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

Download as pdf or txt
Download as pdf or txt
You are on page 1of 210

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
3.6.1 Introduction 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
4.5.1.4 Fatigue Life Prediction 105
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
4.5.2.3 Finite Element Model 109
4.5.2.4 Interaction surfaces 109
4.5.2.5 Fatigue Life Prediction 111
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
4.5.3.5 Finite Element Model 120
4.5.3.6 Boundary Conditions and Loading 121
4.5.3.7 Modeling material properties 122
4.5.3.8 Fatigue Life Prediction 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
5.3.1 Introduction 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
5.3.5.2 Discussion of group 2 results 149
5.3.5.3 Discussion of group 3 results 150
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
5.4.1 Introduction 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 ‎5.19 Schematic drawing for the test variables ..................................................... 158
Figure ‎5.20 The finite element model for the “Built-up Beam” series ........................... 158
Figure ‎5.21 Schematic drawing for the test variables ..................................................... 159

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

Group 1 Group 2 Group 3

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
Passenger wagon 20.4 301000
S 38 92.6 235000
1947-1959 Cargo wagon 34.7 136000
Passenger wagon 20.4 119000
E 362 70.8 318000
2E 362 64.3 18000
1960-1980 Cargo wagon 45.9 880000
E 320 37.6 63000
Passenger wagon 20.4 192000
E 362 70.8 20000
1980-Present
Cargo Wagon 45.9 65000
(Yearly)
Passenger wagon 20.4 41000

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.

4.5.1.4 Fatigue Life Prediction

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

S2-Series (2 in. stagger)

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

S3-Series (3 in. stagger)

Figure 4.20 Test specimens of Josi et al. (1999)

108
4.5.2.3 Finite Element Model

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

4.5.2.5 Fatigue Life Prediction

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.

4.5.3.5 Finite Element Model

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.

4.5.3.8 Fatigue Life Prediction

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
Group 1 Group 2 Group 3

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.

5.3.5.2 Discussion of group 2 results

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.

5.3.5.3 Discussion of group 3 results

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

Figure 5.19 Schematic drawing for the test variables

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

Figure 5.21 Schematic drawing for the test variables

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);
Nf2 = Math.Pow(S2 / 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)
{
// Damage curves constants (ai , ri)
double ai = c1 * (S2 / Sf) + c2;
double ri = Dm0 / Math.Pow(dN2 , ai);

int Partial = n2 - n;

if (Partial >= Increment)


{
n += (int)Increment;

// Extra initial Jump


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

181
//Db += Bjump;
}

// Count additional cycles


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

if (n == Increment && S2 > S1 && TotalCycles != 0)


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

// Count additional cycles


TotalCycles += Partial;
//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;

// Calculate Virtual Life


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

if (TotalCycles >= NVTL2)


{
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 S1, S2, Alpha;
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;

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
Nf1 = Math.Pow(S1 / Sf, -m);
Nf2 = Math.Pow(S2 / Sf, -m);

// 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)
{
// Damage curves constants (ai , ri)
double ai = c1 * (S1 / Sf) + c2;
double ri = Dm0 / Math.Pow(dNo1, ai);

int Partial = n1 - n;

if (Partial >= Increment)


{
n += (int)Increment;

// Extra initial Jump


if (n == Increment && S1 > S2 && TotalCycles != 0)
{

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;
}

// Calculate Virtual Life


NVTL1 = NVTL1 / Increment;

if (TotalCycles >= NVTL1)


{
Terminate = true;
break;
}

if (Terminate)
break;

n = 0;

// Stress Level 2
while (n < n2)
{
// Damage curves constants (ai , ri)
double ai = c1 * (S2 / Sf) + c2;
double ri = Dm0 / Math.Pow(dNo2, ai);

int Partial = n2 - n;

if (Partial >= Increment)

184
{
n += (int)Increment;

// Count additional cycles


TotalCycles += Increment;
}
else
{
n += Partial;

// Count additional cycles


TotalCycles += Partial;
}

// Calculate Virtual Life


NVTL2 = NVTL2/Increment;

if (TotalCycles >= NVTL2)


{
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 S1, S2, Alpha;
double b, Sf;
double Nf1, Nf2;
double TotalCycles = 0;
double Error = 0;

// Initialization
b = Case.BasquinExponent;
Sf = Case.FatigueStrength;
S1 = Case.S1;
S2 = Case.S2;

185
Alpha = Case.Alpha;
Nact = (int)Case.Nactual;

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_BHN_120
 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_1045_Q_T_BHN_336
 Steel_1045_Q_T_BHN_390
 Steel_1045_Q_T_BHN_410
 Steel_1045_Q_T_BHN_500
 Steel_1045_Q_T_BHN_563
 Steel_1045_Q_T_BHN_595
 Steel_4130_BHN_259
 Steel_4130_Q_T_BHN_366
 Steel_4140_Q_T_BHN_293
 Steel_4140_Q_T_BHN_475
 Steel_4142_As_Quenched_BHN_670
 Steel_4142_Q_T_BHN_380
 Steel_4142_Q_T_BHN_400
 Steel_4142_Q_T_BHN_450_b_0_076
 Steel_4142_Q_T_BHN_450_b_0_086
 Steel_4142_Q_T_BHN_475
 Steel_4340_Hot_Rolled_BHN_243
 Steel_4340_Q_T_BHN_275
 Steel_4340_Q_T_BHN_409
 Steel_4340_Su_1172
 Steel_5160_Q_T_BHN_430
 Steel_8630_Cast_BHN_254
 Steel_9262_BHN_260
 Steel_9262_BHN_275
 Steel_9262_BHN_405
 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_4340_bar_Su_1090
 Steel_4340_bar_Su_1482
 Steel_4340_bar_Su_1896
 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_H900_bar_Su_1393
 Steel_H950_bar_Su_1689
 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_2024_T3_Su_490
 Aluminum_2024_T3_Su_496
 Aluminum_2024_T4_Su_476
 Aluminum_2024_T6_Su_475
 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_6061_T6_Hand_Forged_Su_340
 Aluminum_6061_T6_Su_310
 Aluminum_7049_T73_Su_537
 Aluminum_7049_T73_Su_517
 Aluminum_7050_T7351X_Su_517
 Aluminum_7050_T7451_plate_Su_544
 Aluminum_7050_T7451_plate_Su_530
 Aluminum_7050_T7452_Su_537
 Aluminum_7050_T7452_Su_524
 Aluminum_7050_T7651X_Su_599
 Aluminum_7075_T6_Su_572
 Aluminum_7075_T6_Su_579
 Aluminum_7075_T6_Su_565_b_0_105
 Aluminum_7075_T6_Su_565_b_0_172
 Aluminum_7075_T651_Su_580
 Aluminum_7149_T73_Su_503
 Aluminum_7175_T73_Hand_Forged_Su_524
 Aluminum_7175_T73611_Su_524
 Aluminum_7175_T74_Su_510
 Aluminum_7475_T7351_plate_Su_482
 Aluminum_Cast_A356_T6_Su_252
 Aluminum_Cast_A356_T6_Su_266
 Aluminum_Cast_A356_T6_Su_283

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

You might also like