveted Bridges.
Washington, D.C. : National Cooperative Highway Research Program Report 302,
Transportation Research Board, National Research Council, 1987.
Fisher J. W., Yen B. T. and Wang D. Fatigue of Bridge Structure -A Commentary and
Guide For Design, Evaluation and Investigation of Cracking. Bethlehem,
Pennsylvania : ATLSS Report No.89-02, Lehigh university, 1989.
Fisher J.W. Bridge Fatigue Guide. New York: American Institute of Steel Construction,
1977.
Fisher J.W., Yen B.T. and Wang D. 1990. Fatigue Strength of Riveted Bridges
Members. Journal of structural Engineering: Vol. 116, Issue 11: 2986-2981.
Forsberg B. Utmattningshållfasthet hos äldre konstruktionsstål med korrosionsskador.
Master thesis 1993:064E. Division of Steel Structures Luleå University of
Technology, 1993. ISSN: 0349-6023.
Freudenthal A. M. and Heller R. A. 1959. On Stress interaction in fatigue and
cumulative damage rule. Journal of Aero/Space Science: Vol. 26: 431-442.
Frost N. E., March K. J. and Pook L. P. Metal Fatigue. Clarendon Press, Oxford.,
1974.
Gangloff R. 1990. Corrosion Fatigue Crack Propagation in Metals. Environment-Induced
Cracking of Metals, NACE-10: 55-110.
Gatts R. R. 1961. ASME Journal of Basic Engineering: Vol. 83: 529-540.
Goodman J. Mechanics Applied to Engineering (Book Section). London : Longmans
Green, 1899.
Hancq D. Alfred, Walters Andrew J. and Beuth Jack L. 2000. Development of an
Object-Oriented Fatigue Tool. Engineering with Computers: Vol. 16: 131-144.
Hashin Z. and Rotem A., A. 1978. A cumulative damage theory of fatigue failure.
Materials Science and Engineering: Vol. 34, Issue 2: 147-160.
Helmerich R., Brandes K. and Herter J. Full Scale Laboratory Fatigue Tests on
Riveted Railway Bridges. Evaluation of Existing Steel and Composite Bridges,
IABSE Workshop. Lausanne, 1997: 191-200.
Henry D. L. 1955. A theory of fatigue damage accumulation in steel. Transactions of the
ASME: Vol. 77: 913-918.
Hunsche A. and Neumann P. 1988. Crack Nucleation Persistent
Siipbands. Philadelphia: ASTM STP 924: Vol. Basic Questions in Fatigue: I: 26-38.
172
Hwang W. and Han K. S. 1986. Cumulative Damage Models and Multi-stress Fatigue
Life Prediction. Journal of Composite Materials: Vol. 20, Issue 125.
ISO9223 Corrosion of metals and alloys- Classification of corrosivity of atmospheres.
ISO-9223 Corrosion of metals and alloys-Corrosivity of atmospheres-Classification.
Geneva, 1992.
ISO-9224 Corrosion of metals and alloys-Corrosivity of atmospheres-Guiding for the
corrosivity categories. Geneva, 1992.
Josi G., Grondin G. Y. and Kulak G. L. Fatigue of Bearing-Type Shear Splice.
University of Alberta, 1999. Structural Engineering Report No. 227.
Josi Georg, Grondin Gilbert Y. and Kulak Geoffrey L. Fatigue of Bearing-Type Shear
Splice. Structural Engineering Report No. 227. University of Alberta, 1999.
Kandil F. A., Brown M. W. and Miller K. J. 1982. Biaxial low cycle fatigue fracture of
316 stainless steel at elevated temperatures. London : The Metals Society: Vol.
280 :745-755.
Kawahara M. et al. 1988. Corrosion fatigue Crack initiation of carbon steels in
Seawater. ASTM STP 924: Vol. II: 145-163.
Koch G. H. et al. Corrosion cost and prevention strategies in the United States. McLean,
Va. : Report No. FHWA-RD-01-156, Office of Infrastructure Research and
Development, Federal Highway Administration, 2002.
Kramer I. R. 1974. A mechanism of fatigue failure. Metallurgical Transactions: Vol.
5: 1735-1742.
KTS Corrosion Doctors. Kingston Technical Software, 2010.
http://corrosion-doctors.org. (accessed 2011)
Kühn B. et al. Assessment of Existing Steel Structures - Recommendations for
Estimation of the Remaining Fatigue Life. Luxembourg : EUR 23252 EN.: Office
for Official Publications of the European Communities, 2008.
Kulak G. L. High Strength Bolting. CISC, 2005.
Kulicki J.,M. et al. Guidelines for Evaluating Corrosion Effects In Existing Steel
Bridges. National Cooperative Highway Research Program (NCHRP) Report 333,
1990.
Kunihiro T., Inove K. and Fukuda T. Atmospheric exposure study of weathering
steel. Tokyo, Japan : Research Lab. Report Br. 71-08, Ministry of Construction,
1972. (In Japanese).
Leipholz H.H. E. 1986. On the modified S-N curve for metal fatigue prediction and its
experimental verification: Vol. 23, Issue 3: 495-505.
Leis B. N., Gowda C. V. B. and Topper T. H. 1973. Cyclic Inelastic Deformation and
the Fatigue Notch Factor. Symposium on Cyclic Stress–Strain Behavior Analysis,
Experimentation, and Failure Prediction, Bal Harbour, Fla.. - Philadelphia: ASTM
STP 519: 133-150.
173
Li Z. X., Ko J. M. and Chan T. H. T. 2001. Modelling of load interaction and overload
effect on fatigue damage of steel bridges. Fatigue & Fracture of Engineering
Materials & Structures, (June): Vol. 24, Issue 6: 379–390.
Lounis Z. 2007. Aging highway bridges. Canadian Consulting Engineer: 30-34.
Mang F. and Bucak O. Application of the S-N line concept for the assessment of the
remaining fatigue life of old bridge structures. Bridge Management 2. Inspection,
maintenance, 1993. ISBN: 0 7277 1926 2.
Manson S. S. and Halford G. R. 1981. Practical implementation of double linear
damage rule and damage curve approach for treating cumulative fatigue damage.
International Jounal of Fracture: Vol. 17, Issue 2.
Manson S. S., Freche J. C. and Ensign S. R. 1967. Application of double linear damage
rule to cumulative fatigue. In Fatigue Crack Propagation: Vol. ASTM STP
415: 384-412.
Marco S. M. and Starkey W. L. 1954. A concept of fatigue damage. Transactions of the
ASME: Vol. 76: 627-632.
Miner M. A. 1945. Cumulative damage in Fatigue. Journal of Applied Mechanics: Vol.
67: A159-A164.
Morrow J. and Socie D. F. Evolution of Fatigue Crack Initiation Life Prediction
Methods. Materials, Experimentation and Design in Fatigue. Proceedings of Fatigue
'81, Warwick University, England, Sherratt, F., and Sturgeon, J.B. (Editors),
Westbury House. - Guildford, 1981. 3-21.
Moses E. Schilling, E. A., and Raju, K. S. Fatigue Evaluation Procedures for Steel
Bridges. Washington, D.C. : National Cooperative Highway Research Program
Report 299, Transportation Research Board, National Research Council, 1987.
MSC 2011.
http://www.mscsoftware.com/Products/CAE-Tools/MSC-Fatigue.aspx.
(accessed 2011)
NACE Atmospheric Corrosion 2011.
http://events.nace.org/library/corrosion/AtmCorros/iso9223.asp. (accessed 2011)
nCode, 2011.
http://www.ncode.com. (accessed 2011)
Neuber H. Theory of notch stresses: principles for exact calculation of strength with
reference to structural form and material. 2nd Edition. Berlin : Springer, 1958.
Novak S., R. Corrosion Fatigue crack initation behaviour of four structural steels.
Corrosion Fatigue: Mechanics, Metallurgy, Electrochemistry, and Engineering,
ASTMSTP 801, (Book auth. Croolcer T. W., Leis B., N. and Eds.). American
Society for Testing and Materials, 1983.
Out J.M.M., Fisher J.W. and Yen B.T. 1984. Fatigue Strength of Weathered and
Deteriorated Riveted Members. Transportation Research Record 950: Vol. 2: 10-20.
174
Paepegem W. Van and Degrieck J. 2002. Effects of Load Sequence and Block Loading
on the Fatigue Response of Fibre-reinforced Composites. Mechanics of Advanced
Materials and Structures: Vol. 9, Issue 1: 19-35.
Paepegem W. Van, Degrieck J. and De Baets P. 2001. Finite Element approach for
modelling fatigue damage in fibre-reinforced composite materials. Composites: Part
B: Vol. 32: 575-588.
Palmgren 1924. A Die Lebensdauer von Kugellagern. Verfid-mwsrrchinik. Berlin: Vol.
68: 339-341.
Papuga Jan Mapping of Fatigue Damages - Program Shell of FE-Calculation, Ph.D.
Thesis. Prague, 2005.
Paris P. C., Gomez M. P. and Anderson W. P. 1961. A Rational Analytic Theory of
Fatigue. The Trend in Engineering: Vol. 13: 9-14.
Peterson RE. Notch senstivity. Metal Fatigue (Book auth. Sines G. and Waisman JL).
New York : McGraw Hill, 1959.
Radaj D. and Sonsino C. M. Fatigue Assessment of Welded Joints by Local
Approaches. Cambridge, England : Abington Publishing, 1998.
Reemsnyder H. S. 1975. Fatigue Life Extension of Riveted Connections. Journal of the
Structural Division. Proceedings of the American Society of Civil Engineers: Vol.
101, No. ST12: 2591-2608.
Rice R. C., Leis B. N. and Nelson D. V. Fatigue Design Handbook, Second
Edition. Warrendale, PA : Society of Automotive Engineers, 1988.
Schutz W. 1996. A History of fatigue. Engineering fracture mechanics: Vol. 54: 263-
300.
Sehitoglu H. 1983. Fatigue Life Prediction of Notched Members Based on Local Strain
and Elastic-Plastic Fracture Mechanics Concepts. Engineering Fracture Mechanics.
Vol. 18, Issue 3: 609-621.
Smith K. N., Watson P. and Topper T. H. 1970. A stress-strain function for the fatigue
of metals. J. Mater: Vol. 5, Issue 4: 767-778.
Subramanyan S. 1976. A Cumulative Damage Rule Based on the Knee Point of the S-N
Curve. ASME Journal of Engineering Materials: Vol. 98, Issue 4: 316-321.
Taylor D. A Compendium of Fatigue Thresholds and Growth Rates. EMAS, 1985.
Taylor D. 2008. The theory of critical distances: Vol. 75: 1696-1705.
Trethewey K. R. and Chamberlain J. Corrosion for students of science and
engineering. New York : Wiley, 1988.
USACE(EM) US Army Corps of Engineers Engineer Manual Inspection, Evaluation,
and Repair of Hydraulic Steel Structures, 2001: 1110-2-6054.
Vernon W.H.J. 1935. A Laboratory Study of the Atmosphere Corrosion of Metals.
Trans.
175
Xiulin et al. 1996. Fatigue Performance of Old Bridge Steel and the Procedures For Life
Prediction With Given Survivability. Engineering Fracture Mechanics: Vol. 53,
Issue 2: 251-262.
Zhou et al. 1995. Examination of Fatigue Strength (Sr-N) Curves for Riveted Bridge
Members. Engineers Society of Western Pennsylvania. - Pittsburg : 12th
International Bridge Conference.
176
8 APPENDIX A
This section provides a sample C# code used for derivation of the VTLC
parameters using the data provided by Corten et al.
struct CaseStudy
{
public double TotalCyclesPerBlock;
public double Alpha;
public double S1;
public double S2;
public double Nactual;
public double BasquinExponent;
public double FatigueStrength;
public CaseStudy( double FirstStress,

double SecondStress,
double CyclesToFailure,
double Ratio,
double TotalCyclesInBlock,
double Basquin,
double Sf)
{
TotalCyclesPerBlock = TotalCyclesInBlock;
Alpha = Ratio;
S1 = FirstStress;
S2 = SecondStress;
Nactual = CyclesToFailure;
BasquinExponent = Basquin;
FatigueStrength = Sf;
public double n1
{
get { return Alpha * TotalCyclesPerBlock; }
}
public double n2
{
get { return (1 - Alpha) * TotalCyclesPerBlock; }
}
class Program
{
static void Main(string[] args)
{
// Create an array of case studies
CaseStudy[] c = new CaseStudy[31];
// 96000 .. 76000 (Series B, R = 0.79, Rr = 0.2) b = 0.28

c[0] = new CaseStudy(96000, 76000, 55348, 0.4, 10000, -0.13137, 369200);
c[1] = new CaseStudy(96000, 76000, 97499, 0.1, 10000, -0.13137, 369200);
c[2] = new CaseStudy(96000, 76000, 121423, 0.04, 10000, -0.13137, 369200);
c[3] = new CaseStudy(96000, 76000, 129509, 0.01, 10000, -0.13137, 369200);
177
// 96000 .. 66000 (Series A, R = 0.69, Rr = 0.3125) b = 0.10
c[4] = new CaseStudy(96000, 66000, 60187, 0.4, 10000, -0.136314, 381640);
c[5] = new CaseStudy(96000, 66000, 133567, 0.1, 10000, -0.136314, 381640);
c[6] = new CaseStudy(96000, 66000, 221055, 0.04, 10000, -0.136314, 381640);
c[7] = new CaseStudy(96000, 66000, 269650, 0.01, 10000, -0.136314, 381640);
c[8] = new CaseStudy(96000, 66000, 297440, 0.047, 10000, -0.136314, 381640);
// 96000 .. 46000 (Series B, R = 0.48, Rr = 0.52) b = 0.019

c[9] = new CaseStudy(96000, 46000, 83560, 0.4, 10000, -0.13137, 369200);
c[10] = new CaseStudy(96000, 46000, 293900, 0.1, 10000, -0.13137, 369200);
c[11] = new CaseStudy(96000, 46000, 599100, 0.04, 10000, -0.13137, 369200);
c[12] = new CaseStudy(96000, 46000, 974990, 0.01, 10000, -0.13137, 369200);
// 86000 .. 76000 (Series C, R = 0.88, Rr = 0.116) b = 0.146

c[13] = new CaseStudy(86000, 76000, 115378, 0.1 , 10000, -0.13137, 375000);
c[14] = new CaseStudy(86000, 76000, 172902, 0.04, 10000, -0.13137, 375000);
c[15] = new CaseStudy(86000, 76000, 183992, 0.01, 10000, -0.13137, 375000);
// 86000 .. 66000 (Series B, R = 0.77, Rr = 0.233) b = 0.106

c[16] = new CaseStudy(86000, 66000, 127174, 0.4, 10000, -0.13137, 369200);
c[17] = new CaseStudy(86000, 66000, 235288, 0.1, 10000, -0.13137, 369200);
c[18] = new CaseStudy(86000, 66000, 290670, 0.04, 10000, -0.13137, 369200);
c[19] = new CaseStudy(86000, 66000, 333273, 0.01, 10000, -0.13137, 369200);
// 86000 .. 56000 (Series B, R = 0.65, Rr = 0.35) b = 0.0096

c[20] = new CaseStudy(86000, 56000, 458142, 0.1, 10000, -0.13137, 369200);
c[21] = new CaseStudy(86000, 56000, 648485, 0.04, 10000, -0.13137, 369200);
c[22] = new CaseStudy(86000, 56000, 864172, 0.013, 10000, -0.13137, 369200);
// 86000 .. 46000 (Series B, R = 0.53, Rr = 0.47) b = 0.0001

c[23] = new CaseStudy(86000, 46000, 175227, 0.4, 10000, -0.13137, 369200);
c[24] = new CaseStudy(86000, 46000, 611646, 0.1, 10000, -0.13137, 369200);
c[25] = new CaseStudy(86000, 46000, 1122536, 0.04, 10000, -0.13137, 369200);
c[26] = new CaseStudy(86000, 46000, 1809256, 0.013, 10000, -0.13137, 369200);
// 76000 .. 66000 (Series A, R = 0.87, Rr = 0.132)

c[27] = new CaseStudy(76000, 66000, 220850, 0.40, 10000, -0.136314, 381640);
c[28] = new CaseStudy(76000, 66000, 282358, 0.10, 10000, -0.136314, 381640);
c[29] = new CaseStudy(76000, 66000, 338532, 0.04, 10000, -0.136314, 381640);
int Counter = 0;
bool AllSatisfied = true;
double MinSumError = 100000;
#region VTLC1
//for (double c1 = -100; c1 <= -4; c1 -= .5)

for (double c2 = 1.0; c2 <= 1; c2 += .5)
//for (double bi = 0.0001; bi <= 0.002; bi += 0.0001) // 0.0037
//for (double p = 0.02; p <= 0.1; p += 0.001) // 0.0037
for (double d = 100; d <= 100; d += 20) // 0.0037
for (double p = .3; p <= .3; p += .1) // 0.0037
for (double B = 2.7; B <= 2.7; B += 0.1) // 0.0037
for (double D = 05; D <= 05; D += 10) // 0.0037
{
double SumError = 0;
double MinError = 100, MaxError = 0;
double N = 0;
double c1 = 0 ; //, c2 = 2.6;

//double bi = 0; // 0.0016;
//double p = 0.04;
// double d = 460;
// double p = 1;
//double B = 2.7; with c1 = a = 1 // gave best results
178
AllSatisfied = true;
for (int i = 0; i <= 29; i++)

{
double Nact = c[i].Nactual;
N = VTLC2(c1, c2, B, d, c[i], D, false, false, p);
double Error = (Math.Abs(N - c[i].Nactual)/c[i].Nactual)*

100;
if (Error > MaxError)

MaxError = Error;
if (Error < MinError)

MinError = Error;
Counter++;
}
//if (AllSatisfied) //&& MinSumError >= SumError)

{
MinSumError = SumError;
System.IO.StreamWriter fout =
new StreamWriter("C:\\Numbers.txt");
for (int i = 0; i <= 29; i++)

{
N = VTLC2(c1, c2, B, d, c[i], D, false, false, p);
double Error = (Math.Abs(N - c[i].Nactual)/c[i].Nactual)*

100;
fout.WriteLine(N.ToString("N0"));
Console.WriteLine("B = " + B
+ ", d = " + d
+ ", D% = " + D
+ ", p = " + p
+ ", Max Error = " + MaxError);
fout.Close();
}
}
#endregion
// Damage Using Linear Loss in Life with

// Constant Jump magnitude based on the S2/S1 ratio
static double VTLC1( double c1,
double c2,
double B,
double BjumpInitial,
CaseStudy Case,
double Dcritical,
bool ConservativeOnly,
bool PrintError,
double power)
{
// Some variables
int n1, n2, Nact;
double S1, S2, Alpha;
179
double b, Sf;
double Nf1, Nf2;
double Increment = 100; // Cycles
double TotalCycles = 0;
// Initialization
b = Case.BasquinExponent;
Sf = Case.FatigueStrength;
S1 = Case.S1;
S2 = Case.S2;
Alpha = Case.Alpha;
Nact = (int)Case.Nactual;
// Cycles per block

n1 = (int)(10000 * Alpha);
n2 = (int)(10000 * (1 - Alpha));
// Slopes
double m = -1 / b; // S-N Slope at failure
double mv0 = B*m; // Slope of Virgin (virtual Life Curve)
double mv = mv0; // Slope of S-N curve at any time (Very IMPORTANT)
double Dm0 = mv- m; // Reference change in slope
// Failure Lives
Nf1 = Math.Pow(S1 / Sf, -m);
// Virtual target Lives

double NVTL1 = Math.Pow((S1 / Sf), -mv);
double NVTL2 = Math.Pow((S2 / Sf), -mv);
// Loss in Life (dN)

double dN1 = NVTL1 - Nf1;
double dN2 = NVTL2 - Nf2;
double Level1Ratio = dN1 / Nf1;

double Level2Ratio = dN2 / Nf2;
int n = 0;
double Bjump = 0;
double FatigueLimit =0 ;
double DamagePercentage = 0;
bool Terminate = false;
// Stage 1
while (true)
{
n = 0;
Terminate = false;
// Stress Level 1
while (n < n1)
{
// Damage curves constants (ai , ri)
double ai = c1 * (S1 / Sf) + c2;
double ri = Dm0 / Math.Pow(dN1, ai);
int Partial = n1 - n;
if (Partial >= Increment)

{
n += (int)Increment;
// Extra initial Jump

if (n == Increment && S1 > S2 && TotalCycles != 0 )
180
{
double R = S2/S1;
power = 60*R*R - 65*R + 17.1; // equation 1

power = 100 * R * R - 120 * R + 36; // equation 2
double power2 = 18.43 * R - 11.8; // equation 3
power = 0;
double StressRatioModifier = 1-(Math.Pow(S2/ S1, power));
double dDmj = Dm0 / BjumpInitial * StressRatioModifier;
mv -= dDmj;
// Count additional cycles

TotalCycles += Increment;
}
else
{
n += Partial;
TotalCycles += Partial;
}
// Get the change in slope due to accumulated cycles

double dDm = ai * ri * Math.Pow(TotalCycles + Increment / 2, ai - 1)
*
Increment * Math.Pow(Level1Ratio, 1/ai);
// Get the updated slope

mv -= dDm;
// Calculate Virtual Life

NVTL1 = Math.Pow((S1 / Sf), -mv );
if (TotalCycles >= NVTL1)

{
Terminate = true;
break;
}
if (Terminate)
break;
n = 0;
// Stress Level 2
while (n < n2)
{
double ri = Dm0 / Math.Pow(dN2 , ai);

{

if (n == Increment && S2 > S1 && TotalCycles != 0)
{
//Bjump += BjumpInitial;
181
//Db += Bjump;
}

//Nacc = Math.Pow((Dm / ri), (1 / ai)) + Increment;
}
else
{
n += Partial;

{
//Bjump += BjumpInitial;
//Db += Bjump;
}

//Nacc = Math.Pow((Dm / ri), (1 / ai)) + Partial;
}
double dDm = ai * ri * Math.Pow(TotalCycles + Increment / 2, ai - 1)

*
Increment * Math.Pow(Level2Ratio, 1/ai);
// Get the change in slope due to accumulated cycles

//Dm = ri * Math.Pow(Nacc, ai); ;
// Get the updated slope

//mv = mv0 - Dm;
mv -= dDm;

NVTL2 = Math.Pow((S2 / Sf), -mv);
//RemainingLife = NVTL2 - Nacc;

{
Terminate = true;
break;
}
}
if (Terminate)
break;
// Console.WriteLine("Nacc2 = " + Nacc + ", NVTL = " + NVTL + ",

TotalCycles Until Now = " + TotalCycles);
//double PosError = (Nact - TotalCycles)/(double)Nact;
//if (PrintError && Error < ErrorPercent / 100)

// Console.WriteLine("At S1 = " + S1 + ", S2 = " + S2 + ", c1 = " + c1 +
", c2 = " + c2 + ", Binit = " + BjumpInitial + ", B = " + B + ", Nexp/Nact = (" +
(int)TotalCycles + "/" + (int)Nact + ")" + ", Error = " + (Error * 100).ToString("N2") +
"%");
return TotalCycles;
// Damage Using Linear Loss in Life with

// Constant Jump magnitude based on the S2/S1 ratio
182
static double VTLC2(double c1, double c2, double B, double BjumpInitial,
CaseStudy Case, double Dcritical, bool ConservativeOnly, bool PrintError, double power)
{
B = 2;
// Some variables
int n1, n2, Nact;
double b, Sf;
double Nf1, Nf2;
double Increment = 100; // Cycles
// Initialization
S1 = Case.S1;
S2 = Case.S2;
Alpha = Case.Alpha;
Alpha = 1;
// Cycles per block

n1 = (int)(10000 * Alpha);
n2 = (int)(10000 * (1 - Alpha));
// Slopes
double m = -1 / b; // S-N Slope at failure
double mv0 = B * m; // Slope of Virgin (virtual Life Curve)
double mv = mv0; // Slope of S-N curve at any time (Very IMPORTANT)
double Dm0 = mv - m; // Reference change in slope
// Failure Lives
// Virtual target Lives

double NVTLo1 = Math.Pow((S1 / Sf), -mv);
double NVTLo2 = Math.Pow((S2 / Sf), -mv);
double x = Math.Pow(Nf1, B);
double NVTL1 = NVTLo1;

double NVTL2 = NVTLo2;
// Loss in Life (dN)

double dNo1 = NVTLo1 - Nf1;
double dNo2 = NVTLo2 - Nf2;
double Level1Ratio = dNo1 / Nf1;

double Level2Ratio = dNo2 / Nf2;
int n = 0;
double Bjump = 0;
double FatigueLimit = 0;
double DamagePercentage = 0;
bool Terminate = false;
// Stage 1
while (true)
{
n = 0;
Terminate = false;
183
// Stress Level 1
while (n < n1)
{
double ri = Dm0 / Math.Pow(dNo1, ai);

{

{
power = 60*R*R - 65*R + 17.1; // equation 1

power = 100 * R * R - 120 * R + 36; // equation 2
double power2 = 18.43 * R - 11.8; // equation 3
power = 0;
double StressRatioModifier = 1-(Math.Pow(S2/ S1, power));
double dDmj = Dm0 / BjumpInitial * StressRatioModifier;
mv -= dDmj;

}
else
{
n += Partial;
}

NVTL1 = NVTL1 / Increment;

{
Terminate = true;
break;
}
if (Terminate)
break;
n = 0;
// Stress Level 2
while (n < n2)
{
double ri = Dm0 / Math.Pow(dNo2, ai);
184
{

}
else
{
n += Partial;

}

NVTL2 = NVTL2/Increment;

{
Terminate = true;
break;
}
}
if (Terminate)
break;
//double PosError = (Nact - TotalCycles)/(double)Nact;
//if (PrintError && Error < ErrorPercent / 100)

// Console.WriteLine("At S1 = " + S1 + ", S2 = " + S2 + ", c1 = " + c1 +
", c2 = " + c2 + ", Binit = " + BjumpInitial + ", B = " + B + ", Nexp/Nact = (" +
(int)TotalCycles + "/" + (int)Nact + ")" + ", Error = " + (Error * 100).ToString("N2") +
"%");
return TotalCycles;
static double CortenDolan( double Ra,

ref double CorrespondingA,
CaseStudy Case,
double ErrorPercent,
bool ConservativeOnly,
bool PrintError)
{
double B = 0.5;
int n1, n2, Nact;
double b, Sf;
double Nf1, Nf2;
double Error = 0;
// Initialization
S1 = Case.S1;
S2 = Case.S2;
185
Alpha = Case.Alpha;
Nf1 = Math.Pow(S1 / Sf, 1 / b);

Nf2 = Math.Pow(S2 / Sf, 1 / b);
double RHS, LHS;

double A = 0.01;
double MinError = 10000000;
do
{
RHS = Nact * Alpha + Ra * (1 - Alpha) * Math.Pow(Nact, A);
LHS = Nf1;
Error = (Math.Abs(LHS - RHS) / LHS) * 100;
if (MinError >= Error)

{
MinError = Error;
CorrespondingA = A;
// Calcualte CorrespondingN = Nact;
}
A += 0.0001;
} while (A <= 4);
return MinError;
}
186
9 APPENDIX B
This appendix lists the materials database available in CorrFLP:
 Stainless_Steel_30304_Cold_Rolled_BHN_327
 Stainless_Steel_30304_Hot_Rolled_BHN_160
 Stainless_Steel_30304_Su_650
 Stainless_Steel_30310_Hot_Rolled_BHN_145
 Steel_1005_HR_Sheet_Su_359
 Steel_1008_HR_Sheet_Su_363
 Steel_1015_Normalized_Su_414
 Steel_1018_BHN_120
 Steel_1020_HR_Plate_BHN_108
 Steel_1020_Su_455
 Steel_1040_Cold_Drawn_BHN_225
 Steel_1045_Normalized_BHN_153
 Steel_1045_Annealed_BHN_225
 Steel_1045_Q_T_BHN_277
 Steel_4142_As_Quenched_BHN_670
 Steel_4142_Q_T_BHN_450_b_0_076
 Steel_4142_Q_T_BHN_450_b_0_086
 Steel_4340_Hot_Rolled_BHN_243
 Steel_4340_Su_1172
 Steel_8630_Cast_BHN_254
 Steel_Maraging_18Ni_250_BHN_500
 Steel_300M_Su_1958
 Steel_4130_sheet_Su_806
 Steel_4130_sheet_Su_1241
 Steel_4340_bar_Su_862
 Steel_8620H_Case_Su_1600
 Steel_8620H_Core_Su_1510
 Steel_A27_Cast_BHN_135
 Steel_A36_HAZ_BHN_243
 Steel_A36_BHN_160
 Steel_A36_Su_540
187
 Steel_A514_HAZ_BHN_461
 Steel_A514_BHN_303
 Steel_A_517_Grade_F_BHN_256
 Steel_E110_WM_1P_Weld_Metal_BHN_362
 Steel_E110_WM_2P_Weld_Metal_BHN_310
 Steel_E60S_3_WM_1P_Weld_Metal_BHN_233
 Steel_E60S_3_WM_2P_Weld_Metal_BHN_201
 Steel_H1000_Su_1448
 Steel_H1000_bar_Su_1414
 Steel_H1050_sheet_Su_1386
 Steel_HY130_Su_1103
 Steel_IN787_BHN_188
 Steel_ManTen_Su_565
 Steel_RQC_100_Su_863
 Steel_TH1050_sheet_Su_1386
 Steel_TH1050_sheet_Su_1207
 Aluminum_1100_Su_110
 Aluminum_5454_Forged_Su_334
 Aluminum_2014_T6_Su_510
 Aluminum_2014_T6_Hand_Forged_Su_483
 Aluminum_2014_T6_Su_496_44
 Aluminum_5083_0_BHN
 Aluminum_5083_H12_Su_385
 Aluminum_5183_0_Weld_metal_BHN_92
 Aluminum_5456_H311_Su_400
 Aluminum_6061_T6_Forged_Su_389
 Aluminum_6061_T6_Sheet_Su_314
 Aluminum_7050_T7351X_Su_517
 Aluminum_7050_T7451_plate_Su_544
 Aluminum_7050_T7452_Su_537
 Aluminum_7050_T7452_Su_524
 Aluminum_7050_T7651X_Su_599
 Aluminum_7075_T6_Su_565_b_0_105
 Aluminum_7075_T6_Su_565_b_0_172
 Aluminum_7175_T73611_Su_524
 Aluminum_Cast_A356_T6_Su_252
188
10 APPENDIX C
The regression analysis performed in this section was adopted from the method
used by Josi et al. (1999). Regression analysis of the test data was used to derive S-N
curves that can be described using the following equation:
log C  log N f 
log     (C-1)
m m
For simplicity, the notation log  will be replaced by y, log N f  by x,
logC  / m by a and  1/ m by b to have the following form:
y  a  bx (C-2)
Using the least squares fitting method, the slope b and the intercept a of the best
fit line with “n” test results can be obtained from:
S xy
b (C-3)
S yy
where
  
n
S xy   xi  x yi  y
i 1
(C-4)
 
n
S yy   yi  y
2
i 1
and “a” can be obtained from :
a  y  bx (C-5)
To measure the goodness of fit of the regression model, the correlation coefficient
R2 can be used:
189
2
S xy
R 
2
(C-6)
S xx S yy
where
 
n
S xx   xi  x
2
(C-7)
i 1
A statistical comparison of mean values of two sets of test results ( y1 and y2 ) can
be carried out using Student‟s t test :
y1  y2
t
 y 
 y1   y 2  y 2  
2 2
n n  (C-8)
1
  1 2 
n1  1  n2  1  n1n2 
If a 95 percent level of confidence is defined, then t should not be greater than
2.776 in order to conclude there is no significant difference between the two variances.
190
View publication stats

Numerical Tool For Fatigue Life Prediction of Corroded Steel Riveted Connections Using Various Damage Models

Uploaded by

Copyright:

Available Formats

Numerical Tool For Fatigue Life Prediction of Corroded Steel Riveted Connections Using Various Damage Models

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Numerical Tool For Fatigue Life Prediction of Corroded Steel Riveted Connections Using Various Damage Models

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

NUMERICAL TOOL FOR FATIGUE LIFE PREDICTION OF CORRODED STEEL

Thesis · April 2012

Fatigue Life Prediction View project

The user has requested enhancement of the downloaded file.

Presented in Partial Fulfillment of the Requirements

© Ihab El Aghoury, 2012

This is to certify that the thesis prepared

and submitted in partial fulfillment of the requirements for the degree of

Signed by the final examining committee:

of steel bridges. Corrosion is considered to be an influential factor that significantly

this goal, the research is divided into three phases.

and without the effect of corrosion.

behaviour of the material used. A method is proposed to standardize a testing protocol to

twenty-four steel beams subjected to various fatigue and weathering conditions.

accounting for different levels of corrosion.

reviewing and correcting this thesis drafts are greatly appreciated.

is greatly appreciated. Encouraging awards from Concordia University are also

Montréal for their support for me during my study.

1.1 Background, problem definition, and need for research

monitoring and maintenance. One of the main components of an effective structural

Corrosion is considered to be an influential factor that significantly contributes to the

on constant amplitude loading, and being specific to a certain detail. Moreover,

simulating corrosion experimentally along with fatigue loading is challenging, expensive,

The following sections provide a brief description on the amount of deterioration

taking into account the effect of corrosion.

(Lounis, 2007). Existing bridges are experiencing accelerated deterioration due to

time of construction in a period of 39 years.

the metal, pre-existence of micro-cracks. Although fatigue life is affected by several

1.2.1 Methods of Determining Fatigue Life

Traditionally, fatigue life prediction is determined experimentally, but the

is generally neither timely nor statistically comprehensive for a majority of designers.

Modern structural design codes of practice usually refer to well-known damage

Another alternative approach is to predict the fatigue life of a specific structural

will be discussed in details in Chapter 2.

The advancements in the field of numerical simulations of structures using the

researchers to develop post-processing programs to estimate the fatigue life of structural

numerical tool that accounts for the effects of corrosion.

1.2.3 Fatigue of Metals

such as an existing material/manufacturing flaw or an abrupt geometrical discontinuity in

Fatigue failures typically involve minimal levels of plastic deformation.

cross-section occurs. Many examples of catastrophic failures caused by fatigue cracking

have a firm understanding of the phenomenon of fatigue. Many civil engineering

they must carry.

1.2.4 Corrosion of Metals

Corrosion is a process of degradation of a metal by an electrochemical reaction

known example of electrochemical corrosion which is commonly known as rust. This

1- General atmospheric corrosion: is defined as a slow uniform corrosive attack

2- Crevice corrosion: occurs in narrow openings between two contact surfaces,

failure of the paint system, which further promotes corrosion

point). If pitting occurs, it can rapidly accelerate the fatigue damage.

4- Galvanic corrosion: can occur in when steels with different electrochemical

systems are generated.

filaments emanating from one or more sources in random directions.

7- Erosion corrosion: is caused by removal of surface material by action of

associated with the metal removal.

8- Cavitation corrosion: is caused by cavitations associated with turbulent flow. It

rounded micro craters.

stresses (KTS, 2010).