Athanasios I Kolios Thesis 2010

CRANFIELD UNIVERSITY
ATHANASIOS IOANNIS KOLIOS
A MULTI-CONFIGURATION APPROACH TO RELIABILITY BASED

STRUCTURAL INTEGRITY ASSESSMENT FOR ULTIMATE
STRENGTH
SCHOOL OF ENGINEERING
Department Of Offshore, Process & Energy Engineering
Thesis Submitted for the Degree of Doctor of Philosophy

Academic Year: 2010
Supervisor: Professor Feargal Brennan

November 2010
CRANFIELD UNIVERSITY
SCHOOL OF ENGINEERING
Department Of Offshore, Process & Energy Engineering
Academic Year 2010
ATHANASIOS IOANNIS KOLIOS
Supervisor: Professor Feargal Brennan
November 2010
© Cranfield University 2010. All rights reserved. No part of this

publication may be reproduced without the written permission of the
copyright owner.
ABSTRACT
Structural Reliability treats uncertainties in structural design systematically,

evaluating the levels of safety and serviceability of structures. During the past
decades, it has been established as a valuable design tool for the description of
the performance of structures, and lately stands as a basis in the background of
the most of the modern design standards, aiming to achieve a uniform
behaviour within a class of structures. Several methods have been proposed for
the estimation of structural reliability, both deterministic (FORM and SORM) and
stochastic (Monte Carlo Simulation etc) in nature.
Offshore structures should resist complicated and, in most cases, combined

environmental phenomena of greatly uncertain magnitude (eg. wind, wave,
current, operational loads etc). Failure mechanisms of structural systems and
components are expressed through limit state functions, which distinguish a
failure and a safe region of operation. For a jacket offshore structure, which
comprises of multiple tubular members interconnected in a three dimensional
truss configuration, the limit state function should link the actual load or load
combination acting on it locally, to the response of each structural member.
The response of a structure under specific loading conditions can be

determined through Finite Element Analysis Methods. Based on that, advanced
methods (intrusive and non-intrusive) have been developed, such as the
Spectral Stochastic Finite Element Method and the Stochastic Response
Surface method. This Thesis presents a methodology for the structural reliability
assessment of an offshore jacket structure, which has been selected as a
reference application, based on a generalized Stochastic Response Surface
Method. According to the methodology that is proposed, simulation results
obtained by FEA Modelling are combined with numerical reliability procedures,
through multivariate (quadratic) polynomial regression (MPR), in order to
calculate the reliability indices of members. This procedure is particularly useful
as it enables efficient analysis of members under a stochastic perspective,
incorporating design uncertainties. By effectively dividing the analysis into
i
different independent ‘blocks’ in an open sequence, the required number of
Finite Element simulations is greatly reduced. This implies that depending on
the requirements of the structure, different resolutions and tools can be used in
each of those blocks. The reduced resource requirements of this method can
later allow optimization of a complex structure in a closed design procedure.
The novelty of this methodology is in its simplicity and the ability to analyze a
wide range of structures, and further intricate reliability problems, under
complicated load combinations in a more efficient way.
Consideration of appropriate limit state functions is an essential decision for

accurate prediction of the reliability index. For the reference structure, analytical
limit states have been introduced following fundamental failure criteria for
ultimate strength and buckling resistance of ductile members. Hence, after
appropriate modelling of stochastic variables, and based on the derived limit
states, the reliability level is calculated in both a local and a system level. A
sensitivity analysis of the design parameters (surface roughness, variables’
modelling, corrosion deterioration etc) illustrates their effect on the derived
values. On a parallel study, deriving limit states based on the design
requirements of the most widely used design standards for offshore steel
structures (API RP-2A, ISO 19902) but also generic standards for steel
structures (EN 1993, AISC/ANSI), the minimum reliability indices of members
are derived and later compared to the ones obtained by the analytical limit
states. Using the same calculation procedure for the estimation or reliability,
such a comparison can be realistically executed and useful conclusions can be
drawn for the performance of different design standards
The methodology that is derived and presented in this Thesis, can be extended
to the probabilistic assessment of different engineering problems, including
problems of solid mechanics and heat transfer, where detailed analysis is
required for the derivation of the response of the structure. Further, this
methodology can stand as a generic document that can be applied in
conjunction to design standards towards a robust reliability based design.
ii
Keywords:
Offshore Structures, Reliability Assessment, Stochastic Response Surface

Method, Structural Integrity, Design Standards, API RP-2A, ISO 19902, EN
1993, AISC/ANSI
iii
Dedicated to John, Irene, Helen and Michael…
“Της δικαιοσύνης ήλιε νοητέ “Notional sun of justice

και μυρσίνη συ δοξαστική and you myrtle of glory
μη παρακαλώ σας μη don’t, please don’t
λησμονάτε τη χώρα μου…” forget my country…”
Οδυσσέας Ελύτης, (Το Άξιον Εστί, 1959) Odysseas Elytis, (To Axion Esti, 1959)
Νόμπελ Λογοτεχνίας, 1979 Nobel Prize in Literature, 1979
i
ACKNOWLEDGEMENTS
First of all I would like to thank my supervisor, Professor Feargal Brennan, for
his confident supervision and generous support throughout the three years of
this PhD. His advice has mentored my future academic endeavors.
I should express my greatest gratitude to my friend and colleague Dr. Amir

Chahardehi (Cranfield University), not only for proof reading my Thesis but also
for the hours of long discussions we have spent related to fundamental aspects
of Engineering, that have enlightened different perspectives for research.
I would like to thank Professor Iberahin Jusoh, who honors me with his
friendship, from the Technical University of Malaysia for his valuable experience
and advice on the design of offshore structures and his support in my research.
Professor Dimitrios Papantonis (National Technical University of Athens) and

Dr. Pavlos Aleiferis (University College London) have trusted me with their
references; I hope that the contribution of this PhD meets their expectations.
I own appreciation to everyone that has supported this endeavor, especially my

dear friends Mr. Dimitros Katsaros, Mr. Antonios Antoniadis, Mr. George
Kosmidis, and my colleagues in Cranfield University. A special thank to Dr.
Anastasia Athanasoulia.
Finally, I would like to grant the deepest gratitude to my parents, for the
principles they inspired me, and their constant support to all of my dreams.
iii
TABLE OF CONTENTS
ABSTRACT ......................................................................................................... i
ACKNOWLEDGEMENTS...................................................................................iii
LIST OF FIGURES.............................................................................................ix
LIST OF TABLES ...............................................................................................xi
NOMENCLATURE ........................................................................................... xiii
Symbols............................................................................................................ xiii
Abbreviations.................................................................................................... xxi
1 INTRODUCTION – CONTEXT OF STRUCTURAL SAFETY...................... 1
1.1 Introduction ........................................................................................... 1
1.2 Structural Reliability............................................................................... 5
1.3 Overview of Reliability Analysis of Offshore Structures......................... 9
1.4 Design of Offshore Structures ............................................................. 11
1.4.1 Permissible stresses..................................................................... 11
1.4.2 Global Safety Factor ..................................................................... 12
1.4.3 Partial Safety factor ...................................................................... 12
1.4.4 Probabilistic Methods.................................................................... 13
1.4.5 Comments on design methods ..................................................... 13
1.5 Limit State Design ............................................................................... 14
1.5.1 Serviceability Limit State............................................................... 15
1.5.2 Ultimate Limit State....................................................................... 16
1.5.3 Fatigue Limit State........................................................................ 18
1.5.4 Accidental Limit State ................................................................... 18
1.6 Design Standards for Steel Structures ................................................ 19
1.6.1 General......................................................................................... 19
1.6.2 Categorization of Design Standards ............................................. 20
1.6.3 Standards for Offshore Structures ................................................ 20
1.7 Target – Acceptable Reliability............................................................ 24
1.8 Summary............................................................................................. 27
2 RELIABILITY ANALYSIS OF OFFSHORE STRUCTURES ...................... 31
2.1 Basic formulation of the Problem ........................................................ 31
2.2 Background and motivation................................................................. 34
2.2.1 Development of Structural Reliability Applications........................ 34
2.2.2 Development of offshore industry ................................................. 35
2.2.3 Application of Reliability Analysis in the offshore industry ............ 36
2.3 Response Analysis.............................................................................. 38
2.3.1 Static Analysis .............................................................................. 38
2.3.2 Dynamic Analysis ......................................................................... 39
2.3.3 Deterministic and Stochastic Processes....................................... 40
2.3.4 Selection of type of analysis ......................................................... 40
2.3.5 System Response......................................................................... 42
2.3.6 Practical Methods of Analysis ....................................................... 45
2.3.7 Modelling of Post-Failure Behaviour ............................................. 46
2.3.8 Methods in Computing System Reliability..................................... 47
2.4 Review of Stochastic Methods ............................................................ 51
2.4.1 Stochastic Expansions.................................................................. 51
v
2.4.2 Spectral Stochastic Finite Elements ............................................. 53
2.5 Summary............................................................................................. 58
3 NUMERICAL METHODS FOR STRUCTURAL RELIABILITY ANALYSIS 59
3.1 Introduction ......................................................................................... 59
3.2 Numerical Methods ............................................................................. 59
3.2.1 Deterministic Methods .................................................................. 59
3.2.2 Simulation Methods ...................................................................... 76
3.3 Stochastic Response Surface Method ................................................ 87
3.3.1 General Concept........................................................................... 87
3.3.2 Review and Notation..................................................................... 90
3.3.3 Adaptive Response Surface Method ............................................ 92
3.3.4 Algorithms of the Stochastic Response Surface Method .............. 94
3.4 Regression methods ........................................................................... 95
3.4.1 Linear Regression......................................................................... 95
3.4.2 Multivariate Regression ................................................................ 97
3.4.3 Alternative Regression Methods ................................................... 98
3.5 Numerical model Developed ............................................................... 99
3.5.1 Description.................................................................................... 99
3.5.2 Verification Process .................................................................... 102
3.5.3 Validation of the FORM/SORM code .......................................... 108
3.5.4 Codes Included........................................................................... 112
3.6 Summary........................................................................................... 112
4 STOCHASTIC MODELING OF ENVIRONMENTAL LOADING AND LOAD
CAPACITY OF OFFSHORE STRUCTURES ................................................. 115
4.1 Introduction ....................................................................................... 115
4.2 Classification of Loads ...................................................................... 115
4.3 Environmental Loading...................................................................... 117
4.3.1 Design Environmental Conditions............................................... 118
4.3.2 Environmental Loads .................................................................. 119
4.3.3 Wave Modelling .......................................................................... 124
4.3.4 Joint Probability of Environmental Parameters ........................... 132
4.3.5 Estimation of Extreme and Design Values.................................. 136
4.3.6 Fluid Loading on Offshore Structures ......................................... 140
4.3.7 Response of Structure under Environmental Loading ................ 143
4.4 Capacity of Offshore Structures ........................................................ 144
4.4.1 Resistance Model ....................................................................... 144
4.4.2 Material Data .............................................................................. 145
4.4.3 Geometry Data ........................................................................... 147
4.4.4 Capacity of Members and Joints................................................. 148
4.4.5 Geotechnical Data ...................................................................... 149
4.4.6 Fatigue Data ............................................................................... 153
4.4.7 Corrosion Modelling.................................................................... 154
4.5 Summary........................................................................................... 163
5 APPLICATION OF THE RESPONSE SURFACE METHOD IN A TYPICAL
JACKET STRUCTURE .................................................................................. 165
5.1 Introduction ....................................................................................... 165
5.2 Limit State Formulation ..................................................................... 165
5.2.1 Maximum shear stress (MSS) – Tresca or Guest Theory........... 168
vi
5.2.2 Distortion energy (DE) – von Mises Stress Theory ..................... 170
5.2.3 Ductile Coulomb-Mohr (DCM)..................................................... 172
5.2.4 Comments on Failure Criteria ..................................................... 174
5.2.5 Buckling Limit state..................................................................... 176
5.3 Application of Joint Probability Distribution........................................ 179
5.4 Reference Structure .......................................................................... 187
5.5 Component Structural Reliability Assessment................................... 191
5.5.1 Base case................................................................................... 191
5.5.2 Sensitivity analysis of design parameters ................................... 196
5.5.3 System Reliability Integration ..................................................... 206
5.6 A Note on the dynamic loading of structures..................................... 213
5.7 Summary........................................................................................... 215
6 COMPARISON OF THE ULTIMATE STRENGTH RELIABILITY
PERFORMANCE OF RELEVANT DESIGN PROCEDURES......................... 217
6.1 Introduction ....................................................................................... 217
6.2 API RP-2A: Recommended practice for planning, designing and
constructing fixed offshore platforms LRFD ................................................ 218
6.2.1 Design Provisions ....................................................................... 218
6.2.2 Numerical Results....................................................................... 224
6.3 ISO 19902:2002: Petroleum and natural gas industries-general
requirements for offshore structures ........................................................... 229
6.4 BS EN 1993-1-1:2005 Eurocode 3: Design of Steel Structures ........ 241
6.5 ANSI/AISC 360-05: Specification for structural steel buildings.......... 254
6.5.2 Numerical Application ................................................................. 259
6.6 Discussion......................................................................................... 262
6.7 Summary........................................................................................... 267
7 CONCLUSIONS AND RECCOMENTATIONS ........................................ 269
7.1 Summary........................................................................................... 269
7.2 Contribution of this PhD .................................................................... 271
7.3 Future recommendations: ................................................................. 277
REFERENCES............................................................................................... 281
APPENDICES ................................................................................................ 313
APPENDIX A.................................................................................................. 313
APPENDIX B.................................................................................................. 322
APPENDIX C.................................................................................................. 327
APPENDIX D.................................................................................................. 329
vii
LIST OF FIGURES
Figure 1: Response Surface Method Flow Chart................................................ 3

Figure 2: Structural Design Consideration of the ULS...................................... 17
Figure 3: Iterative procedure for Risk Assessment [48].................................... 29
Figure 4: Relationship between β and Probability of failure.............................. 33
Figure 5: Definition of Reliability Index ............................................................. 33
Figure 6: Scatter of response (Max Base Shear-MN) by different Analysis
Methods [75] ............................................................................................. 41
Figure 7: Classification of methods for system reliability assessment [76] ....... 43
Figure 8: Models of Post-Failure Behaviour ..................................................... 46
Figure 9: Intrusive and non-intrusive formulation [93]....................................... 53
Figure 10: First and Second order approximations........................................... 60
Figure 11: Transformation to the U-space [93] ................................................. 61
Figure 12: Algorithm of HL Reliability Index Calculation................................... 68
Figure 13: Normalized Tail Approximation [116]............................................... 70
Figure 14: Area calculation with Monte Carlo Simulation (500 and 5000
sampling points) ........................................................................................ 77
Figure 15: Inverse transformation method........................................................ 78
Figure 16: Latin Hypercube Method ................................................................. 79
Figure 17: Monte Carlo Simulation Method ...................................................... 80
Figure 18: Design Point Simulation .................................................................. 83
Figure 19: Directional Simulation...................................................................... 85
Figure 20: Axis-Orthogonal Simulation............................................................. 87
Figure 21: Regression vs. Order of Polynomial ................................................ 89
Figure 22: Different Sampling Approaches [160].............................................. 89
Figure 23: Calculation of reliability based on Direct Simulation ...................... 100
Figure 24: Calculation of reliability based on Normal Response Surface Method
................................................................................................................ 100
Figure 25: Calculation of reliability based on Adaptive Response Surface
Method .................................................................................................... 101
Figure 26: Reference structure....................................................................... 105
Figure 27: ABAQUS and Matlab code deformed model ................................. 106
Figure 28: FEA model in DNV Genie software ............................................... 109
Figure 29: Methods for combining current and wave [186]............................. 121
Figure 30: Ranges of appropriate wave theories [1]...................................... 124
Figure 31: Ranges of appropriate wave theories [193]................................... 125
Figure 32: Coordinate system of wave propagation [194] .............................. 126
Figure 33: Applicability regions of wave theories based on the relative error on
the fit of the two nonlinear free surface boundary conditions [195] ......... 127
Figure 34: CD-KC number for different values of Re and β=Re/KC [199] ...... 130
Figure 35: CM-KC number for different values of Re and β=Re/KC [199] ...... 130
Figure 36: Frequency domain analysis [232].................................................. 142
Figure 37: Static and Dynamic analysis response curve ................................ 144
Figure 38: Continuous load process and potential exceedance of the
deteriorating structural Resistance [268]................................................. 156
Figure 39: Strength Deterioration Structural Reliably Problem [269] .............. 156
ix
Figure 40: Corrosion depth as a function of sea water temperature [269]...... 157
Figure 41: Effect of water velocity on early loss corrosion [268]..................... 157
Figure 42: Different corrosion mechanism as a function of time [269]............ 159
Figure 43: Thickness of corrosion wastage as a function of time [275] .......... 161
Figure 44: Model for maximum pit depth as a function of exposure period [276]
................................................................................................................ 162
Figure 45: The MSS theory for plane stress problems (two nonzero principal
stresses).................................................................................................. 169
Figure 46: Distortion Energy Theory for plane stress problems...................... 172
Figure 47: Mohr Cycles and Coulomb-Mohr Failure criterion ......................... 173
Figure 48: Coulomb-Mohr Failure criterion for Plane Stress problems........... 174
Figure 49: Linear Plot of real data (Hs (m), Tp (s))......................................... 180
Figure 50: Logarithmic Plot of real data (Hs (m), Tp (s)) ................................ 180
Figure 51: Linear Plot of regenerated data (Hs (m), Tp (s)) ........................... 181
Figure 52: Logarithmic plot of LogN bivariate joint distribution (smoothened) (Hs
(m), Tp (s)) .............................................................................................. 183
Figure 53: Logarithmic plot of Weibull bivariate joint distribution (smoothened)
(Hs (m), Tp (s))........................................................................................ 183
Figure 54: Logarithmic plot of LogN-Weibull bivariate joint distribution
(smoothened) (Hs (m), Tp (s))................................................................. 184
Figure 55: Comparison of best fit on the tail region of the CDF for Hs (m)..... 185
Figure 56: The Reyleigh distribution of wave heights ..................................... 187
Figure 57: FEA model of a jacket structure developed in DNV GeniE software
................................................................................................................ 190
Figure 58: Thickness deterioration as a function of time ................................ 203
Figure 59: Reliability index deterioration of critical members (b454, b458) .... 204
Figure 60: Reliability index deterioration of members (b403-b103) ................ 205
Figure 61: Post failure behaviour.................................................................... 208
Figure 62: Critical Failure paths...................................................................... 208
Figure 63: Displacement vs overturning moment (intact structure) ................ 209
Figure 64: Displacement vs load factor (intact structure) ............................... 209
Figure 65: Displacement vs overturning moment (one member removed)..... 211
Figure 66: Displacement vs overturning moment (two members removed) ... 211
Figure 67: Displacement vs load factor (one member removed) .................... 212
Figure 68: Displacement vs load factor (two members removed)................... 212
Figure 69: Buckling curves ............................................................................. 247
Figure 70: Stress distribution.......................................................................... 275
Figure 71: Reliability distribution..................................................................... 275
Figure 72: Reliability Based Optimization....................................................... 279
Figure 73: Corrosion Cell ............................................................................... 326
Figure 74: Numbering of the structural members ........................................... 327
x
LIST OF TABLES
Table 1: Examples of Limit States according to DNV ....................................... 16

Table 2: Reliability Index and Reliability Classes [7] ........................................ 28
Table 3: Values of acceptable annual probabilities of failure (PF) [14]............. 28
Table 4: Comparative estimates of target Pf [35] ............................................. 29
Table 5: Reliability Index and Probability of Failure vs. Sampling Number....... 92
Table 6: Stochastic Loads Consideration ....................................................... 105
Table 7: Verification of FEA code: results (kPa) ............................................. 106
Table 8: Sensitivity Analysis of MCS Probability of failure results .................. 107
Table 9: MCS vs ARSM Probability of failure results...................................... 107
Table 10: Stochastic loads on verification model............................................ 109
Table 11: Results of comparative analysis of test case structure ................... 111
Table 12: Values of hydrodynamic coefficient for circular cylinders [202] ...... 131
Table 13: Yield Strength Properties [235]....................................................... 146
Table 14: Young’s modulus and Poisson’s ratio............................................. 146
Table 15: Variation in thickness of steel plates [239]...................................... 147
Table 16: Probability Density Function for Soil Characteristics [254] ............. 151
Table 17: Typical annual corrosion rates [270], [271]..................................... 158
Table 18: Phases in pitting corrosion and calibrated functions for model
parameters [276] ..................................................................................... 162
Table 19: MSS Failure Criterion for Plane stress problems............................ 170
Table 20: CM Failure Criterion for Plane stress problems.............................. 173
Table 21: Limit states for different Failure Criteria.......................................... 176
Table 22: Distribution coefficients................................................................... 182
Table 23: Wave Height Statistical Correlations [286] ..................................... 186
Table 24: Design load input parameters for dimensioning of the structure..... 190
Table 25: Properties of stochastic variables................................................... 191
Table 26: Reliability Index of Horizontal Members ......................................... 192
Table 27: Reliability Index of Vertical Members.............................................. 192
Table 28: Reliability Index of Legs Members.................................................. 193
Table 29: Minimum Reliability index of members, incorporating 8 different
directions................................................................................................. 195
Table 30: Reliability indices for buckling limit states (10 critical members) .... 196
Table 31: Reliability index of smooth and rough cylinders (30 critical members)
................................................................................................................ 197
Table 32: Reliability index for different wave theories (20 members with critical
effect) ...................................................................................................... 198
Table 33: Parameters of equivalent normal distributions................................ 199
Table 34: Reliability indices for equivalent normal distributions...................... 200
Table 35: Reliability indices for cases of 25% reduced loads......................... 201
Table 36: Reliability indices for S355 and S275 steels................................... 202
Table 37: Reliability indices for 20 years for different corrosion models......... 204
Table 38: Bounds of System Reliability .......................................................... 213
Table 39: Effective length and bending reduction factors............................... 224
Table 40: Limit States according to API-LRFD ............................................... 225
Table 41: Reliability indices for Leg Members ................................................ 225
xi
Table 42: Reliability indices for Horizontal Brace Members ........................... 226
Table 43: Reliability indices for Vertical Brace Members................................ 227
Table 44: Effective length and moment reduction factors for member strength
checking .................................................................................................. 236
Table 45: Limit States according to ISO 19902 .............................................. 237
Table 49: Classification of members .............................................................. 242
Table 50: Imperfection factors for buckling curves ......................................... 246
Table 51: Limit States according to EN 1993 ................................................. 250
Table 55: Limit States according to ANSI/AISC 360-05 ................................. 260
Table 59: Minimum Reliability indices for Leg Members ................................ 265
Table 60: Minimum Reliability indices for Horizontal Brace Members ............ 266
Table 61: Minimum Reliability indices for Vertical Brace Members ................ 266
Table 62: Reduction factors of different standards ......................................... 266
Table 63: Critical Members............................................................................. 276
Table 64: Shallow and deep water approximation to linear wave theory [194]315
Table 65: Φn' and ηn' the coefficients [194]..................................................... 317
Table 66: Stokes fifth-order wave theory [194] ............................................... 318
Table 67: The coefficients for Stokes fifth-order wave theory [311]................ 319
Table 68: Typical coating systems used for offshore structures..................... 323
Table 69: Requirements of Galvanic and an Impressed-Current Corrosion
Protection System ................................................................................... 326
Table 70: Dimension of group of members .................................................... 328
xii
NOMENCLATURE
Symbols
‫ܣ‬ Coefficient of Reyleigh Parameter

‫ܣ‬ Cross Sectional Area
‫ܣ‬ Transformation matrix
‫ܣ‬ூ, ‫ܣ‬஽ Characteristic Areas
‫ܣ‬௘ Effective net area, calculated as the product of the net area ‫ܣ‬௡
and the shear lag factor ܷ specified for different connections of
tension members.
‫ܣ‬௚ Gross area of section based on design wall thickness
‫ܣ‬௚ Gross area of member
‫ܤ‬ Matrix Relating Strains to Displacements
‫ܥ‬ Global Damping Matrix
‫ܥ‬ Buckling Parameter
‫ܥ‬ Multiplication Correction
‫ܥ‬ Torsional constant
‫ܥ‬ Total Loss of Material Function
‫ݔ(ܥ‬, ߠ) Randomness Matrix
‫ܥ‬஽ Drag Coefficient
‫ܥ‬ெ Inertia Coefficient
‫ܥ‬ௌ Shape Coefficient
‫ܥ‬௠ ,௬ , ‫ܥ‬௠ ,௭ Moment Reduction Factors
‫ܥ‬௠ ௬ , ‫ܥ‬௠ ௭ Reduction factors
‫ܥ‬௫ Elastic Critical Buckling Coefficient ‫ܦ‬ Outside diameter
‫ܦ‬ Diameter of Cylindrical Members
‫ܦ‬ Elasticity Matrix
‫ܦ‬ Spatial Domain
‫ܦ‬௞ Relative Distance Between Sampling Points and Current Design
Point
‫ܧ‬ Load Effect
‫ܧ‬ Modulus of Elasticity
‫ܧ‬ Vector of Environmental Conditions
‫ܧ‬ Young’s Modulus of Elasticity
‫)ݑ(ܧ‬ Expected Value of Response
‫ܧ‬ത(‫)ݔ‬ Mean value of stochastic process
‫ܧ‬௉ Probability of Excedance during ‫ܮ‬௙
‫ܧ‬ௗ Design Value of Load Effect
‫ܧ‬௞ Characteristic Value of Load Effect
‫ܨ‬௫೔(‫ݔ‬௜) Marginal Cumulative Distribution Function
‫ܨ‬௛௖ Nominal critical hoop buckling strength
‫ܨ‬௕௡ Nominal Bending Strength
‫ܨ‬௖௡ Nominal Axial Compressive Strength
‫ܨ‬௖௡ Nominal Axial Compressive Strength
xiii
‫ܨ‬௖௥ The greatest of (without exceeding 0.6‫ܨ‬௬ )
‫ܨ‬ௗ Design Value of variables
‫ܨ‬௘ Elastic critical buckling
‫ܨ‬௘௬ , ‫ܨ‬௘௭ Euler buckling strengths
‫ܨ‬௜௡௧ Internal reaction force vector
‫ܨ‬௨ Specified minimum tensile stress
‫ܨ‬௩௡ Nominal Shear strength
‫ܨ‬௩௧௡ Nominal torsional strength
‫ܨ‬௫௖ Nominal Inelastic Local Buckling Strength
‫ܨ‬௫௘ Nominal Elastic Local Buckling Strength
‫ܨ‬௬ Nominal Yield Strength
‫ܨ‬௬ Specified minimum yield stress
‫ܩ‬ Shear Modulus
‫ݐ(ܩ‬, ‫)ݖ‬ Gust Factor
‫ܩ‬஼ெ Failure Limit State for CM failure criterion
‫ܩ‬஽ா Failure Limit State for DE failure criterion
‫ܩ‬ெ ௌௌ Failure Limit State for MSS failure criterion
‫ܩ‬௘௤ Equivalent Limit State
‫ܪ‬ Wave Height
‫ܪ‬ Transformed Orthogonal Matrix
‫ܪ‬ଵ/ଵ଴ 1/10 Highest Wave Height
‫ܪ‬௠ ௔௫ Maximum Design Wave Height
‫ܪ‬௢ Average Wave Height
‫ܪ‬௦, ‫ܪ‬ଵ/ଷ Significant Wave Height
‫ܪ‬௪ Wave Height
‫ܫ‬ Moment of inertia
‫)ݖ(ܫ‬ Turbulence Intensity
‫)࢞(ܫ‬ Indicator Function
‫ܫ‬௣ Polar moment of inertia
‫ܫ‬௣ Polar Moment of Inertia
‫ܬ‬ Polar Moment of inertia
‫ܭ‬ Effective length factor in plane of bending, equals 1 unless
different analysis indicates a smaller value
‫ܭ‬ Effective Length Factor
‫ܭ‬ Effective Length Factor
‫ܭ‬ Global Stiffness Matrix
‫ܭ‬ Covariance Matrix
‫ܥܭ‬ Kuelegan-Carpenter Number
‫ܭ‬ூ஼ Stress Intensity Factors
‫ܭ‬௬ , ‫ܭ‬௭ Effective length factors
‫ܮ‬ Laterally un-braced length of member
‫ܮ‬ Length of Member
‫ܮ‬ Load
‫ܮ‬ Un-braced Length
‫ܮ‬ Wave Length
‫ܮ‬௘௥ Buckling length in the buckling plane
xiv
‫ܮ‬௙ Service Life
‫ܮ‬௩ Distance from maximum to zero shear force
‫ܮ‬௬ , ‫ܮ‬௭ Un-braced lengths
‫ܯ‬ Applied Bending moment
‫ܯ‬ Bending Moment
‫ܯ‬ Global Mass Matrix
‫ ܯ‬ாௗ Design value of the moment
‫ ܯ‬ாௗ Bending moment
‫ ܯ‬ே ,ோௗ Design plastic moment resistance reduced due to axial force
‫ ܯ‬௕,ோௗ Design buckling resistance moment
‫ܯ‬௖ Available flexural strength
‫ ܯ‬௖௥ Elastic critical moment for lateral-torsional buckling
‫ܯ‬௥ Required flexural strength
‫ܯ‬௥ Required flexural strength
‫ܯ‬௨ Regression Design Matrix
‫ ܯ‬௩,௧ Torsional Moment due to Factored Actions
‫ ܯ‬௩௧ Torsional moment
‫ܯ‬௖ Design flexural strength
ܰ Total Realizations
ܰாௗ Design value of the compression force
ܰாௗ Design value of the compression force
ܰாௗ , ‫ ܯ‬௬,ாௗ , ‫ ܯ‬௭,ாௗ Design values of the action at point of consideration
ܰோௗ , ‫ ܯ‬௬,ோௗ , ‫ ܯ‬௭,ோௗ Design Values of the Resistance depending on the cross
sectional classification and including any reduction that may be
caused by shear effects
ܰ௕,ோௗ Design buckling resistance of the compression member
ܰ௖௥ Elastic critical force for the relevant buckling mode
ܰ௙ Successful Realizations
ܰ௫௫ Axial Force
ܲ Orthogonal Eigenvector Matrix
ܲ Tensile Force
ܲ Vector of Parameters for the CPS
ܲ஺ True Probability of Event
ܲா Estimated Probability in LHC
ܲ௅ Probability of linear zed domain
ܲ௖ Available axial compressive strength
ܲ௖ Available axial tensile strength
ܲ௖ Design tensile or compressive strength
ܲ௖௥ Critical Load for Unstable Bending
ܲௗ Design Value of Probability of Failure
ܲ௙ Probability of Failure
ܲ௥ Required axial strength
ܲ௥ Required axial compressive (tensile) strength
ܲ෠ா Estimated Probability of an Event
ܳ Bernoulli’s constant
ܳ Probability of Excedance
xv
ܴ Nodal Load Vector
ܴ Structural Resistance
ܴ Time-dependent Load
ܴ݁ Reynolds Number
ܴ෠, ܵመ Standard Normalized Random Variables
ܴଶ Coefficient of Determination
ܴ௎ Ultimate Lateral Load Capacity
ܴௗ Design Value of Resistance
ܴ௞ Characteristic Value of Resistance
ܴ௡ Nominal strength, as specified for each case of loading
ܴ௨ Required strength
ܵ Elastic section modulus
ܵ Elastic Section Modulus
ܵ First moment of area
ܵி (߱) Wave Force Transfer Function
ܵோ Reference Lateral Loading
ܵௐ (߱) Sea Surface Spectrum Function
ܵ௧ Material Tensile Strength
ܵ௨௖ Material Ultimate Compressive Stress
ܵ௨௧ Material Ultimate Tensile Stress
ܵ௬ Material Yield Strength
ܶ Correlated Random Vector
ܶ Wave Period
ܶ஼ Design torsional strength
ܶாௗ Design value of the torsional moment
ܶ௉ Spectral Peak Period
ܶ௉ Peak spectral period
ܶோ Return Period
ܶோௗ Design torsional resistance of the cross section.
ܶ௥ Required torsional strength
ܷ Vector of Standardized Independent Variables
ܷீ (‫)ݖ‬ Gust Wind Speed at z m above SWL
்ܷௌ Speed of Current
்ܷ௓ Speed of Current at z-level
ܷ௪ (1݉ , ‫)ݖ‬ 1-minute Mean Sustained Wind Speed at 10 m above Sea Water
Level
ܸ Beam Shear
ܸ Member Shear
ܸாௗ Design value of the shear force
ܸாௗ Shear force
ܸ௖,ோௗ Design shear resistance
ܸ௡ Design shear strength
ܸ௥ Required shear strength
ܹீ Weight Matrix
ܹ ௘௟,௠ ௜௡ , ܹ ௘௙௙,௠ ௜௡ Maximum elastic stress
ܺ Matrix of Dependent Variables
ܺ Stochastic Variables vector
xvi
ܺ Uncorrelated Random Variables
ܻ Matrix of Independent Variables
ܻ Transformed Matrix
ത
ܻ Estimated Values
ܼ Plastic section modulus about the axis of bending
ܼ Plastic Section Modulus
ܼ௘ Elastic Section Modulus
ܼ Safety Margin, Limit state equation
ܾ௜ First Order Regression Coefficients

ܾ௜ Polynomial Coefficients
ܿ Celerity
ܿ௜ Linear Coefficients
ܿ௜ Second Order Regression Coefficients
݀ Still water depth
݀ Total Water Depth
݀(‫)ݐ‬ Corrosion Wastage Thickness
݀ஶ Long term thickness corrosion wastage
݀௜௝ Mixed Term Regression Coefficients
݁ Regression Error
݂ Identification Coefficient
݂ Total Hydrodynamic Wave Force
݂(‫ݔ‬, ‫)ݕ‬ Bivariate Distribution Function
݂௫೔(‫ݔ‬௜) Probability Density Function
݂௛ Hoop stress due to factored hydrostatic pressure
݂௕,௛ Representative Bending Strength in the presence of external
hydrostatic pressure
݂௕ Bending Stress
݂௕ Representative Bending Strength
݂௕௬ , ݂௕௭ Bending stresses about y and z- axis
݂௖,௛ Representative Axial Compressive Strength in the presence of
external hydrostatic pressure
݂௖ Axial Compressive Stress
݂௖ Representative Axial Compressive Strength
݂௘,௬ , ݂௘,௭ Euler Buckling Strengths
݂௘ Euler Buckling Strength
݂௧,௛ Representative Axial Tensile Strength in the presence of external
hydrostatic pressure
݂௧ Representative Axial Tensile Strength
݂௧ Axial Tensile Stress
݂௩௧ Torsional shear stress
݂௫(‫)ݔ‬ Initial Distribution Function
݂௫௘ Representative Elastic Local Buckling Strength
݂௬ Representative Yield Strength
݂௬௖ Representative Local Buckling Strength
xvii
݂௬௖ Representative Load Buckling Strength
݂ఔ Representative Shear Strength
݃(‫)ݐ‬ Gust Factor Component
݃(‫)ݔ‬ Indicator function
݃(ܷ) Transformed Limit State Surface
݃෤(‫)ݔ‬ Approximate Limit State Function
݅ Radius of gyration about the relevant axis
݇ Area of gyration
݇ Safety Factor
݇ Wave Number
݇(ߠ) Global Stiffness Matrix
݇௬௬ , ݇௬௭, ݇௭௬ , ݇௭௭ Interaction factors (App. A and B of standard)
݉ Number of Segments in LHC
݊ Limit State Safety Factor
‫݌‬ Hydrostatic Pressure
‫݌‬௑ (‫)ݔ‬ Sampling Distribution
‫݌‬௠ ௡(‫ݔ‬ ) Regression Polynomials
‫ݎ‬ Governing radius of gyration
‫ݎ‬ Correlation coefficient
‫ݎ‬ Element Stiffness Matrix
‫ݎ‬ Radius of Gyration
‫ݎ‬ Radius of Gyration
‫)ݐ(ݎ‬ Displacement Vector
‫ݏ‬ Global Safety Factor
‫ݏ‬଴ Target Global Safety Factor
‫ݐ‬ Time Parameter
‫ݐ‬ Time Variables
‫ݑ‬ Coefficient vector
‫ݑ‬ Poisson’s Ration
‫ݑ‬ Strain Energy
‫)ߠ(ݑ‬ Gaussian Random Response
‫ݑ‬௜ Transformed ‫ݔ‬௜ into their Standardized Forms
∗
‫ݑ‬௜ Transformed Intermediate Coordinates
‫ݒ‬ Poisson’s Ratio
‫ݑ‬௠ ௔௫ Maximum Effect in Response
‫ݑ‬௦௧௔௧௜௖ Static Effect in Response
‫ݓ‬ Sea water density
‫ ݓ‬஽௞ Weight Factor
‫ீ ݓ‬௞ Weight Factor
∗
‫ݔ‬௜ Intermediate Coordinates
‫ݔ‬ Subscript relating symbol to strong axis bending
‫ݕ‬ Subscript relating symbol to weak axis bending
‫ݕ‬ො௕௘௦௧ Best Design Value
‫ݖ‬ Depth below still water surface
‫ݖ‬ Distance above the Seabed
xviii
߁௡ n-th order polynomial of variables
߁௣ Hermite Polynomials
߂ߊ ௬,ாௗ , ߂ߊ ௭,ாௗ Moments due to the shift of the centroidal axis
߉ Eigen-value Matrix
ߔ Cumulative Distribution Function
ߔ, ߖ Stream Function (velocity potential)
ߔ ௎ିଵ() Inverse Standardized Normal Distribution Function
ߙ Imperfection factor
ߙ, ߚ, ߛ, ߣ Coefficients of Weibull Bivariate Distribution
ߙ௅் Imperfection factor
ܽ
ത Coefficients Matrix
ܽௗ Material’s Uncertainties
ܽ௜ Indicator Variables
ܽ௜ Sensitivity Factors
ܽ௞ Characteristic Value of Material’s Uncertainties
ߚ Reliability index
ߚு௅ Hasofer-Lind reliability index
ߚௗ Design Value of Reliability index
ߛ Partial Factor
ߛ஽ Hydrostatic pressure load factor
ߛெ ଴ Resistance partial factor of cross section applicable to all classes
ߛெ ଵ Resistance of members to instability
ߛெ ଵ Resistance of members to instability
ߛெ ଶ Resistance of cross section in tension to fracture
ߛோ,௕ Partial Resistance Factor for Bending Strength
ߛோ,௖ Partial Resistance Factor for Axial Compressive Strength
ߛோ,௧ Partial Resistance Factor for Axial Tensile Strength
ߛோ,ఔ Partial Resistance Factor for Shear Strength
ߛ௠ Material Reduction Factor
ߛ௢ Partial safety factor
ߜ௜௝ Delta Function
ߝ Perturbation Parameter
ߝ Residual
ߝ Strain
ߝ Zero Mean Error Function
ߝ௙ Maximum Strain (Elongation) before Failure
ߟ Free Surface Profile
ߠ Random Process
ߠ Wave Phase Angle
ߠௗ Model’s Uncertainties
ߠ௞ Characteristic Value of Model’s Uncertainties
ߣ Column Slenderness Parameter
ߣ Eigen-values
ߣ௬ , ߣ௭ Column slenderness parameters
ߤ(‫)ݔ‬ Mean Value
xix
ߤௗ (‫)ݐ‬ Mean Value of Deterioration Model
ߤఄ Vector of Mean Values
(ߦଵ, ߦଶ) Gaussian variables
ߩ Air Density
ߩ Correlation coefficient
ߩோௌ Correlation Coefficient between R and S
ߪ Stress
ߪ௚෤ Standard Deviation of the Approximate Limit-State Function
ߪᇱ von Mises stress
ߪଵ, ߪଶ, ߪଷ Principle Stresses
ߪ௛ Hoop Stress due to forces from factored hydrostatic pressure
ߪ஺ , ߪ஻ Principle Stresses for plane stress problems
ߪ௔,௫௫ Axial Stress
ߪ௔௩ Hydrostatic Stress
ߪ௕,௫௬ , ߪ௕,௫௭ Bending Stress
ߪ௕,௬ , ߪ௕,௭ Bending Stress about the member y and z-axis
ߪ௕ Bending Stress
ߪ௖ Axial Compressive Stress
ߪ௖௥௜௧ Critical Stress
ߪௗ (‫)ݐ‬ Standard Deviation of Deterioration Model
ߪ௠ ௔௫ Maximum Stress
ߪ௣௘௥ Allowable Stress)
ߪ௤ Compressive Axial Strength due to capped end hydrostatic action
ߪ௧ Axial Tensile Stress
ߪ௫,ாௗ Design Value of the Longitudinal Stress at the point of
consideration
ߪ௫,ாௗ Design value of the longitudinal stress due to moment and axial
force
ߪ௫ Standard Deviation
ߪ௭,ாௗ Design Value of the Transverse Stress at the point of
consideration
߬ாௗ Design Value of the Shear Stress at the point of consideration
߬௕ Maximum Beam Shear Stress
߬௖ Coating Life
߬௠ ௔௫ Maximum Shear Stress
߬௧ Torsional Shear Stress
߮ Resistance factor, as specified for each case of loading
߮஼ Resistance factor for compression
߮௕ Resistance Factor for Bending Strength
߮௕ Resistance factor for flexure
߮௕ Resistance factor for hoop buckling strength
߮௖ Resistance Factor for Axial Compressive Strength
߮௜(‫)ݔ‬ Base Functions
߮௧ Resistance factor for tension (as described above)
߮௧ Resistance Factor for Axial Tensile strength
߮௩ Resistance Factor for Member Shear Strength
xx
߮௫(‫)ݔ‬ Joint Probability Density Function
߯ Reduction factor for the relevant buckling mode
߯௅் Reduction factor for lateral-torsional buckling.
߰ Eigen-functions
߰ Reduction Factor
߰ Reduction Factor
߱ Angular Frequency
Abbreviations
‫ܥܵܫܣ‬ American Institute of Steel Construction

‫ܵܮܣ‬ Accidental Limit State
‫ܫܵܰܣ‬ American National Standards Institute
‫ܫܲܣ‬ American Petroleum Institute
‫ܣܵܧܴܣ‬ Adaptive Reliability Estimation Response Surface Algorithm
‫ܯܥܤ‬ Brittle Coulomb-Mohr
‫ܸܱܥ‬ Coefficient of Variation
‫ܵܲܥ‬ Corrosion Protection System
‫ܦܱܶܥ‬ Crack Tip Opening Displacement
‫ܯܥܦ‬ Ductile Coulomb-Mohr
‫ܧܦ‬ Distortion Energy
‫ܸܰܦ‬ Det Norske Veritas
‫ܴܶܦ‬ Damage Tolerance Ratio
‫ܨܦܧ‬ Empirical Distribution Function
‫ܰܧ‬ Euro Norm
‫ܣܧܨ‬/‫ܯܧܨ‬ Finite Element Analysis/Methods
‫ܵܮܨ‬ Fatigue Limit State
‫ܯܴܱܨ‬ First Order Reliability Method
‫ܨܴܮܪ‬ Hasofer Lind – Rachwitz Fiessler
‫ܱܵܫ‬ International Standardization Organization
‫ܵܪܮ‬ Latin Hypercube Simulation
‫ܦܨܴܮ‬ Load Resistance Factor Design
‫ܯܵܮ‬ Least Square Method
‫ܵܥ ܯ‬ Monte Carlo Simulation
‫ܯܯ‬ Modified Mohr
‫ܵܰ ܯ‬ Maximum Normal Stress
‫ܲܲ ܯ‬ Most Probable failure Point
‫ܵܵ ܯ‬ Maximum Shear Stress
‫ܯܱܵܨܸ ܯ‬ Mean Value First Order Second Moment
ܰ‫ܣܥܣ‬ National Association of Corrosion Engineers
ܱܴ‫ܣ‬ Organizational Reliability Analysis
ܲ‫ܧܥ‬ Polynomial Chaos Expansion
ܴܳ‫ܣ‬ Quantitative Risk Analysis
ܴ‫ܨܫ‬ Residual Resistance Factor
ܴܵ‫ܦܣܧ‬ Response Surface with Adaptive Experimental Design
xxi
ܴܴܵ Reserve Strength Ratio
ܵ‫ܵܮ‬ Servicability Limit State
ܱܴܵ‫ܯ‬ Second Order Reliability Method
ܴܵ‫ܣ‬ Structural Reliability Analysis
ܴܵܵ‫ܯ‬ Stochastic Response Surface Method
ܵܵ‫ܧ‬ Error Sum of Squares
ܵܵ‫ܯܧܨ‬ Spectral Stochastic Finite Element Method
ܴܵܵ Regression Sum of Squares
ܵܵܶ Total Sum of Squares
ܹܵ ‫ܮ‬ Still Water Level
ܶ‫(ܣܰܣ‬2) Two point Adaptive Nonlinear Approximations
ܶ‫ܲܮ‬ Tension Leg Platform
ܷ‫ܵܮ‬ Ultimate Limit State
ܹ ܵ‫ܦ‬ Working Stress Design
xxii
1 INTRODUCTION – CONTEXT OF STRUCTURAL
SAFETY
1.1 Introduction
Current evolution in engineering practices has allowed more critical and

complicated structures to be designed with greater confidence than in the past.
In addition, the constantly increasing cost of construction materials, due to
higher demand over the last years, indicates that design optimization can
provide competitive structures, compromising performance characteristics for
more cost effective designs. Structural Reliability has been established as a
valuable design tool for the description of the performance of structures;
therefore and towards this purpose, it can serve as a design restriction that will
ensure derivation of preferable structures which comply to minimum safety
requirements. The requirements set above, indicate the demand of a more
systematic assessment of the uncertainties of the basic design variables; the
functional and environmental loads, geometrical and model parameters, as well
as material properties.
Cases with limited levels of randomness can be treated deterministically,

applying a magnification factor on the loading or a reduction on the capacity
modelling, to account cumulatively for the effect of uncertainties. This
simplification in the design process produces most of the times oversized
designs without providing accurate information on the service life performance
of the structure, or ensuring adequate levels of structural safety. In contrast,
when the level of uncertainty is high, a stochastic approach of the design
variables seems essential. Following this approach, statistical representation of
the design parameters will provide the response of the structural member or
system in a stochastic way, allowing a better understanding of its service life
performance.
This contribution will focus on the reliability assessment of offshore structures,

and particularly frame-type jacket structures due to the important environmental
1
phenomena they have to withstand throughout their service life and the
complicated failure mechanisms that they form. This Thesis will report the
development of the Stochastic Response Surface Method (SRSM) that allows
reliability assessment of a structure, in the form of individual design blocks. This
methodology, contrary to complicated Stochastic Finite Element Analysis
Methods that demand deep knowledge of the mathematical background in order
to express explicitly uncertainty within the analysis, treats the simulation
procedure as a ‘black box’, extracting the response of the structure and later
processing it, within an individual reliability assessment routine.
The methodology that is proposed allows reliability assessment of structural

components through a sequence of individual steps that permit calculation of
reliability, based on easy to program procedures. After execution of a finite
series of simulations, the response of each member can be identified and a
quadratic polynomial response surface can be formulated based on the values
of the limit states that are examined through data regression analysis. Later, a
separate routine can account for the estimation of the reliability index based on
one of the available numerical techniques. Those discrete steps, which may
employ different tools and procedures for each task, can handle several
problems that are difficult to be modelled in one unified simulation code. For the
problem of the design of offshore structures that has been studied, reliability
analysis took place using the specialized software DNV SESAM for a confident
representation of the response of structural members through appropriate
modelling of the environmental loads acting on the structure. Following this
methodology, specialized commercial tools for different applications may be
employed for the probabilistic assessment of several engineering problems.
Figure 1, initially presents this procedure, which will be developed in the next
chapters.
The First Chapter of this Thesis presents the background of structural reliability
and the context of structural safety. Evolution of design methods and design
standards is discussed. Target reliability requirements are included, based on
2
requirements of regulatory bodies and classification societies, in order to set the
general requirements of a structural reliability assessment.
Identification of variables Chapter 4
Construction of Design Matrix Par. 3.3.1
Execution of FEA Simulations

Par. 3.5.2.1
Post-Processing of Results
Formulation of Limit States Values Chapter 5&6
Data Regression Analysis Par. 3.4
Reliability Index Calculation Par. 3.2
Figure 1: Response Surface Method Flow Chart
The Second Chapter presents a review of reliability assessment of steel and

offshore structures. Selection of the appropriate type of response analysis and
consideration of the integration from component to system reliability is
discussed. Finally, a comprehensive review of the Stochastic Methods for
reliability assessment will provide the background for the development of the
methodology that will be applied later in this contribution.
The Third Chapter includes the numerical procedures for the computation of
structural reliability. Deterministic methods, First and Second Order Reliability
Method (FORM/SORM), as well as Simulation Methods will be presented and
the background to the codes that have been written for the scope of this Thesis
will be set. A review of the Stochastic Response Surface Method (SRSM) and
the methodology that is introduced and will be applied in the later chapters are
analytically discussed. Finally Regression techniques are included as
fundamentals of the (weighted) regression analysis that is applied in the SRSM,
and a variation of the conventional least square method (LSM) will be proposed
for more accurate regression and prevention form ill conditioned systems of
3
equations. Verification of the mathematical tools that are developed is included,
providing confidence on their later application.
In the Fourth Chapter, the environmental loading and capacity modelling of

Offshore Structures will be analytically presented. Modelling of wave loads
according to different wave theories and the correlation between significant
wave height and peak spectral period based on joint distribution of statistical
distributions is presented. Wind and current modelling, are also covered. An
extensive review of literature data for material properties and methodologies for
the consideration of corrosion is included. Geotechnical data regarding piling
properties are also presented.
Once the stochastic variables and a reference structure have been identified, in
the Fifth Chapter the reliability of structural tubular members of a typical
offshore structure is assessed for ultimate strength under different combinations
of stochastic and deterministic loads. Limit states are formulated based on
fundamental failure criteria of ductile structural members, in order to
comprehensively represent multiple load actions that they are subjected to. A
sensitivity analysis of the design parameters (surface roughness, variables’
modelling, corrosion deterioration etc) is also included. System reliability
integration, based on the reliability performance of the structural members and
the failure mechanism of the structure, is also discussed.
The Sixth Chapter introduces the most widely used standards for the design of
offshore structures API RP 2A [1], [2], and ISO 19902 [3]. Further, ANSI/AISC
360-05 [4], and EUROCODE 3 [5], are included as generic codes of the design
of steel structures with strong probabilistic background. For the same reference
case structure, limit state functions will be examined based on their individual
design requirements and the minimum reliability indices of members are derived
and later compared to the ones obtained by the analytical limit states, using the
same calculation procedure.
4
The final chapter of this Thesis, Chapter Seven, gathers the conclusions of this
endeavour and proposes some future work towards a completely robust design
of steel structures.
1.2 Structural Reliability
In [6], the most complete definition of reliability, which has been adopted with
minor alterations from most current design standards, summarizes that
“reliability is the ability of a structure to comply with given requirements under
specific conditions during the intended life for which it was designed”. In this
definition, the important elements of design requirements, service life period,
and design conditions are included. A more practical, direct approach defines
reliability of a structure as the opposite of its probability to fail.
Reliability is the entity that will compromise the two requirements of design;
structural integrity and economy. In [7], a fundamental requirement for design
states that “a structure should be designed and executed in such a way that it
will, during its intended life time with appropriate degrees of reliability and in an
economic way sustain all actions and influences likely to occur during execution
and use and remain fit for the use for which it is required”. Structures can be
designed to have nearly zero probability to fail. Absolute no failure is an
oversimplification and can never be achieved because every forthcoming event
cannot be realistically predicted. Therefore, failures are accepted up to a level
that all parties involved in the design and operation of the structure will agree.
Structural reliability works on the prediction of the probability of exceedance of

the structural restrictions imposed by the design requirements at any stage of its
service life. The probability of occurrence of such an event is directly correlated
to its reliability, and once this is derived, design alterations can be identified, in
order to either improve structural reliability, or optimize already adequate
designs. Techniques of structural reliability, following computational resources
and numerical methods evolution, can be applied in wider multidisciplinary
5
design environments, considering the joint effect of multiple uncertain variables.
Practice can verify that structures that have been designed deterministically,
neglecting analytical modelling of uncertainty in variables, can have a greater
probability of failure compared to less expensive structures of similar service
that have been designed following a stochastic approach.
Within different application fields, different methods for reliability assessment

can be distinguished including Quantitative Risk Analysis (QRA), Structural
Reliability Analysis (SRA) and Organizational Reliability Analysis (ORA), which
is not applicable in the case of structural design. In order to better understand
the use of Structural Reliability Analysis, it is important to consider its interaction
with other disciplines.
Structural Reliability Analysis is related to the estimation of the probability of

failure of a structure or a structural member for given loading conditions.
Numerical implementation should be based on the most up-to-date modelling
techniques. Combination of fundamentals of structural reliability and modelling
techniques can allow calibration of design standards as will be discussed in a
later section. Quantitative Risk Assessment relates to the evaluation of the
overall risk of potential failure to humans, safety, environment and assets. The
main steps should include the following, according to [8]:
 Identification of hazards
 Assessments of frequency of initiating events
 Accident development
 Consequence Assessment
 Calculation of risks
A classification between probability concepts, distinguishes frequentistic to

Bayesian Probability. The first refers to the statistical interpretation of the
outcomes of stochastic experiments and its approach to probability can be
adequately predicted when the number of experimental iterations is increased
respectively. Bayesian probability, proposed by Bayes [9], is considered
subjective as it is based on the knowledge of each individual decision maker
6
rather than in the outcome of a repeated experiment and it is considered as an
expression of a “degree of belief” [10], [11]. Although in literature [12] the two
concepts are considered fundamentally different, both of them satisfy the
mathematical theory of probability. Structural Reliability Assessment follows the
Bayesian approach for the definition of probability since, although it uses data
from experiments or experience of existing structures, the basic scope is to
mathematically estimate the value of reliability. Presence of uncertainty in
materials, fabrication defects, effects of loading etc, constitutes each structure
as a unique outcome of the ‘design experiment’.
Statistical modelling of uncertainties is analytically studied in [13], while in [14] a

classification of uncertainty in structural design distinguishes the following types
of uncertainty:
 Physical (intrinsic or inherent) uncertainty describes the natural

randomness of a quantity. Typical examples of this type are the yield
stress affected by production variability (manufacturing defects) or the
variability in the wave and wind loading.
 Measurement uncertainty which is caused by errors in instruments or
instrumental configurations and sample disturbance due to external
factors (eg. ‘noise’ in experimental measurements).
 Statistical uncertainty which occurs due to inadequate data or information
such as a limited number of samples.
 Model uncertainty due to imperfections and idealizations made in the
physical model, formulations for load and resistance variables as well as
the allocation of statistic distribution to the main variables.
Structural Reliability Methods are classified according to their Level, Moment

and Order:
 Level, refers to the extent of information that the reliability problem

incorporates.
 Moment, refers to the order of statistical moments applied to better
represent the stochastic nature of an uncertain variable.
7
 Order, refers to the polynomial order used for the approximation of the
limit state surface.
From the above attributes, that of Level describes the development of reliability
methods [14]:
 Level I methods are deterministic reliability methods that only use one
characteristic value to describe each uncertain variable. Common design
standard formats, load-resistance and allowable stress, belong to this
category. Those methods correspond to standard deterministic design
methods. They can be combined with more advanced, higher level
methods in the case of partial safety factors calibration, which can
optimize the application of those methods.
 Level II methods use two values for the representation of each uncertain
variable (eg. the mean and the variance) and a supplementary measure
of the correlation between the variables (eg. covariance). Reliability
determined following such methods can be geometrically interpreted as a
relative distance from the mean value.
 Level III methods introduce the joint probability distribution of the sum of
the uncertain variables, calculating directly the probability of failure for a
limit state function. Advanced mathematical techniques such as
numerical integration, approximate analytical methods, including the First
and Second Order Reliability Methods, and simulation methods, such as
the Monte Carlo Simulation and the Directional Sampling methods,
belong to this category.
 Level IV reliability methods, are the most advanced, introducing the
element of target cost to the principles of engineering in order to derive a
technically feasible and at the same time economically optimized
solution. These methods can set an acceptable target reliability level for
the application of Level III methods.
8
1.3 Overview of Reliability Analysis of Offshore Structures
Offshore structural reliability analysis becomes of particular importance recently

considering numerous changes within offshore industry. Introduction of the
Load Resistance Factor Design format in standards has significantly contributed
to a more systematic design of offshore structures. Further, the introduction of a
‘goal setting regime’ [15] which stands as a requirement as well as a target
restriction in the structural design, allows more flexibility in the procedure of the
design of offshore structures. This has resulted to the establishment of basic
guidelines for a thorough reliability based design.
The above changes combined with the increasing need for a better
understanding of the performance of structures throughout their service life in
aspects of inspection, maintenance and reliability have created a wider
acceptance framework for reliability assessment methods. More accurate
modeling techniques and tools together with a higher available computational
capacity, allow for analytical assessment of the reliability evaluation of
structures in different stages and under different loading and capacity
conditions.
Structural reliability analysis can provide significant benefit to the potential

safety and cost; however the level of confidence of a reliability assessment
strongly depends on the uncertainty consideration, accuracy of modeling and
simplifying assumptions made as it has already been mentioned in Chapter
One. Structural problems, and even more extensively in offshore environments,
are in most cases non-deterministic, with limited information and knowledge in
both the conceptual and the design phase. Therefore, risk quantification yields
for stochastic (random) variables to be considered.
From the variables considered, the main concern is the environmental loads
due to its great randomness. In Chapter Four, a probabilistic consideration of
the joint environmental load will be presented. Extensive literature is available
for the meteocean conditions mainly for the Gulf of Mexico and the North Sea
sites, since those areas serve as a baseline for the development of design
9
standards. Application of those data in different regions might provide
inaccurate results, and calibration of the statistical properties of variables to
specific areas is essential. Stochastic modeling of capacity of structures and
structural members is less significant, as shown in [16], affecting the reliability
results in a lower scale. This phenomenon is also illustrated by design
standards consideration for partial safety factors, where those that account for
loads are significantly higher than those for material properties.
During the last decades, momentous developments have occurred in the

methodology as well as the tools for calculation of structural reliability. In a
component level, methods such as the First and Second order reliability
methods (FORM/SORM) have been widely used, proposing modifications to
account for complicated formulations of limit state functions and transformation
of complicated statistical distributions to a normalized ‫ݑ‬-space. Further,
simulation techniques, such as the widely known Monte Carlo Simulation, have
been introduced overcoming limitations of the deterministic techniques. Chapter
Three that follows presents the theory and the solution algorithms of those
methods. In a structural system reliability analysis level, different methods are
used from a component-based approach to a fully probabilistic analysis where
advanced mathematical and computational effort is required. In [17], the well-
known theory for system reliability assessment of the “branch and bound”
method is described.
Based on the context presented so far the research problems that will be
covered within this Thesis are summarized as follows:
I. Development of a methodology for the reliability assessment of a

complex frame type structure, using sequential steps of high capabilities
tools and widely used numerical techniques.
II. Provide analytical guidelines for the modeling of stochastic variables for
environmental loading and capacity of offshore jacket structures.
III. For a reference structure, identify critical members and failure paths that
will result to global failure of the structure and examine the sensitivity of
the derived values of reliability to significant design parameters.
10
IV. For the same reference structure, evaluate the reliability performance of
different design standards.
1.4 Design of Offshore Structures
Development of our knowledge of structures and their behaviour, and the tools
that are available, have advanced design procedures and methodologies
applied on the design of structures. In this section, a classification of design
methods will be presented, based on the way uncertainties are treated, and with
a view for classification of design standards to follow.
1.4.1 Permissible stresses
This is the first widely accepted approach to systematic design, also noted as
“allowable (working) stress method”. It is in line with the linear elastic theory.
The condition that the design should satisfy is:
ఙ೎ೝ೔೟
ߪ௠ ௔௫ < ߪ௣௘௥ or ߪ௠ ௔௫ < (1-1)
௞
The coefficient ݇, also noted SF by safety factor, is the only explicit measure
considered to account for all types of uncertainties. The maximum acting stress
on the structure should not exceed the critical value of the materials divided by
the coefficient k. The expressions above refer to a local effect on the structure
and therefore the comparison should be applied at its most exposed locations
(maximum stress). The basic principle of this method does not allow any
treatment of non linearity, stress distribution, ductility of materials and structural
members [18].
The inability of the method to consider analytically all the imposed uncertainties
of variables and models, the strict consideration of linear performance as well
as the fact that it does not consider combinations of loads, impose a great factor
of conservativeness in the design outcome. Different actions that should be
examined as well as use of different materials can produce different and
unreliable results.
11
1.4.2 Global Safety Factor
The method of global safety factors is based on the relation between the mean
values of the structural resistance ܴ and the load effects ‫ܧ‬. The ratio of the two
specifies the quantity of the global safety factor.
ܴ
‫=ݏ‬ > ‫ݏ‬଴ (1-2)
‫ܧ‬
The value of ‫ݏ‬଴ should be defined and is the target that the designer should aim
to meet. Contrary to the permissible stresses method, this method takes into
account the structural behaviour of members and their cross sections,
geometric non linearity, stress distribution and ductility in the individual
calculation of R and S.
The disadvantage of the method is that although R and S are calculated in a

more scientific method, the only explicit value is that of the global safety factor
‫ݏ‬. In addition, no special consideration of uncertainties of modeling is made and
therefore the results are forced to become conservative to account for any
unpredicted and unfavorable events. Combinations of loads and use of different
materials still cannot be handled.
1.4.3 Partial Safety factor
The method of Partial Safety Factors is the most up-to-date used method to the
establishment of design methodologies. It is also called ‘Limit State Method’
because it is applied in parallel with the concepts of limit states design for
different design conditions. The method is advanced considering that it gives
potential for mathematical optimization in several aspects. It can be
summarized as follows:
‫ܧ‬ௗ (‫ܨ‬ௗ , ݂ௗ , ܽௗ , ߠௗ ) < ܴௗ (‫ܨ‬ௗ , ݂ௗ , ܽௗ , ߠௗ ) (1-3)
Where: ‫ܧ‬ௗ and ܴௗ represent the design values of actions effects and resistance
respectively, ‫ܨ‬ௗ = ߰ ∙ ߛி ∙ ‫ܨ‬௞ design values of variables describing the actions,
௙
݂ௗ = ఊ ೖ describes the material properties, ܽௗ describes the geometrical
೘
12
uncertainties, and ߠௗ the model uncertainties. The design values derive from
the corresponding characteristic values of the variables (‫ܨ‬௞, ݂௞, ܽ௞, ߠ௞ ), applying
the required partial factors γ, reduction factors ψ and any other specified factor,
which are the control values of the reliability of the design. It is obvious that this
method compared to the previous ones is the most analytical one, making the
fewest simplifying assumptions. Use of the method, allows handling of load
combinations as well as multiple different materials.
1.4.4 Probabilistic Methods
The Probabilistic Design Methods are the most advanced that have been
proposed. Their basic requirement is that during the service life of a structure
the probability of failure does not exceed an acceptable design value. This can
be expressed as:
ܲ௙ ≤ ܲௗ or ߚ > ߚௗ (1-4)
The above two expressions are equivalent. The design values that should be
fulfilled can be determined by the specifications of the structure. Those methods
are not widely used yet and undergo some controversy due to their increased
complexity. However, they can achieve optimized results leading to lighter and
more economically efficient structures.
In addition to the Partial Safety Factor method, concepts of probabilistic

analysis can be used to optimize the values of the partial safety factors. This
procedure is noted as calibration and can be found on the Annexes or guiding
material of the recently introduced modern structural codes [19]. Probabilistic
design methods are very important for the design of special cases or novel
structures where previous experience does not exist and application of
accepted methodologies is not applicable.
1.4.5 Comments on design methods
The design methods as presented above, starting from the permissible loads
method and heading to the fully or partially probabilistic methods become more
13
complicated, demanding greater engineering and mathematical skills. However
towards the same direction, the level of conservativeness and therefore the
over sizing of the structures is reduced leading to more efficient structures with
a better understanding of their service life performance.
Modern Standards and Codes follow the Partial Safety factor methods, giving
the engineer the ability for further optimization. This trend has been
countersigned by the fact that the most widely used standard for offshore
platforms API RP-2A – WSD (Working Stress Design) [1], has been also
published from 1993 in a Partial Factor format API RP-2A – LRFD (Load
Resistance Factor Design) [2].
1.5 Limit State Design
Since the common trend in modern design is the design according to limit
states, this will be presented more analytically in the following sections. The
general design requirement is to provide structures with adequate safety
margins in order to account for all types of uncertainties affecting its integrity
(load and capacity variability, modelling idealizations etc). A simplistic definition
of limit state design indicates that the demand (load) of a structural system
should under no conditions exceed its capacity (resistance). Considering a case
of multiple loading, the safe region criterion should be expressed as:
‫ܦ‬ௗ = ߛ௢ ෍ ‫ܦ‬௞௜(‫ܨ‬௞௜, ߛ௞௜) < ‫ܥ‬ௗ = ‫ܥ‬௞/ߛெ (1-5)

௜
In the above expression, index k represents the characteristic value of a load

or resistance variable while index ݀ the design values that incorporates the
required magnification or reduction to account for consideration of uncertainties.
Load variables are magnified with load factors ߛ௞௜ in order to account for
unforeseen events, while the capacity is diminished with the material factor ߛெ in
order to account for capacity uncertainties (material properties, quality of
construction, corrosion etc). A further partial safety factor ߛ௢ is added to
14
consider the seriousness of the examined limit state to the integrity of the
structure.
The characteristic (nominal) value of a variable is defined by its statistical

properties. For a capacity variable, it can be based on the lower bound or 95%
excedance value, while for a load variable, the characteristic value on an upper
bound or a 5% excedance value. Derivation of partial safety factors is based on
either previous experience or through a rigorous procedure that provide
acceptable levels of safety and performance. In the previous methodology of
allowable (or working) stress design the basic concept was to make sure that
the response of the structure due to loads acting on it will remain below specific
levels throughout the service life of the structure. The limit state design
approach systematically examines the response of the structure under various
conditions it might have to withstand, as a combination of loads and capacity.
For offshore and marine structures, several limit states are proposed by
regulatory bodies and classification societies that should be examined within a
comprehensive design. Table 1, adopted by [20], presents some common limit
states for the four main types of limit states that should be considered:
 Serviceability limit state (SLS)

 Ultimate limit state (ULS)
 Fatigue limit state (FLS)
 Accidental limit state (ALS)
1.5.1 Serviceability Limit State
This type of limit states, refers to conditions where, due to extensive

deformation, vibration or noise, the structure’s functionality is influenced. The
factors mentioned are in many cases correlated. Criteria that have been
established based on practice experience of the functionality of the structure are
expressed in the form of maximum allowable deflection [5], or similar
restrictions that should be fulfilled in order for the structure to operate without
requirements of maintenance or further intervention. Buckling phenomena are
15
often incorporated in order to control the behaviour of the structure and prevent
cases of extensive deflection.
Loss of structural resistance (excessive yielding and buckling)
Failure of components due to brittle fracture
Loss of static equilibrium of the structure, or of a part of the structure,

considered as a rigid body, e.g. overturning or capsizing
Ultimate limit
states (ULS)
Failure of critical components of the structure caused by exceeding the
ultimate resistance (in some cases reduced by repeated loads) or the
ultimate deformation of the components
Transformation of the structure into a mechanism (collapse or

excessive deformation).
Fatigue limit
states (FLS) Cumulative damage due to repeated loads.
Structural damage caused by accidental loads
Accidental Ultimate resistance of damaged structures

limit states
(ALS) Maintain structural integrity after local damage or flooding
Loss of station keeping (free drifting).
Deflections that may alter the effect of the acting forces
Deformations that may change the distribution of loads between

supported rigid objects and the supporting structure
Serviceability
limit states Excessive vibrations producing discomfort or affecting non-structural
(SLS) components
Motion that exceed the limitation of equipment
Temperature induced deformations
Table 1: Examples of Limit States according to DNV
1.5.2 Ultimate Limit State
In many cases, this is the most important limit state that should be considered
since it checks the ability of the structure to resist plastic collapse or reach to
ultimate strength. For an analytical description of a structural member, its post
buckling behaviour should be considered in order to avoid additional
conservatism on the strength of the element. Figure 2, illustrates this approach
16
(point B) compared to a more traditional approach (point A) where the point of
elastic buckling determines the strength of the element.
Figure 2: Structural Design Consideration of the ULS
The safety margin of such members cannot be directly evaluated. However, the
practice trend is to design structures for ultimate strength. In cases of structures
that have suffered any type of damage, a new ultimate strength should be
calculated, based on the deterioration of their capacity. Structural design should
also take into account the response of the structure in the case of a potential
failure. The structure should be designed to fail in a ductile rather than brittle
manner in order to allow progressive collapse, by redistribution of stresses in
alternative load paths, rather than failing suddenly without providing any
warning and therefore potential of intervention. Ductility in a design can be
facilitated by design techniques such as avoidance of high stress
concentrations, weld defects, allowing some level of plastic deformations etc.
The design problems that are studied in this Thesis mostly deal with ultimate
limit states, so in the later sections, several issues as well as design standards
provisions will be discussed in greater depth.
17
1.5.3 Fatigue Limit State
Fatigue Limit State is particularly important in structures that undergo significant

cyclic phenomena. Large offshore structures, such as wind turbines are
designed for their fatigue life in addition to ultimate strength. Offshore structures
have a long service period that might exceed 20 or even 40 years. This, in
conjunction with the inspection intervals, affects the reliability requirements of
the structural design. The effect of fatigue is preliminarily a local effect,
concerning welded joints and areas of stress concentrations.
The Fatigue Limit State criteria are based on a cumulative fatigue damage of a
structure under repeated fluctuation of loading. The fatigue damage at a crack
initiation is affected by many factors such as the stress ranges experienced
during load cycles, local stress concentration characteristics and the number of
stress range cycles.
1.5.4 Accidental Limit State
This category of limit states aims to limit the consequences in a case of a

failure, such as avoidance of loss of life and assets, environmental pollution and
financial losses. The criteria that should be satisfied are based on accidental
scenarios and associated performance that should be decided upon analytical
risk assessment. Progressive collapse in case of failure, impact, excessive
loads due to human error or machinery failure, explosions etc are some of the
scenarios that should be addressed. In order to have an economically efficient
design, a trade off should be made between ultimate safety and prevention
costs, setting realistic survivability consequences.
For Accidental limit state design, the integrity of the structure should be
assessed initially in a global level (accident events) and later in a post-accident
assessment to account for the real impact on the structure. In the case of an
impact on a jacket structure for example the major requirement is to sustain its
stability avoiding total collapse. Once this has been evaluated, the remaining
18
capacity of the structure should be assessed in order to ensure that it will
remain functional after such instance.
1.6 Design Standards for Steel Structures
1.6.1 General
Historical development has provided design methods with empirical,

experimental and theoretical knowledge of mechanics and probabilistic
concepts. The systematic recording of this knowledge under a sound scientific
foundation can derive a methodology that will allow design of specific structures
in a way that defined levels of reliability may be obtained. The analytical
documentation of this methodology can help in the composition of structural
standards and design codes. In the following section, after a brief classification,
standards applied in the design of offshore structures will be introduced.
In [21], the background principles of modern offshore standards and guidelines

are summarized as follows:
i. Design criteria are formulated in terms of limit states

ii. Semi-Probabilistic Methods for ultimate limit strength design, have been
calibrated by reliability or risk analysis methodologies
iii. Fatigue design checks depending on consequences of failure and
access for inspection
iv. Explicit accidental collapse design criteria to achieve damage tolerance
for the system
v. Consideration of loads that include payload; wave, current and wind
loads, ice, earthquake loads as well as accidental loads (fires,
explosions, ship impacts)
vi. Global and local structural analysis by finite element methods for ultimate
strength and fatigue design checks
vii. Nonlinear analyses to demonstrate damage tolerance in view of
inspection planning and progressive failure due to accidental damage
19
The elements referred above, are followed in order to set safety requirements to
avoid ultimate consequences such as fatalities, environmental or property
damages. The corresponding regulatory regimes set different acceptance
criteria for these consequences as it will be discussed later.
1.6.2 Categorization of Design Standards
Following the categorization of the design methods that have been presented
so far, the available structural Codes and Standards are distinguished as either
allowable (working) stress or limit state design.
The allowable stress codes, consider that the stress under the maximum
loading conditions should not exceed the material yield or ultimate strength
divided by an appropriate safety factor. Typical values for the safety factor can
be in order of magnitude of 1.5 for yield strength or 2.5-3.0 for the ultimate
strength [22]. As already discussed, the explicit calculation of the safety factor is
a very demanding task in order to incorporate all sources of uncertainty that can
impose a considerable degree of conservativeness in the resultant design.
In the limit state design, the structure is designed to resist specific loading
conditions described in each corresponding limit state. Commonly, the loads are
multiplied by partial safety factors, the resistance divided by safety factors and
combinations of loads are considered. Application of those standards can be
more complicated involving several aspects of decision making from the
engineer but will generally produce more favourable results.
1.6.3 Standards for Offshore Structures
1.6.3.1 API RP-2A: Recommended practice for planning, designing and

constructing fixed offshore platforms WSD/LRFD
This recommended practice was introduced by the American Petroleum

Industry in 1969. It is based on sound engineering principles, extensive testing
and field application experience. In its initial publication it followed the working
design stress format. In 1989, a draft was issue in a Load Resistance Factor
20
Design format that was publicly released in 1993, based on the fact that two
decades of application of the WSD edition of the standard had provided
structures with sufficient reliability performance [7]. This standard is based on
data for the Gulf of Mexico environmental conditions; however complementary
documentation is available for different regions [23].
1.6.3.2 ISO 19902:2002: Petroleum and natural gas industries-general

requirements for offshore structures
The publication of the ISO 19902 [3] offshore structure standard for Fixed Steel
structures represents the culmination of significant efforts over many years.
Providing guidance and procedures for the design and fabrication of offshore
fixed steel structures, the Standard has been developed based on established
standards for fixed offshore steel structures and through direct input from many
of the countries actively engaged in the development of the offshore, including
the United States, United Kingdom, France, Norway, Canada etc. ISO
Standards are been adopted by several countries as National Standards,
introducing them as legal requirements that allow certification of structures from
world widely recognized certification bodies.
As for API, significant work has been carried for the applicability of the ISO
standard in different regions around the world [24]. This consideration, proposes
methodologies for custom derivation of load factors that lead to optimization of
resulting designs.
1.6.3.3 BS EN 1993-1-1:2005 Eurocode 3: Design of Steel Structures
EUROCODEs is a set of standards produced by the European Commission in

order to “establish a set of common technical rules for the design of buildings
and civil engineering works which will ultimate replace the differing rules in the
various Member States” [25]. EUROCODEs embody National experience and
research, presenting a world class standard for structural design. The standard
was formally released in 2007 and was subjected to a three years period until
conflicting National Standards were withdrawn. The verification procedure is
21
based on the limit state concept. One of the most important aspects of those
standards is that they allow for design based on probabilistic methods [26],
giving the opportunity for further design optimization.
EN 1993:2005 [5] refers to the design of Steel structures. Although it is not yet
used for the design of offshore structures, its highly scientific, probabilistic
background provides a systematic design methodology that, if used in
conjunction with established standards for this class of structures, may provide
more efficient, optimized designs.
1.6.3.4 ANSI/AISC 360-05: Specification for structural steel buildings
This Specification provides the generally applicable requirements for the design
and construction of structural steel buildings and other structures. The American
Institute of Steel Construction incorporates in a single document both allowable
stress design (ASD) and load and resistance factor design (LRFD) methods,
allowing design according to provisions of either method. Based on previous
experience and up-to-date technical knowledge, it aims to provide design
guidelines for commonly used steel-framed buildings and other, similar
structures.
1.6.3.5 Other Standards for Offshore Structures
Petroleum industry involved with offshore oil gas platforms and marine/naval
engineering have assisted crucially the process of systematic design of offshore
structures, due to their high demands on reliability and safety. Apart from the
standards that have been referred to so far, Lloyd’s published the “LRS Code
for Offshore Platforms” [27] in 1988 and recently in 2007, Germanischer Lloyd,
published the “Guideline IV – Industrial Services: Offshore Technology” [28].
Det Norske Veritas published “DnV: Rules for the classification of Offshore
Installations” [29] in 1989, and recently in 2008 the “Offshore Standard OS-
C101: Design of offshore steel structures, General – LRFD Method” [30].
Finally, the Norwegian petroleum industry has introduced the NORSOK
standards, which refer to ISO, EN 1993 and is currently on its 5th edition [31].
22
1.6.3.6 Limitations of Structural Standards
Although in general use of standards results in design of structures with

acceptable reliability, limitations arise for their application on novel and special
structures, due to the fact that they primarily refer to specific structures and are
presented in a high level that generally provides limited detail information and
guidance on the background of the methodology they follow [32]. In this aspect
the concept of reliability based design method can provide adequate results for
the design of novel structures.
Adopting the target reliability requirements from relevant standards, partial

safety factors can be calculated independently, avoiding unwanted
conservativeness imposed. Further, in areas of high uncertainty, design details
are approached in such a way that the consequences of failure can be reduced
(eg. structural redundancy, etc). The former can be realized by combination of
different standards where appropriate, resulting in solutions that provide a
reliable design that meets the specifications set.
During fabrication and service of the structure, safety elements can be

introduced such as quality control, alignment control, visual inspection,
instrumented monitoring and proof loading. Those practices provide information
about the structure, additional to those available at the design stage, reducing
the overall uncertainty. Once the manufacturing process is completed, a
structural integrity monitoring system can compare real data to ones initially
calculated, verifying the conditions of the structure. Data obtained, can provide,
throughout its service life, all the necessary information having good confidence
levels for life-cycle fitness-for-service assessments including cases following
unforeseen events such as local collision or component failure. Therefore,
current reliability can be calculated, identifying the actual condition of the
structure and indicating the actions that should be taken for any required
intervention as well as the ability of the structure to work above the initially
considered service life. Finally, the database that has been created, can provide
substantial information for relevant optimized future structures and systems.
23
1.7 Target – Acceptable Reliability
Structural design aims to develop structures that are able to perform

adequately, compromising cost and safety. The consequence of failure is an
important parameter that should be assessed in order to specify the potential of
injury or life loss, economic losses (direct and indirect), environmental pollution
etc. Quantification of consequences is a very difficult task affecting the outcome
of a reliability assessment [33]. Previous experience for a class of structures is
an essential guide for the determination of target reliability levels for similar
structures.
On the other hand, innovative structures cannot follow provisions of existing

standards that refer to different structures since loading behaviour and
consequences of failure strongly depend on their service, even if more
conservative loading factors are adopted. This practice would drive the cost of
the design without ensuring sufficient levels of reliability. One typical example of
this problem, refers to the design of offshore wind turbines with jacket type
foundation; although the general layout of the structure is similar to that of an
offshore oil and gas platform, the loads added due to the rotor, the fact that the
structures are unmanned and the large scale of production, constitute standards
that refer to the latter application unable to ensure sufficient levels of reliability.
For such cases, a robust reliability based design should be adopted [34] that
would allow from basis derivation of load factors applicable to each case.
In the determination of the target reliability of structures the following

fundamental parameters should be assessed. This is particularly important for
cases when different structures or concepts are compared:
 Interpretation of calculated reliability. This identifies derived reliability not

as a property of the structure, but as the level of our knowledge of its
performance. This means that the level of uncertainty modelling (quantity
and quality of information) will influence the results of the reliability
assessment.
24
 Reference Period. The reliability of a structure should be measured for a
specific period of time. This period is usually one year or its specified
service life. This is a very important parameter for the inspection and
maintenance requirements of the structure as well as for its
reassessment in cases when the initially planned service life is to be
exceeded and requalification assessment is required. Table 2, presents
an example of target reliability for different structures and different
reference periods.
 Scale of Reliability. Values of target reliability may be provided either in a
local-component level or on a global-structural system level. Limit states
may refer to the response of a member, but exceedance of this limit state
in the failure region might not lead to structural failure. Typical example
of this parameter is a local failure of a brace of a frame structure with
structural redundancy. In such a case, load paths should be identified in
a way that all possible failure mechanisms are identified and through
integration to system reliability the total reliability of the structure may be
derived.
 Structural modelling. The tools and techniques used for calculation of
reliability, both in estimating the response of the structure as well as in
the reliability calculation methods might affect the accuracy of the results.
 Stochastic modelling. This parameter refers to the statistical
representation of the stochastic variables. The more accurate the
statistical model the better uncertainties in modelling are considered.
Design according to standards can achieve minimum levels of target reliability.

Some standards clearly state the target reliability they aim for, such as the
EUROCODEs, while others rely on producing sufficient structures when the
provisions of the code are followed as close as possible. Obtaining a better
understanding of structures, may be update the reliability performance of
standards.
As it has been already mentioned, marine and offshore engineering has

provided valuable experience on the establishment of agreed target reliability
25
levels for different classes of structures. In [35], an analytical assessment on the
development of target reliability for offshore structures can be found. Evaluation
of the performance of ABS (American Bureau of Shipping) ‘Rules for ship
structures’ that was carried out in [36] predicts the reliability index achieved is
between 3.15-3.65. In a study [37] of API recommended practices for Offshore
Jacket structures (LRFD edition) estimates an annual average estimated
probability of failure in the order of 4x10-4 (ߚ~3.35). DNV [14] suggests
calibration of target reliabilities based on existing cases with previous
experience; however a very comprehensive table is proposed and adopted by
the DNV group of standards, distinguishing classes of consequence and
redundancy of a structure. Table 3 presents those recommended values of
acceptable annual probabilities of failure. ISO proposes a model that
incorporates the possibility of injury or fatality in the case of a collapse with a
mathematical correlation that accounts for the number of people at risk. A lower
boundary of 10-6 Probability (annual death per failure) is set, corresponding to a
ߚ~4.75). Evaluation of AISC, in studies performed in [38] and [39] for structural
members of a bridge, predict target reliability in the ultimate limit state (ߚ~3.5).
Finally, EUROCODEs, prescribe a target probability of failure of about ܲ௙ =
10ିସ [25]. From the figures that were summarized here, it can be observed that
in the more general standards, a more conservative approach of reliability is
observed. This fact is reasonable since they should account for different types
of uncertainty for application in different structures.
Evaluation of risk is a dynamic task based on information update and

experience gained from similar cases. Serious accidents such as the Piper
Alpha offshore platform accident [40], is an example of events that initiated
assessment of rules and codes for the avoidance of similar phenomena through
a knowledge-based approach. Some analytical methods have been proposed
for a systematic determination of risk and reliability [35]:
 A risk based approach, focuses in the quantification of consequences in

an absolute measuring unit, such as loss of lives per accident or cost of
26
failure in economic units [41], [42]. This method involves great
subjectivity on the valuation of human life loss.
 Life cycle cost analysis, is a method applicable especially in cases of
structures that can operate well with regular maintenance interventions.
It is based on the maximization of a cost function with the restriction of
minimized reliability [43], [44].
 Social tolerable risk for failure and fatalities. This method incorporates
the different reaction of societies to hazards and risks’ providing more
strict safety requirements in general, since the element of cost efficiency
becomes secondary [45]. In [46] and [47] mathematical expressions are
proposed for the calibration of the safety requirements according to the
social impact factor and the level of warning before failure. .
Generic standards, such as [14], [6] propose analytical methodologies for the
reliability assessment of structures. In [48] a procedure is described for this
purpose and has been adopted by many of the existing standards as a way to
achieve calibration of target reliability levels. Figure 3, illustrates in an iterative
block diagram format this risk assessment procedure. Further, Table 4
summarizes some target probabilities of failure for offshore applications based
on what has been discussed so far [35].
1.8 Summary
In this Chapter, the context of structural reliability has been set, incorporating
different sources of uncertainty in structural design. Classification of available
methods has been presented, based on the extent of information that the
problem treats, and the accuracy in the representation of the design variables.
Evolution of design methods from permissible (allowable) stresses to limit states
formats was also included, with a view for classification of design standards,
and a reference to the consideration of appropriate limit states for the latter
class. Finally, after identification of the most important available standards for
the design of offshore and steel structures, and some comments on the
27
limitations they involve, an investigation of the target reliability levels that are
documented from widely accepted certification bodies have been presented.
Reliability index β
Consequences for loss of
Examples of
Reliability human life, economical,
βd for building and civil
Classes social and environmental βa for
Td=50 engineering works
consequences Ta=1 year
year
Bridges, Public
3-high High 5.2 4.3
Buildings
Residential and
2-normal Medium 4.7 3.8
office buildings
Agricultural
1-low Low 4.2 3.3 buildings,
greenhouses
Table 2: Reliability Index and Reliability Classes [7]
Class of failure Consequence of failure
Less serious Serious
-3 -4
PF=10 PF=10
I - Redundant structure
(βt=3.09) (βt=3.71)
-4 -5
PF=10 PF=10
II - Significant warning before the occurrence
of failure in a non-redundant structure
(βt=3.71) (βt=4.26)
-5 -6
PF=10 PF=10
III - No warning before the occurrence of
failure in a non-redundant structure
(βt=4.26) (βt=4.75)
Table 3: Values of acceptable annual probabilities of failure (PF) [14]
28
Allowable system
Source
failure probability
-6
Risk Analysis (analytical Assessment) 10 /yr
-5
CSA 10 /yr
-6 -5
DNV 10 -10 /yr
-7
ISO-1000 people 10 /yr
-5
Professional recommendations 10 life time
-7 -5
Social Criteria 10 -10 /yr
-7 -5
Existing Structures 10 -10 /yr
Table 4: Comparative estimates of target Pf [35]
START
DEFINITION OF THE SYSTEM
HAZARD IDENTIFICATION
PROBABILITY ANALYSIS CONSEQUENCE ANALYSIS
RISK ESTIMATION
RISK EVALUATION
NO
RISK ESTIMATION ACCEPTABLE RISK
YES
STOP
Figure 3: Iterative procedure for Risk Assessment [48]
29
2 RELIABILITY ANALYSIS OF OFFSHORE
STRUCTURES
2.1 Basic formulation of the Problem
Following an initial, general approach, the behaviour of a structure can be

determined by the values of loads (actions) or load effects ‫ ܮ‬acting on it and its
load bearing capacity (resistance) ܴ. The following correlation between the two
variables can form the acceptance criterion of the structure for a specific failure
mode – limit state:
ோ
ܴ − ‫ > ܮ‬0 or >1 (2-1)
௅
The safety margin, of the structure can be expressed as:
ܼ=ܴ−‫ܮ‬ (2-2)
In practice, both resistance as well as loading effects, involve a number of

variables or material properties, subject to several sources of uncertainty. In the
critical case where the resistance and load values are equal, limit state
equations can be formed as:
ܼ(ܺ) = 0 (2-3)
In the case when ܼ(ܺ) ≥ 0 the structure operates in the safe region while when
ܼ(ܺ) < 0, it is considered in the failure region. For each limit state, the
probability of failure can be expressed as:
ܲ௙ = ܲ{ܼ(ܺ) < 0} (2-4)
Alternatively, considering probabilistic models for the assessment of the

variables ܺ = ൣܺଵ, ܺଶ,… , ܺ௡ ൧ and simplifying that they are described by time
independent joint probability density function ߮௫(‫)ݔ‬, the expression of the
probability of failure can be described with the integral:
31
ܲ௙ = න ߮௫(‫ݔ݀)ݔ‬ (2-5)
௓(௑)ழ଴
The expression above can be extended to become applicable to some cases of

time dependent quantities which can be transformed into time independent
ones [18]. For cases where this is not feasible, the process of calculating the
probability of failure becomes much more complicated and in practice should be
assisted by different numerical methods and software products [6].
Instead of using the term of “probability of failure”, the equivalent term of

reliability index ‘ߚ’ is usually referred to in the design standards and relevant
documentation. This is the negative value of the standardized normal variable,
corresponding to the probability of failure:
ߚ = −ߔ ௎ିଵ൫ܲ௙൯ (2-6)
Where, ߔ ௎ିଵ(ܲ௙), is the inverse standardized normal distribution function. The

benefit of using this notation is that ߚ can provide results for several types of
statistical distributions based on deterministic methods as it will be discussed in
Chapter Three. Figure 4, illustrates the correlation between the reliability index
and the probability of failure. Considering a simple case of a single load acting
on a member and its resistance, both distributed normally, the correlation of
their means, standard deviations and ߚ can be shown in Figure 5.
A quantitative definition of risk derives it as the product between the probability

of occurrence of an adverse event and its consequences [22]. The first is
influenced by the reliability of the structure while the latter of its function and
specifications. This implies that different values of risk can exist for different
combinations of these parameters. Unmanned offshore structures for example
can experience failures that halt the operation/production but without fatalities
and therefore the measured consequences can be considered lower; the same
risk can be achieved with designs of increased probability of failure but lower
consequences in the case of potential failure. Taking this into account, the
calculated reliability levels can be interpreted.
32
0
-1
Probability of failure (log) -2
-3
-4
-5
-6
-7
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Reliability Index
Figure 4: Relationship between β and Probability of failure
0.012
Load
Resistance
0.01
0.008
0.006
pdf
0.004
0.002
0
50 100 150 200 250 300 350 400 450 500 550
Load/Strength
Figure 5: Definition of Reliability Index
An interesting parameter of the reliability index which represents the relative

positions of the Load and Resistance distributions is its performance throughout
the operational life-cycle. The load effects may increase with time due to crack
growth etc, moving the corresponding curve to the right while the resistance
might decrease due to deterioration of fracture toughness or other age related
33
mechanisms. This fact leads to a decrease in the relative difference between
mean values, which reflects to a decrease in the reliability index and therefore
an increase in the probability of failure. The importance of this characteristic is
significant for the design process and the engineer should incorporate this time
dependent component into the design model in order to avoid unwanted
residual uncertainties in the calculations.
2.2 Background and motivation
2.2.1 Development of Structural Reliability Applications
Conventional deterministic structural design has been established successfully

during the past years in the design, construction, maintenance and inspection of
structures. Although, in general, it provides structures of sufficient performance,
uncertainties are considered in a generalized way incorporating a significant
degree of conservatism. Further, requalification of structures that have
exceeded their predetermined service life demands reassessment on a basis
that will ensure safety in further operation. Both of the above facts introduce
reliability analysis, providing the framework for effective decision making in
cases where uncertainties cannot be incorporated following a deterministic
approach.
An initially application of structural reliability, as it is presented in [49], can be

found back in 1926 by Mayer, in an attempt to quantify the safety of a structure,
treating uncertain variables by using the mean and standard deviation. Initial
applications of reliability theory are focused in the field of aerospace, electronics
and nuclear industry. Interest in structural reliability begins in 1947 when in a
study of Columbia University [50], safety margins and safety factors are
analyzed in a way that allows their systematic derivation. Evolution in
computation of reliability is traced in 1969 where the Mean Value First Order
Second Moment method (MVFOSM) was proposed [51]. This is an initial
approach to the derivation of the reliability index based on mean values of load
34
and resistance variables. The fact that this method was not taking into account
different formulations of limit state functions, initiated the Hasofer and Lind [52]
reliability index methodology to be formulated which was the first method to
geometrically approach the reliability index calculation as the minimum distance
of the failure surface from the origin of the normalized space. Methods for
handling non-normal random variables have been introduced and will be
discussed later in this Thesis. Following the development of FORM, SORM
managed to deal with more complicated limit state functions with multiple
minimization points and cases of non linear curves. In a very important review
paper [53], summarizing the general belief of the 1990 decade, has illustrated
that the development of deterministic methods had been completed. In addition,
competent simulation techniques have evolved, able to produce accurate
results at the expense of greater numerical effort.
Development of structural reliability is now moving towards structural

optimization schemes with reliability to stand as the basic design restriction. As
this is a relatively new practice, important issues of time performance should be
included, constituting this as a time-variant analysis [54]. Apart from this,
another very important issue is the transformation of the state-of-the-art into
state-of-practice methods in order to allow application in more essential
engineering problems.
2.2.2 Development of offshore industry
The first offshore drilling activities started in 1947, when the first platform was
built in Louisiana in the Gulf of Mexico. This was a very short structure of 5.1 m
height. During the last decades, deployment of offshore structures is moving in
deeper waters. Cognac platform [55] has been installed in a depth of 338 m,
followed by Shell’s Bullwinkle platform [56] standing in 415 m of water. An
economic sensitivity analysis, had set 300 m to be the restriction for fixed
offshore platforms however this number has been revised to 615 m.
Evolution of technology has come up with new configurations of offshore

structures in order to accommodate the harsh conditions of deeper waters. In
35
mid 1990s, Shell has installed a tension leg platform (Auger TLP) in depth of
872 m [57] setting the up-to-date record for any installed offshore structure. In
[58], a review of installed offshore platforms refers to a total of 5238 platforms,
15% of which are very heavy structures (more than 5000 tn). Due to the
depleting resources in the oil reserves, the fluctuation in prices of oil and the
world’s constantly increasing demand in resources, the challenge of exploration
in harsher environments is now active. For 2005, an estimate on deployed
structures considered approximately 1100 active rigs [59].
Design of oil and gas platforms, has stood as the basis for significant
developments in offshore wind industry. Deployment of wind turbines offshore
goes back to 1930, however it was only in 1990 when the first offshore unit (220
kW, 38 m hub height, 7 m water depth) was deployed in the North Sea. Similar
structures were deployed in Denmark at the same time, accommodating wind
turbines rated up to 500 kW each at depths of up to 6 m. Several projects were
completed up to 2000, increasing the capacities of the wind turbines supported,
the depths of deployment, and the distance from shore considering harsher
environmental conditions. 2000 can be considered a breakthrough for wind
industry, initiating England’s involvement in the field, deploying units of 2 MW
on the Blyth wind park (800 m from shore, up to 11 meters water depth). Water
depths of 20 m have been reached by 2002 in Denmark, moving even further
offshore (3.5 km). The state-of-the-art installed support structure so far, refers to
the Beatrice wind farm, built in 2007, consisting of 2x5 MW wind turbines in a
site 22 km from the Scottish coast, and 45 m of water depth [60]. The constant
demand of clean energy constitutes efficient design of deep water foundations
an essential task, considering that, contrary to oil and gas platforms, wind
turbine support structures are aimed to be built in massive production.
2.2.3 Application of Reliability Analysis in the offshore industry
Application of Reliability Methods appears to provide particular benefits in the

marine environment, due to the randomness and the restricted knowledge of
the phenomena structures resist. Proper modeling of the structures’ capacity
36
and loads encountered can predict its probability of failure through reliability
analysis. This analysis, for offshore structures, includes component and system
reliability analysis, specific design assessment and various forms of cost-benefit
optimization and maintenance planning. For a fixed offshore unit, the benefits
reliability analysis can provide can be summarized as follows [61]:
I. Achieve uniformity in the reliability at a component level

II. More efficient utilization of material properties, compared to over-sizing
followed by deterministic procedures
III. Account directly for randomness and uncertainties in engineering
parameters
Reliability assessment was introduced in the field of offshore structures in

1980s, following the ‘goal setting’ philosophy in the safety and economic
consideration [62]. Application of reliability methods for offshore structures was
initiated and motivated for ship hulls [63], [64] and was later applied in the
reliability of offshore structures. Important work is carried out in [65] and [66]
applying reliability analysis theory for the safety assessment of existing
structures, allowing efficient maintenance planning based on risk involved under
certain conditions of a deteriorating structure.
Reliability Analysis is also applicable in the optimization of structural design.

Setting a target level of reliability, structures can be designed compromising risk
and cost. In [67], a reliability-based design format for jacket platforms under
wave loads is presented by resizing members until the target reliability is
sufficiently approached. Reliability based design as well as design optimization
is in its infancy since several restrictions should be set for realistic designs to be
derived. However, moving towards more economic structures, this topic should
be investigated further.
The demands of reliability methods in mathematics skills, have led to the

development of reliability calculation software. PROBAN, developed by DNV, is
a general purpose reliability program able to analyze different failure modes
simultaneously, determine conditional probabilities, and derive partial safety
37
factors [68]. Further, STRUREL/COMREL [69] and RASOS are some common
software packages used for the reliability of offshore structures. Apart from
those standalone general application programs, incorporation of reliability
calculation procedures is attempted by well established FEA software (ANSYS,
SOFISTIK etc) in order to provide users with a friendly interface that demand
minimum knowledge of the background theory of reliability analysis.
2.3 Response Analysis
A fundamental decision to be made before the analysis is the identification of

the type that is required. Several analytical methods are available and may be
categorized as static or dynamic, linear or non-linear, deterministic or
stochastic. Combinations of the above categories can identify the analysis
method to be employed, depending on the properties of the structure under
consideration. The type of response of a structure may require different types of
analysis. In [70], an analytical block diagram for global response analysis of
marine structures is presented, guiding selection of the most appropriate
method.
2.3.1 Static Analysis
The general equilibrium equation for static analysis can be expressed as

follows:
ࡷ࢘ = ࡾ (2-7)
Where: ࡷ is the global stiffness matrix formed from the combination of the
element stiffness matrices; ࢘, is the vector of unknown nodal displacements and
ࡾ is the nodal load vector. The typical finite element analysis based on the
above equilibrium equation should follow the steps as described in [71]:
 Discretization, where the actual structure is approximated by an

assembly of finite interconnected elements
38
 Element Analysis, where the stiffness properties of each individual
element is determined and any loading is transformed into equivalent
nodal forces
 System Analysis, where the individual elements of the structure are
merged to form the elements stiffness matrix ࡷ and load vector ࡾ. The
above equation, will then determine the nodal displacement vector ࢘.
 Post-processing of the results, will derive the stresses from translational
and rotational displacements in each of the structural members.
Non-linear problems can generally be solved with both analytical (e.g., the Ritz
method) and numerical methods. The numerical methods seem to be the most
prominent, based upon the principle of stepwise integration of the problem such
that any non-linear structural problem can be transformed into a series of linear
problems. Reduction of an analytical non-linear structural system, can diminish
the problem to finding the displacement vector ࢘(࢚) that produces an internal
reaction force vector ࡲ࢏࢔࢚(࢚) that balances the applied forces ࡾ(࢚). The
expression of the equilibrium equation is formulated as follows and can be
solved incrementally with corrective iteration:
ࡲ࢏࢔࢚(࢚)࢘(࢚) = ࡾ(࢚) (2-8)
2.3.2 Dynamic Analysis
Following the same consideration, dynamic analysis includes time dependency,

damping and inertia as local and system effects:
ࡹ ࢘̈ + ࡯࢘̇ + ࡷ ࢘ = ࡾ(࢚, ࢘̇ , ࢘̈ ) (2-9)
Where: ࡾ() is the time-dependent load, ࢘(࢚) the displacement, ࡹ the global
matrix, ࡯ is the global damping matrix, and ࡷ the global stiffness matrix. Mass
and damping properties of the system, can be derived as the assembly of the
properties of each element. The internal reaction forces for any element can
then be computed by use of virtual work equations. The above equation is
applicable both for linear and non-linear systems.
39
2.3.3 Deterministic and Stochastic Processes
Deterministic process is a process for which it is possible to describe the exact

magnitude of the load at any given time. A deterministic analysis involves an
initial consideration of the statistical data for environmental loading. For extreme
response analysis, for example, a suitable event could be defined as the wave
which is expected to cause the most severe response. This requires that the
structural model is exposed to a unidirectional, periodic wave. The loading is
calculated in the time domain at given points in time during a wave cycle.
Contrary to deterministic processes, a stochastic process is described by the

use of probabilities. Therefore, a stochastic load or response may not be fully
described by exact magnitude at a given time, but rather by the probability
(statistical distribution) by which it will exceed some specified value. Further
discussion of modelling of environmental and capacity variables will be
presented in Chapter Four of this Thesis.
2.3.4 Selection of type of analysis
Literature can provide a range of methods for the selection of the most
appropriate analysis, considering that all relevant global and local effects,
dynamic as well as non-linear, are satisfactorily accounted for in the analysis
[72], [73]. In [74], a study based on different types of analysis for a reference
structure and for the same loading conditions was executed, including
combinations of linear and non-linear analysis, with regular and random wave
loads. Figure 6, adopted by [75], illustrate the results comparing the maximum
base shear that was derived.
40
Figure 6: Scatter of response (Max Base Shear-MN) by different Analysis
Methods [75]
Different approaches can be considered to account for dynamic effects, either

directly by dynamic analysis, or by applying ‘correction factors’ to the results of
a static analysis, such that the analysis may be referred to as ‘quasi-static’. For
the choice between a static and a dynamic analysis approach the following
recommendations may be considered [70]:
 For typical fixed offshore structures (e.g., jacket type structures) the
effects of dynamics, for extreme global response analysis, should be
included when the global natural period of the structure is greater than
three seconds.
41
 For floating structures, including compliant structures, a dynamic analysis
should always be undertaken to identify contribution of extreme response
dynamic effects.
 A dynamic analysis should be executed in order to establish structural
response when impulse or resonant effects may be governing.
 Model and/or prototype measurements should be considered for
structures or effects not amenable to analytical calculations.
2.3.5 System Response
Significant developments have been identified in the latest decades in the area
of system reliability assessment. A structural system with multiple failure paths
can be represented by a series of parallel sub-systems, with each subsystem
representing a failure mode. Starting from a component reliability level, the
combined structures reliability of the system can be then calculated. In the case
of complex structures such as offshore platforms, there is a number of potential
failure paths and structural components which makes this approach not
practical. Thus efforts have concentrated in developing more accurate system
reliability methods for these structures. Figure 7, presents some of the methods
for system reliability assessment [76].
The increased available computational resources for a system reliability

analysis of an offshore platform has given motivation for the development of a
number of “search algorithms”, in order to identify the most dominant failure
paths and calculate the combined system probability of failure [76]. For the case
of large structures, such as offshore platforms, the search algorithms technique
such as the ‘branch and bound method’ [77] can provide sufficient results.
Alternative methods towards identification of dominant failure paths are the
selective enumeration techniques [78] and the marginal probability and leading
probability methods [79]. Among methods, the ‘branch and bound method’ has
the drawback of computational cost while the enumeration techniques cannot
guarantee identification of every potential failure path.
42
Range of Methods
A Full probabilistic
B Search algorithms
 Selective enumeration
 Brunch and Bound
 Leading probability
C Pushover assisted by
 Simulations
 Sequence methods
 Response surface
D Simplified models
E “Component” based
Figure 7: Classification of methods for system reliability assessment [76]
Identification of the most dominant failure path can be also performed by the so
called ‘pushover analysis’. This analysis will identify deterministically the most
critical members but will not take into account the effect of possible residual
strength after failure which may redistribute loads and result in different
sequences of failure and different combination of members. However, in [80],
for cases under extreme loading conditions, it is found that the reliability index
of the failure path identified through a deterministic pushover analysis closely
identifies the one obtained after extensive searches or simulations.
From all of the approaches that have been proposed, the component based
approach is the simplest. In this approach the whole structure is treated as one
component, with either deterministic resistance or an associated coefficient of
variation (COV) suitable to quantify the probability of system collapse of a fixed
platform [81], [82]. Based on this approach it is proposed that with accurate
representation of the resistance, a good non-linear model and a competent
43
analyst it is possible to reduce significantly the modeling uncertainties
associated with the resistance.
Another, simplified, system reliability method, that can be applied as a

preliminary design tool for configuration of new platforms, has been based on a
series system where the components in series are the deck, each platform bay
and the foundations [83]. This method, considers within each component
parallel elements including deck, legs, braces, joints and piles. For each
component to fail, failure of all parallel elements should happen. The use of
simplified analytical procedures to estimate reference storm lateral loadings and
the ultimate capacities of platforms are comparably well in agreement to those
derived from more complex analysis.
The selection on the various methods discussed depends on the application.

Classification in Figure 7, distinguish the methods closer to “Α” associated in
general with higher level of complexity which is more suitable as research tools,
while methods towards the other end of the scale (towards “Ε”) would be more
appropriate for practical assessment [76]. However, the choice of the methods
would also be influenced by other factors such as the availability of
computational and analytical tools, significance of local effects such as
foundation uncertainties in the overall system reliability and the expertise of the
analyst.
In the application that is presented in Chapter Five, the pushover analysis

approach was selected in combination with simulation method in order to
address the possible variations and their effect. The difficulty associated with
this approach was the limiting number of simulations which can be performed
given the size of the problem and the high computational demands. Simple
approaches found in literature [81], [84] can provide useful conclusions in the
implementation of the above combination.
44
2.3.6 Practical Methods of Analysis
Among the various methods available in literature, this section will present a
practical procedure for the assessment of structural system reliability. The main
assumption indicates that failure will occur at one instant, for example when the
lateral wave load reaches a maximum value, implying that failure will occur over
a short period of time, during which the load is applied proportionally. This
assumption makes the reliability calculation problem time-independent.
Local failure on a structural member will alter the structural stiffness of the
structure, redistributing stresses in alternative load paths. Residual strength is
modelled by applying appropriate forces at the nodes of the failed members, or
by changing the properties of the failed members. Once the system has been
redefined, a new stress calculation should be performed, initiating an iterative
process, which after a sufficient number of successive member failures has
occurred, the structure will be considered to have failed when the structural
failure criterion (collapse or large displacement) is met. Therefore, this
procedure can identify a potential load path. In real structures, due to their
complexity, a large number of possible failure paths exist. For this reason a
search technique should be used in order to identify the important paths.
Pushover analysis is a method that can be followed for this purpose until the
platform collapses.
This manual search of failure paths, usually expressed in the form of failure
trees, can provide sufficient information for a system reliability assessment [85].
The failure tree is a representation of all possible failure paths in the structure.
Nodes represent damaged states of the structure and the branches represent
member failures in the corresponding damaged structure. The number in the
node is the element number. In the failure tree, each path represents a failure
mode, which can be modeled as a parallel system since the structure can only
fail when all its members reach their limit state. Α redundant system might have
several failure modes. In such case, each of the failure modes can be modeled
45
as a parallel system and all modes, in turn, can be modeled as a series system
to find the reliability of the complete system.
2.3.7 Modelling of Post-Failure Behaviour
Reliability of a structural member will strongly depend on whether it behaves on

a brittle or a ductile way. This does not mainly refer to the material of the
structure but in the behaviour of the potential members to fail. The response of
the component's post-failure behaviour is one of the key factors that determine
the effective redundancy of a structure. The two extreme types of failure are the
perfect brittle and the prefect ductile failure performance. The first type
becomes completely ineffective after failure, eliminating completely its load-
bearing capacity. If a failure element maintains its load-bearing capacity after
failure it is categorized as ductile. Real materials, in most of cases lay between
the two extreme categories. One model which can be incorporated in the
probabilistic analysis is the bi-linear, two state model. In the non-failure
condition, the component is linear elastic, while in the failed condition the
component still behaves linearly but with a modified stiffness matrix. Figure 8,
illustrates this behaviour.
Ductile
Force
Semi-brittle
Brittle
Displacement
Figure 8: Models of Post-Failure Behaviour
46
With this type of models, various component behaviours, ranging from brittle to
prefect plastic, can be described. In the semi-brittle model, the member force
increases elastically to the member capacity or resistance. After failure, if the
axial deformation in the element is increased beyond its failure value, the
element force abruptly drops to a fraction, φ, of its non-failure capacity. Based in
literature [86] a deterministic value of φ = 0.4 can be adopted for members
failing in compression and φ = 1.0 for tension failure. This assumes ductile
tension failure behaviour, maintaining the failure load and an abrupt drop by
40% capacity when failing in compression.
2.3.8 Methods in Computing System Reliability
Accurate estimation of reliability can be hesitated by the multiple failure paths

existing and any potential correlation between the failures. The later, should be
primarily identified since this correlation will affect the final results of the
calculations. Generally there are two approaches; the Hohenbichler
approximation [87] and the bounding method [88].
Hohenbichler approximation computes the probability of failure from the multi-

normal distribution function, providing a very accurate prediction of the
probability of failure of the structure. However, for complex structures, the
method becomes complicated from a computational point of view [79]. For such
structures, use of bounding techniques for the calculation of the probability of
failure, which is the simple bound and Ditlevsen bounds [89], are proven to be
more efficient methods. Τhe Ditleνsen bound is a narrower bound compared to
the simple bound, providing a smaller range between the upper and the lower
bound. The computation of the Ditleνsen bound is calculated through numerical
methods in order to calculate the joint probabilities.
The simple bound method can be also applied for the computation of structural
reliability however the results derived lack in accuracy compared to the
Hohenbichler approximation and the Ditleνsen bound. On the other hand, the
limited requirements in computational demands constitute the simple bound
method widely used. The simple bound is a range between the maximum and
47
minimum value of the probability of failure; however this range can be
significantly wide and it can only serve as a rough indication of the system
reliability. Before proceeding with the calculation of the simple bound method,
the types of system has to be identified, as series system, parallel system or
combination of both.
2.3.8.1 Classification
Α series system, or as is called the weakest link system, is the system where
failure corresponds to failure of the weakest element in the system. In
describing the status of each member, each element is assumed to be either in
a functioning or in a failed state. This consideration can be expressed by
introducing a binary state indicator variables ai (=1 for functioning members,
and =0 for failed members), considering n number of elements in the structure.
The simple bound for the derivation of the probability of failure for series system
[79], is defined as:
max ܲ(ܽ௜ = 0) ≤ ܲ௙௦ ≤ 1 − ෑ ൫1 − ܲ(ܽ௜ = 0)൯ (2-10)

௜ୀଵ
௜ୀଵ
The lower bound in the above equation is equal to the exact value of Pfs if there
is full dependence between all elements and the upper bound correspond to no
dependence between any pair of elements. When the probability of failure of
one element is predominant in relation to the other failure elements, the
probability of failure of series system is approximately equal to the predominant
probability of failure and the gap between the upper bound and lower bound is
narrow. In the opposite case, when the probabilities of failure are in the same
order, the simple bounds are wide.
For a parallel system, it is considered to be in a functioning state if at least one

element is functioning. Simple bounds for the probability of failure for parallel
system are defined as:
48
௡
ෑ ܲ(ܽ௜ = 0) ≤ ܲ௙௦ ≤ min ܲ(ܽ௜ = 0) (2-11)

௜ୀଵ
௜ୀଵ
Following a parallel concept as for the series systems, the lower bound in the
above equation is equal to the exact value of Pfp if there is no dependence
between any pair of members and the upper bound corresponds to full
dependence between all elements.
2.3.8.2 System effects
System effects in fixed offshore platforms can distinguish deterministic effects

which relate to the redundancy of the system and probabilistic effects which
relate to the randomness of the member capacities under stochastic loading.
Deterministic system effects relate to the redundancy incorporated into the
structure, which allows load redistribution after the first member failure and
results in a higher ultimate load capacity. Due to this, it can result to lower
requirements of reliability for individual members.
For perfectly balanced structures, the system effects for overload capacity
beyond first member failure are due to the randomness in the member
capacities. Α balanced structure in this sense refers to a structure where, in a
linear analysis, the first member to fail has the same probability as for all other
members. For a more realistic unbalanced structure, system effects are from
both deterministic and probabilistic effects. Deterministic effects are due to the
fact that the remaining members in the structure can still carry the load after one
or several members have failed, while the probabilistic effects are due to the
randomness in the member capacities [90]. Structural behavior beyond first
member-failure depends on the degree of static indeterminacy, ability of
structure to redistribute the load and ductility of individual members, as well as
by aspects such as wave-in-deck loading, the behavior of the joints and the
behavior of the foundation. In order to assess system effects, there are a
number of factors, such as reserve strength and residual strength that can be
derived from the analysis of a structural model.
49
The failure of only one part of a system may not limit the capacity of the
structure as a whole; instead, a sequence of component failures may occur
before the ultimate strength is reached. The reserve strength ratio (RSR) is
generally defined as:
‫݁ܿ݊ܽݐݏ݅ݏ݁ݎ ݉ݎ݋݂ݐ݈ܽ݌݁ݐܽ ݉݅ݐ݈ݑ‬

ܴܴܵ = (2-12)
݀݁‫݀ܽ݋݈ ݊݃݅ݏ‬
RSR can be quoted in terms of ratios of platform base shear or overturning

moment. For every platform, a different value of RSR will be obtained for every
different load case. Apart from that, in order to make comparisons between
different RSR values, one should note that the definition which is used to
quantify this value is the same. In [83] and [91], an extensive study on the
definition and the use of RSR value was undertaken developing a four-tier
system for the assessment of structures. An alternative definition of RSR may
be defined is:
ܴ௎
ܴܴܵ = (2-13)
ܵோ
Where: Ru is ultimate lateral load capacity of the platform and SR is a reference

lateral loading. The primary objective in the four-tier system is to allow a simple
assessment and re-qualification of platforms. The more complicated levels in
this system would be used for more complex platforms including intense
analyses for re-qualification. Another definition of RSR value, used by Shell
[82], is given as:
݁݊‫݁ݏ݌݈݈ܽ݋ܿݐܽ ݀ܽ݋݈݈ܽݐ݊݁ ݉݊݋ݎ݅ݒ‬

ܴܴܵ = (2-14)
‫݀ܽ݋݈݈ܽݐ݊݁ ݉݊݋ݎ݅ݒ݊݁ ݊݃݅ݏ݈݁݀ܽ݊݅݃݅ݎ݋‬
Considering an undamaged structure, some residual capacity may exist, the

magnitude of which can be described by its degree of in-determinacy. The
effect of a certain damage scenario can be assessed by the concept of residual
strength. This can be an important indicator of structural behavior, and can be
defined by the residual resistance factor (RIF), generally defined as follows [92]:
50
݀ܽ݉ ܽ݃݁݀ ‫݁ݎݑݐܿݑݎݐݏ‬ᇱ‫݁ݏ݌݈݈ܽ݋ܿݐܽ ݀ܽ݋݈݈ܽݐ݊݁ ݉݊݋ݎ݅ݒ݊݁ݏ‬
ܴ‫= ܨܫ‬ (2-15)
݅݊‫݁ݎݑݐܿݑݎݐݏݐܿܽݐ‬ᇱ‫݁ݏ݌݈݈ܽ݋ܿݐܽ ݀ܽ݋݈݈ܽݐ݊݁ ݉݊݋ݎ݅ݒ݊݁ݏ‬
The ratio of the ultimate capacity of the damaged structure, when compared to
the ultimate capacity of the intact structure, can also give a useful indication of
platform behavior [92]. This can be defined as the damage tolerance ratio
(DTR), which can be expressed as follows:
݀ܽ݉ ܽ݃݁݀ ‫݁ݎݑݐܿݑݎݐݏ‬ᇱ‫ݕݐ݅ܿܽ݌ܽܿ݁ݐܽ ݉݅ݐ݈ݑݏ‬

‫= ܴܶܦ‬ (2-16)
݅݊‫݁ݎݑݐܿݑݎݐݏݐܿܽݐ‬ᇱ‫ݕݐ݅ܿܽ݌ܽܿ݁ݐܽ ݉݅ݐ݈ݑݏ‬
The value of DTR characterizes the weakening of the structure caused by the
damage. For example, a DTR of 0.9 would indicate a 10% loss in the reserve
strength.
The damage tolerance ratio has been calculated in order to investigate the
residual strength in the application of Chapter Five. Push over analysis was
carried out in 8 wave directions of the platform; 45 degrees apart as indicated
from standards for orthogonal (4 legs) cross section, to derive the DTR for each
direction. The most critical member, i.e. the member that has the highest
probability of failure was removed for each direction. After analyzing the first
damage, the second critical member was removed iteratively, in order to study
the effect of multiple damages.
2.4 Review of Stochastic Methods
2.4.1 Stochastic Expansions
Extensive use of Finite Elements in various fields of engineering has evolved

using stochastic field theory in structural engineering. The critical issue towards
this scope is the discrete representation of stochastic variables and the
corresponding interpretation of stochastic responses [93]. Stochastic expansion
is an efficient tool for reliability analysis. The purpose of stochastic expansions
is to consider uncertainties through a series of polynomials in order to
investigate the reliability of a system. Karhunen-Loeve expansion (K-L) and
51
Polynomial Chaos Expansion (PCE) can serve the above purpose. Combination
of the above two techniques together with principles of the Finite Element
Analysis Methods provide a useful tool, the Spectral Stochastic Finite Element
Method (SSFEM), towards an analytical assessment of the reliability of
structural systems [94].
Applied in the Spectral Stochastic Finite Element Method (SSFEM) [95],

Polynomial Chaos Expansion (PCE) has been successfully used to represent
uncertainty in a variety of applications, including structural response. This
method uses orthogonal polynomials of random variables. Most commonly, the
random variables are standard-normal, and Hermite polynomials are used in
SSFEM. Efficiency of this method has found multiple applications in various
engineering problems such as two dimensional elasticity [95], soil mechanics
[96], heat conduction [97] and composite materials [98].
In [99] the probabilistic collocation method is introduced; according to this the

responses of stochastic systems are projected onto the PCE. In [100] the
limitation of the probabilistic collocation method for large-scale models was
indicated and a different approach was suggested, using a stochastic response
surface method that uses the partial derivatives of model outputs with respect to
model inputs. In [101] combination of PCE with MCS was applied while in [102]
extended PCE were used to represent different distribution functions by using
the Askey scheme.
Stochastic Expansions could be classified in two categories: the non-intrusive

and intrusive formulation procedures, as shown in Figure 9 [93]. An intrusive
formulation is the one in which the representation of uncertainty is expressed
explicitly within the analysis of the system. Practically this refers to methods that
use PCE and KL expansions to directly modify the stiffness matrix of a finite
element analysis procedure. SSFEM [103] and the stochastic Galerkin FEM
[104] are both intrusive formulations. On the other hand, non intrusive
formulations, represents uncertainties in a non explicit way, treating the analysis
code as a “black box” without requiring access to the analysis code. This
52
method is called Stochastic Response Surface Method, and will be studied in
depth in the following sections.
Although this Thesis does not deal with the Spectral Stochastic Finite Element
Method, for reasons of completeness, it will be briefly presented in the following
sections, as the combination of stochastic expansions and Finite Element
Methods. The Stochastic Response Surface Method will be presented in the
next chapter in greater detail since it will be used in the numerical part of this
Thesis.
Figure 9: Intrusive and non-intrusive formulation [93]
2.4.2 Spectral Stochastic Finite Elements
The key aspect of SSFEM, is the appropriate transformation of a complicated

random quantity, to a set of simpler random quantities, easier to assess.
Towards this scope, two different stages can be distinguished:
 One involving representation of the random processes used to model the

corresponding random properties.
 One involving the solution process.
The properties mentioned in the first stage, are assumed to be represented

through their second-order statistics, varying continuously over space. Each of
those processes is considered as an uncountable set of random variables which
should be replaced by a finite set, able to be truncated at a presecribed level
that represents sufficient accuracy. Karhunen-Loeve expansion, or different
53
expansions eigensolutions of self-adjoint operators [105], can be employed
towards this scope.
As far as the solution process is concerned, it consists of a vectorial random

process whose members represent the random solution at the nodes of the
finite element discretization. For the description of the solution in a formulation
independent to the unknown properties, Polynomial Chaos Expansion should
be employed, degrading the problem to the calculation of deterministic
coefficients representing the solution process with respect to this expansion. In
[106] and [107], expansions that refer to non-Gaussian processes can be found,
as a variation of the conventional Polynomial Chaos expansion.
2.4.2.1 Karhunen-Loeve (KL) Transform
Stochastic analysis faces the challenge of appropriate representation of

uncertainty in a computationally feasible way. The concept of random fields can
adequately represent space and time varying variables such as distributed
loads and material properties. Large numbers of variables often cause problems
to uncertainty analysis, affecting the accuracy of analysis and therefore the
predicted reliability. This fact, especially for cases with highly correlated
variables, yields a demand for compromising reduction in dimensionality and at
the same time ensuring accuracy for the uncertainty analysis.
Karhunen and Loeve, have represented the continuous-time random process in

terms of ortho-normal coordinate functions derived from the covariance function
while in [108] principle components were used to analyze the correlation among
different variables. Further, various methodologies have been presented in
order to achieve reduction of the variables and their correlation. In [109], rank
correlation is used to generate the resultant dependent random variables and in
[110] a modification of this approach was proposed using ranks of the sample
elements. In [111], [112], [113] a complete non-intrusive procedure combined K-
L transform, PCE and LHS to account uncertainties in properties in structural
reliability assessment. Finally, orthogonal decomposition is often used to
generate correlated random variables.
54
K-L transform is a more advanced tool than the orthogonal transform method
and will be briefly analyzed in this section in order to later derive the
methodology of the SSFEM. It is an important tool in computational application
since it can reduce the dimensionality of the problem generating correlated
random variables [114].
Considering the stochastic process ‫߯(ܧ‬, ߠ), the corresponding K-L expansion
will be based on the spectral expansion of its covariance function ‫ܥ‬ாா (‫ݔ‬, ‫)ݕ‬,
where ‫ ݔ‬and ‫ ݕ‬denote spatial coordinates and ߠ indicates the random quantity.
Common properties of the covariance function are symmetry and positive
definition, with mutually orthogonal eigen-functions. The expansion can be
formulated as:
‫߯(ܧ‬, ߠ) = ‫ܧ‬ത(‫ )ݔ‬+ ෍ ඥߣ௜ߦ௜(ߠ)߰ ௜(‫)ݔ‬ (2-17)

௜ୀଵ
Where: ‫ܧ‬ത(‫ )ݔ‬denotes the mean of the stochastic process, {ߦ௜(ߠ)} a set of
orthogonal random variables, {߰ ௜(‫ })ݔ‬are the eigen-functions and {ߣ௜} are the
eigenvalues, that can be evaluated as the solution to the following equation:
න ‫ܥ‬ாா (‫ݔ‬, ‫ ߰)ݕ‬௜(‫ߣ = ݕ݀)ݕ‬௜߰ ௜(‫)ݔ‬ (2-18)

ࣞ
Where: ࣞ denotes the spatial domain over which the process ‫߯(ܧ‬, ߠ, ) is
defined.
The expression of (2-18) denotes that the random fluctuations have been
decomposed into a set of deterministic functions in the spatial variables by
combination of random coefficients that are independent of these variables. The
vector of the random variables {ߦ௜} will follow the properties of ‫߯(ܧ‬, ߠ).
2.4.2.2 Polynomial Chaos expansion
Polynomial Chaos Expansions is a stochastic method used for the

representation of the correlation between the basic design variables and the
response of the structure or the structural members. Considering that the
55
solution process is a function of the material properties, the nodal response
ො can be formed stochastically by a set of non linear functions {ߦ௜(ߠ)}.
vector ‫ݑ‬
This functional dependence can be expanded in terms of polynomials in
Gaussian random variables, referred to as Polynomial Chaoses [115], as
follows:
௫ ௫ ௜భ
‫ܽ = )ߠ(ݑ‬଴߁଴ + ෍ ܽ௜భ ߁ଵ ൫ߦ௜(ߠ)൯+ ෍ ෍ ܽ௜భ௜మ ߁ଶ ቀߦ௜భ (ߠ), ߦ௜మ (ߠ)ቁ+ ⋯. (2-19)

௜భୀଵ ௜భୀଵ ௜మୀଵ
Where: ߁௡ ൫ߦ௜భ , … … , ߦ௜೙ ൯ denotes the ݊-th order Polynomial Chaos of the
variables ൫ߦ௜భ , … … , ߦ௜೙ ൯. {ߛ௜(ߠ)}, being a one-to-one mapping to a set with
ordered indices and truncating the Polynomial Chaos expansion the above can
be reduced to:
‫ = )ߠ(ݑ‬෍ ‫ݑ‬௝ߛ௝(ߠ) (2-20)

௝ୀ଴
In those polynomials, the inner product 〈ߛ௜ߛ௞〉, which is defined as the statistical
average of their product, is equal to zero for ݆≠ ݇. Once ‫ݑ‬௝ are calculated, the
process ‫ )ߠ(ݑ‬can be assessed. Increasing the amount of random variables or
the order ot PCE polynomials can refine those series. Further, PCE can be
used to represent non-Gausian processes (variables). Combination of
Karhunen-Loeve expansion and the Polynomial Chaos expansion will be
presented in the next section in order to formulate a Spectral Stochastic Finite
Element Method procedure.
2.4.2.3 Formulation of Spectral Stochastic Finite Element Method
The formulation of the global stiffness matrix for uncertain stiffness can be
expressed as:
ࡷ (ߠ) = න ࡮ ் ࡯(‫ݔ‬, ߠ)࡮݀ࢹ (2-21)

ఆ
56
Where: ࡮ is the matrix relating strains to nodal displacements and ࡯(‫ݔ‬, ߠ) is the
randomness matrix. Introducing the K-L expansion for ࡯(‫ݔ‬, ߠ), (2-21) can be
transformed as:
‫ݔ‬
ഥ + ෍ ߦ (ߠ) න ࡮ ் ࡯૚(‫ࢹ݀࡮)ݔ‬
ࡷ (ߠ) = ‫ܭ‬ (2-22)
݅
݅=1
ఆ
ഥ, the above can be expressed as:

Redefining, ߦ௢ = 1 and ࡷ ૙ = ࡷ
‫ݔ‬
ࡷ (ߠ) = ෍ ߦ݅(ߠ) න ࡮ ் ࡯૚(‫ࢹ݀࡮)ݔ‬ (2-23)

݅=1
ఆ
While, truncating the K-L expansion, the final matrix equation for the finite
element model is transformed to:
൭෍ ߦ௜(ߠ) ࡷ ࢏൱ ‫݂ = )ߠ(ݑ‬ (2-24)

௜ୀ଴
Combining PCE with K-L expansions through (2-20) and (2-24), the nodal
response values can be obtained as:
௉ ெ
ቌ෍ ෍ ߦ௜(ߠ)ߛ௜(ߠ) ࡷ ࢏ቍ ‫ݑ‬௝ = ݂ (2-25)

௝ୀ଴ ௜ୀ଴
Enforcing orthogoniality to the above equation, a system of linear equations can

be obtained as:
‫ݑܭ‬
ො= ݂ (2-26)
Where: ࡷ ࢏࢑ = ∑ெ௜ୀ଴ ࡷ ࢏〈ߦ௜(ߠ)ߛ௜(ߠ)ߛ௞(ߠ)〉 , ݆, ݇ = 0, … , ‫݌‬, and ‫ݑ‬

ො is an extended
solution vector containing ‫ݑ‬௝ as its sub-vectors. Solution of the system for the
coefficient vector ‫ݑ‬
ො, provides the statistical properties of the nodal response
values.
57
2.5 Summary
This chapter has dealt with fundamental aspects of the reliability analysis of
offshore and steel structures. After basic mathematical formulation of the
reliability estimation problem, the motivation of a reliability framework and the
evolution in its application on the assessment of structures was presented.
Selection of the appropriate type of analysis was also discussed in conjunction
with methods for integration from a local to a global level of reliability
calculation, distinguishing a practical method for the deterministic assessment
of the non linear performance of structures to failure. Finally, a review of
stochastic expansions was included, followed by a brief formulation of the
Spectral Stochastic Finite Elements Method, in order to set the background for
the development of the Stochastic Response Surface Method, which will be
analytically discussed in the next chapter.
58
3 NUMERICAL METHODS FOR STRUCTURAL
RELIABILITY ANALYSIS
3.1 Introduction
In this Chapter the numerical procedures for the computation of reliability will be
presented. Solution of the integral of the joint probability distribution function
imposes difficulties in the calculation of the probability of failure equation.
Therefore, use of limit state function approximations can be employed in order
to overcome this problem. Appropriate selection of the proper technique, can
handle many engineering problems of great importance. In the following
sections, Level III reliability analysis methods will be presented analytically, as
they will be applied later in this Thesis. Both deterministic and probabilistic
formulations will be discussed. The Stochastic Response Surface Method
(SRSM) will be presented and after a literature review of this method and
multivariate regression techniques, the methodology that will be used in this
Thesis will be derived.
3.2 Numerical Methods
3.2.1 Deterministic Methods
The methods and algorithms that will be discussed in this section refer to
deterministic handling of the limit state functions through geometrical
approximation of the stochastic variables. Random variables are characterized
by their moments. Initially the First-order Second Moment Reliability Method
(FOSM) will be presented that will stand as a basis for the First-order and later
the Second-order Reliability Methods (FORM/SORM). Analytical description of
those methods can be found in [116].
The multiple variables participating in the probability of failure calculation have

yielded several methods that would simplify this procedure. Employment of first
59
and second order Taylor series expansion is a common practice applied to
linearize the limit state equation. This practice refers respectively to the First
and Second Order Moment methods. FOSM is referred to as mean value first
order second moment method (MVFOSM) and is a simplistic method that
cannot provide results of sufficient accuracy for very low probabilities of failure
or non linear limit state functions [117]. Addition of the second term, in SOSM
methods cannot sufficiently handle this problem.
In order to overcome the above difficulties, a geometrical solution, the safety

index approach, transforms the problem to a mathematical optimisation problem
of finding the point of the limit state surface with the minimum distance to the
origin of the standard normal space. In [52], the Hasofer and Lind (HL)
algorithm is introduced transforming the vector of the design stochastic
variables ܺ into a vector of standardized independent variables ܷ. The design
point in the ܷ-space represents the point of greatest probability density and is
called the Most Probable failure Point (MPP). The transformed limit state
surface ݃(ܷ) = 0 can be approached with first or second order approximations
and therefore account for First and Second Order Reliability Methods
(FORM/SORM).
yn
Limit-state function
MPP
Second-order
Approximation
First-order
Approximation
β
0 yi
Figure 10: First and Second order approximations
The interpretation of the above approach is that in FORM, the limit state surface
is approximated by a tangent plane at the MPP while for SORM the MPP is
60
approximated by a curve as it can be seen in Figure 10. FORM gives inaccurate
results in cases of highly non linear limit state surfaces and large curvatures.
Figure 11: Transformation to the U-space [93]
3.2.1.1 First Order Reliability Methods
3.2.1.1.1 The Mean Value FOSM (MVFOSM)
The Mean Value FOSM (MVFOSM) simplifies the calculation procedure of the
probability of failure of a limit state function. The characterization “first-order”
derives from the employment of first order expansions for the linearization of the
initial function, expressing inputs and outputs as the corresponding mean and
standard deviation. This simplifying approximation, neglects higher moments,
and therefore increases the subsequent model uncertainty. According to this
method, the limit-state function is approximated by the first-order Taylor series
expansion at the mean value point. Considering X to be the vector of statistically
61
independent variables, the approximate limit-state function at the mean can be
expressed as:
݃෤(ܺ) ≈ ݃(ߤఄ ) + ∇݃(ߤఄ )் ∙ ൫ܺ௜ − ߤ௫೔൯ (3-1)

்
Where: the vector of mean values is ߤఄ = ൛ߤఄభ, ߤఄమ , … ߤఄ೙ ൟ , and the gradient of
g evaluated at μΧ as:
்
߲݃(ߤ௑ ) ߲݃(ߤ௑ ) ߲݃(ߤ௑ )
ߘ݃(ߤ௑ ) = ቊ , ,…, ቋ (3-2)
߲‫ݔ‬ଵ ߲‫ݔ‬ଶ ߲‫ݔ‬௡
The mean (expected) value of the approximated limit-state function ݃෤(ܺ) is:
ߤ௚෤ ≈ ‫ߤ(݃[ܧ‬ఄ )] = ݃(ߤఄ ) (3-3)
Following some fundamental statistics transformations, the standard deviation

of the approximate limit-state function is calculated as:
ߪ௚෤ = ඥܸܽ‫݃[ݎ‬෤(ܺ)] = ඥ[ߘ݃(ߤఄ )்]ଶ ∙ ܸܽ‫)ܺ(ݎ‬

ଵ
௡ ଶ ଶ (3-4)
߲݃(ߤ௑ )
= ൥෍ ቆ ቇ ∙ ߪ௫ଶ೔൩
߲‫ݔ‬௜
௜ୀଵ
The reliability index β is then calculated as:
ߤ௚෤ ݃(ߤఄ )
ߚ= ≈ ଵ
ߪ௚෤ ଶ ଶ (3-5)
߲݃(ߤ௑ )
ቈ∑௡௜ୀଵ ൬ ൰ ∙ ߪ௫ଶ೔቉
߲‫ݔ‬௜
For cases of linear limit state functions, the above expression of the reliability
index can be analytically derived by expressing the safety margin between a
resistance R and a loading S of a system with normal variables as:
݃(ܺ) = ܴ(ܺ) − ܵ(ܺ) (3-6)
The resulting mean value ߤ௚ and standard deviation ߪ௚ are respectively:
ߤ௚ = ߤோ − ߤௌ (3-7)
62
ߪ௚ = ටߪோଶ + ߪௌଶ − 2 ∙ ߩோௌ ∙ ߪோ ∙ ߪௌ (3-8)
Where: ߩோௌ is the correlation coefficient between ܴ and ܵ, and ߤோ , ߤௌ , ߪோ , ߪௌ the

mean values and standard deviations of the ܴ and ܵ variables. The reliability
index ߚ is calculated as:
ߤ௚ ߤோ − ߤௌ
ߚ= = (3-9)
ߪ௚ ඥߪோଶ + ߪௌଶ − 2 ∙ ߩோௌ ∙ ߪோ ∙ ߪௌ
And for the case of uncorrelated variables, where: ߩோௌ = 0:
ߤ௚ ߤோ − ߤௌ
ߚ= = (3-10)
ߪ௚ ඥߪோଶ + ߪௌଶ
In cases of nonlinear limit-state functions, the approximate limit-state surface

can be derived by linearization of the initial limit-state function at the mean value
point. In the generalized case with multiple independent variables, the failure
surface is a represented by a hyper plane that is defined as a linear-failure
function:
݃෤(ܺ) = ܿ଴ + ෍ ܿ௜ ∙ ‫ݔ‬௜ (3-11)

௜ୀଵ
ߤ௚෤ = ܿ଴ + ܿଵߤ௫భ + ܿଶߤ௫మ + ⋯ + ܿ௡ ߤ௫೙ (3-12)
ߪ௚෤ = ඩ ෍ ܿଵଶ ∙ ߪ௫ଶ೔ (3-13)

௜ୀଵ
The MVFOSM method is a simplistic method for the calculation of reliability

indices, using minimum representation of basic variables. This fact diminishes
the range of applicability of the method due to the following reasons:
 Non linearity or large variations cannot be efficiently handled by that

method, since linearization of the limit-state function about the mean
values can lead to inaccurate results.
63
 The MVFOSM method is dependent on different - mathematically
equivalent - formulations of a same problem; both for linear and non
linear expressions of the limit state function.
3.2.1.1.2 Hasofer and Lind
The reliability index can be interpreted as the geometrical distance defined from
the origin of a ‫ݑ‬-dimensional space to the Most Probable failure Point (MPP) on
the failure surface. The Hasofer and Lind reliability index method transforms the
expansion point from the mean value point to the MPP improving the approach
of the MVFOSM. To expand this practice for problems with multiple variables,
Hasofer and Lind [52] proposed a linear transformation of the basic variables ‫ݔ‬௜
into a set of normalized and independent variables ‫ݑ‬௜.
For the basic case with two independent, normally distributed variables of
strength ܴ, and stress, ܵ, Hasofer and Lind transformed the initial variables to
standard normalized ones, following:
ܴ − ߤோ ܵ − ߤௌ
ܴ෠ = , ܵመ= (3-14)
ߪோ ߪௌ
Where: ߤோ and ߤௌ are the mean values and ߪோ and ߪௌ are the standard
deviations of R and S, respectively. Following, the limit-state surface ݃(ܴ, ܵ) =
ܴ − ܵ = 0 should be transformed from the original (ܴ, ܵ) coordinate system into
the limit-state surface, in the standard normalized ൫ܴ෠, ܵመ൯coordinate system as:
݃ ቀܴ൫ܴ෠൯, ܵ൫ܵመ൯ቁ = ݃ො൫ܴ෠, ܵመ൯= ܴ෠∙ ߪோ − ܵመ∙ ߪௌ + (ߤோ − ߤௌ) = 0 (3-15)
The distance from the origin in the ൫ܴ෠, ܵመ൯ coordinate system to the failure
surface ݃ො൫ܴ෠, ܵመ൯= 0 is equal to the safety-index:
(ߤோ − ߤௌ)
෢ ∗=
ߚ = ܱܲ (3-16)
ඥߪோଶ + ߪௌଶ
The point ܲ∗ ൫ܴ෠∗ , ܵመ∗ ൯ on ݃ො൫ܴ෠, ܵመ൯= 0, is the Most Possible failure Point (MPP),
and corresponds to this shortest distance. In the general case with ݊ normally
64
distributed and independent variables, the failure surface is described by a
nonlinear function:
݃(ܺ) = ݃({‫ݔ‬ଵ, ‫ݔ‬ଶ, … , ‫ݔ‬௡ }் ) = 0 (3-17)
Transformation of the variables into their standardized forms should follow:
‫ݔ‬௜ − ߤ௫೔
‫ݑ‬௜ = (3-18)
ߪ௫೔
Where: ߤ௫೔ and ߪ௫೔ represent the mean and the standard deviation of ‫ݔ‬௜,
respectively. The mean and standard deviation of the standard normally
distributed variable ‫ݑ‬௜ are zero and unity, respectively.
The failure surface ݃(ܺ) = 0 in X-space is mapped into the corresponding

failure surface ݃(ܷ) = 0 in U-space. Due to the rotational symmetry of the
second-moment representation of U, the geometrical distance from the origin in
U-space to any point on ݃(ܷ) = 0 refers to the number of standard deviations
from the mean value point in X-space to the corresponding point on ݃(ܺ) = 0.
The safety-index β is the shortest distance from the origin to the failure surface
݃(ܷ) = 0, as:
ଵൗ
ߚ= min (ܷ ் ∙ ܷ) ଶ (3-19)
௎∈௚(௎)ୀ଴
This value of ߚ is called the ‘Hasofer and Lind (HL) safety-index’ ߚு௅. The point
ܷ ∗ (‫ݑ‬ଵ∗ , ‫ݑ‬ଶ∗ , … , ‫ݑ‬௡∗ ) on ݃(ܷ) = 0 is the design point, which can provide the
corresponding vector point in the ܺ-space.
Based on the above theoretical presentation of the method, the problem of the
calculation of the reliability index ߚ can be derived as the solution of a
constrained optimization problem in the standard normal ܷ-space.
ଵൗ
݉ ݅݊݅݉ ݅‫݁ݖ‬: ߚ(ܷ) = (ܷ ் ∙ ܷ) ଶ with ݃(ܷ) = 0 (3-20)
Several algorithms are available that can solve this problem, such as
mathematical optimization schemes or other iteration algorithms. In [118],
65
several constrained optimization methods were used in order to solve this
optimization problem, including primal methods (feasible directions, gradient,
projection, and reduced gradient), penalty methods, dual methods, and
Lagrange multiplier methods [116]. Each method’s applicability depends on the
nature of the problem that is investigated. Following, the HL and HL-RF
methods are presented, as the most commonly used algorithms.
The HL algorithm was proposed by Hasofer and Lind, allowing consideration of

normally distributed random variables. Rackwitz and Fiessler extended the HL
method in order to handle non-Gaussian statistical distributions, forming the
extended HL-RF method. Assuming that the (linear or non linear) limit state
surface with n normally distributed and independent random variables X can be
expressed as:
݃(ܺ) = ݃({‫ݔ‬ଵ, ‫ݔ‬ଶ, … , ‫ݔ‬௡ }் ) = 0 (3-21)
Based on the transformation the limit-state function becomes:
்
݃(ܷ) = ݃ ቀ൛ߪ௫భ ‫ݑ‬ଵ + ߤ௫భ , ߪ௫మ ‫ݑ‬ଶ + ߤ௫మ , … , ߪ௫೙ ‫ݑ‬௡ + ߤ௫೙ ൟ ቁ = 0 (3-22)
The normal vector from the origin ܱ෠ to the limit-state surface ݃(ܷ) generates an
intersection point ܲ∗ . The distance from the origin to the MPP is the safety-index
β. The first-order Taylor series expansion of ݃(ܷ) at the MPP ܲ∗ is:
௡
߲݃(ܷ ∗ )
݃෤(ܷ) ≈ ݃(ܷ ∗)
+෍ ∙ (‫ݑ‬௜ − ‫ݑ‬௜∗ ) (3-23)
߲ܷ௜
௜ୀଵ
From the transformation:
߲݃ො(ܷ) ߲݃(ܺ)
= ∙ ߪ௫೔ (3-24)
߲ܷ௜ ߲‫ݔ‬௜
The minimum distance from ܱ෠ to the ݃෤(ܷ) surface may be given as:
66
߲݃(ܷ ∗ )
݃(ܷ ∗ ) − ∑௡௜ୀଵ ∙ ߪ௫೔ ∙ ‫ݑ‬௜∗
߲‫ݔ‬௜
ܱ෠ܲ∗ = ߚ =
ଶ (3-25)
ඨ ∑௡௜ୀଵ ൬߲݃(ܷ )
∗
∙ ߪ௫೔൰
߲‫ݔ‬௜
The direction cosine of each of the transformed variables, often called

sensitivity factor, is given as follows, expressing the relative effect of the
corresponding random variable on the total variation.
߲݃(ܷ ∗ ) ߲݃(ܺ ∗ )
∙ ߪ௫೔
߲‫ݑ‬௜ ߲‫ݔ‬௜
ܽ௜ = cos ߠ௫భ = cos ߠ௨భ = − = ଵൗ (3-26)
|∇݃(ܷ ∗ )| ଶ ଶ
߲݃(ܺ ∗ )
ቈ∑௡௜ୀଵ ൬ ∙ ߪ௫೔൰ ቉
߲‫ݔ‬௜
The coordinates of the point P* are computed as:
‫ݔ‬௜∗ − ߤ௫೔
‫ݑ‬௜∗ = = ܱ෠ܲ∗ cos ߠ௫భ = ߚ cos ߠ௫భ (3-27)
ߪ௫೔
And, transforming them into the original space ܺ:
‫ݔ‬௜∗ = ߤ௫೔ + ߚߪ௫೔ cos ߠ௫భ , (݅= 1, 2, … , ݊) (3-28)
Since P* is a point on the limit-state surface, it should satisfy:
݃({‫ݔ‬ଵ∗ , ‫ݔ‬ଶ∗ , … , ‫ݔ‬௡∗ }் ) = 0 (3-29)
In cases where the failure surface may contain several points corresponding to
stationary values of the reliability-index function (multiple MPP problem), it may
be necessary to use several starting points in order to find all the values
{ߚଵ, ߚଶ, . . . , ߚ௠ }, deriving the HL safety-index as:
ߚு௅ = ݉ ݅݊{ߚଵ, ߚଶ, . . . , ߚ௠ } (3-30)
67
DEFINITION OF LSF
INITIAL DESIGN POINT
CALCULATION OF
GRADIAENTS AT THE INITIAL
DESIGN POINT
CALCULATION OF β AND
DIRECTION COSINES
COMPUTATION OF NEW
DESIGN POINTS XK AND UK
CALCULATE GRADIENTS AT
THE NEW DESIGN POITNS
CALCULATION OF β AND
DIRECTION COSINES
NO
CONVERGENCE CRITERION
YES
DEFINE DESIGN POINTS XK

OR MPP
Figure 12: Algorithm of HL Reliability Index Calculation
The difference between the MVFOSM method and the HL method is that the HL
method approximates the limit-state function using the first-order Taylor
expansion at the design point X(k) or U(k) instead of the mean value point μX
[116]. Further, the MVFOSM method is a straight forward procedure, while the
HL method needs several iterations to converge, especially for nonlinear
problems. The HL method usually provides better results than the mean-value
method for nonlinear problems. The quality of the linearized limit-state function,
݃෤(ܷ) = 0 will determine the accuracy of the calculation of the probability of
failure ܲ௙.
68
3.2.1.1.3 Hasofer Lind - Rackwitz Fiessler (HL-RF) Method
In the Hasofer-Lind reliability index method, the random variables X are

assumed to be normally distributed. In cases of non-Gaussian variables, the
reliability calculation procedures as presented so far are inefficient. Many
structural reliability problems involve non-Gaussian random variables; therefore
it is necessary to find a way to treat such problems. There are many methods
available for conducting transformations to the normalized space, such as those
found in [119], and [120]. A simple, approximate transformation called ‘the
equivalent normal distribution’, or ‘the normal tail approximation’, is presented in
this section. The main advantages of this transformation are:
i. Avoidance of the multi-dimensional integration of the main integral

for the probability of failure.
ii. Transformation of non-Gaussian variables into equivalent normal
variables is accomplished prior to the solution of the final equations
avoiding overcomplicated procedures.
iii. Results often agree with the exact solution of the multi-dimensional
integral of the fundamental probability integral.
Considering, mutually independent variables, the transformation is given as:
‫ݑ‬௜ = ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜)൧ (3-31)
Where: ߔ ିଵ[. ] is the inverse of ߔ [. ].
Employment of Taylor series expansion of the transformation at the MPP ܺ ∗,

neglecting nonlinear terms [121], can provide the equivalent normal distribution
as:
߲
‫ݑ‬௜ = ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜∗ )൧+ ൫ൣߔ ିଵ‫ܨ‬௫೔(‫ݔ‬௜)൧൯|‫ݔ‬௜∗ ∙ (‫ݔ‬௜ − ‫ݔ‬௜∗ ) (3-32)
߲‫ݔ‬௜
Where:
߲ ିଵ ݂௫೔(‫ݔ‬௜)
ߔ ൣ‫ܨ‬௫೔(‫ݔ‬௜)൧= (3-33)
߲‫ݔ‬௜ ߮൫ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜∗ )൧൯
69
After substitution of the above and rearrangement of terms:
‫ݔ‬௜ − ൣ‫ݔ‬௜∗ − ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜∗ )൧߮൫ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜∗ )൧൯/݂௫೔(‫ݔ‬௜∗ )൧ ‫ݔ‬௜ − ߤ௫೔′

‫ݑ‬௜ = = (3-34)
߮൫ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜∗ )൧൯/݂௫೔(‫ݔ‬௜∗ ) ߪ௫′
೔
Where: ‫ܨ‬௫೔(‫ݔ‬௜) is the marginal cumulative distribution function ݂௫೔(‫ݔ‬௜), is the

probability density function, and ߤ௫′ and ߪ௫′ are the equivalent means and
೔ ೔
standard deviations of the approximate normal distributions, which are given as:
߮൫ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜∗ )൧൯

ߪ௫′ = (3-35)
೔ ݂௫೔(‫ݔ‬௜∗ )
ߤ௫′ = ‫ݔ‬௜′ − ߔ ିଵൣ‫ܨ‬௫೔(‫ݔ‬௜∗ )൧∙ ߪ௫′ (3-36)

೔ ೔
Another way to get equivalent normal distributions is to match the cumulative

distribution functions and probability density function of the original, non-normal
random variable distribution, and the approximate or equivalent normal random
variable distributions at the MPP [119].
Figure 13: Normalized Tail Approximation [116]
This normalized-tail approximation is shown in Figure 13. Using the above

procedure, the transformation of the random variables from the ܺ-space to the
ܷ-space can be easily achieved, and the performance function ݃(ܷ) in ܷ-space
is approximately obtained.
70
The RF method is also called the HL-RF method, since the iteration algorithm
was originally proposed by Hasofer and Lind and later extended by Rackwitz
and Fiessler to include random variable distribution information. For the
extended RF algorithm, the same steps as shown in Figure 12 should be
followed considering one extra block for the transformation of variables before
the definition of the initial design point.
The HL and HL-RF algorithm, approximates the limit-state function, ݃(ܷ), by the
first-order Taylor expansion at the MPP. For nonlinear problems, this approach
cannot adequately approximate ߚ, requiring several iterations until
convergence, especially for poorly linearization of nonlinear functions ݃(ܷ).
Different approximation could be used alternatively, such as the Two-point
Adaptive Nonlinear Approximations (TANA) [122], including TANA and TANA2.
This set of approximations uses Taylor series expansion in terms of adaptive
intervening variables, overcoming non-linearity problems by using updated
information through the iteration process. TANA2 further includes a correction
term for second-order terms.
3.2.1.2 Second Order Reliability Methods
FORM approximation provides adequate results when the limit-state surface

has only one minimal distance point and the function is nearly linear close to the
design point. For cases where the failure surface has large or irregular
curvatures (high nonlinearity), the failure probability estimated by FORM, using
the safety-index β, may give unreliable and inaccurate results [121]. Introducing
second-order Taylor series expansions (or other polynomials) may overcome
this problem.
Various nonlinear approximate methods have been proposed in the literature. In

[123], [124], [125], [126], [127], and [128] SORM have been developed using
the second order approximation to simplify the original surfaces. In [129] and
[130] second-order approximations are used, forming approximate curvatures in
order to avoid exact second-order derivatives calculations of the limit state
surface.
71
3.2.1.2.1 Orthogonal Transformations
Transformation of the U-space to a rotated new standard normal Y-space would

simplify the integration of the limit state function. This can be achieved by
introducing an orthogonal transformation matrix H. Formulation of this is initiated
by defining an initial matrix as:
−߲݃(ܷ ∗ )⁄߲ܷଵ −߲݃(ܷ ∗ )⁄߲ܷଶ −߲݃(ܷ ∗ )⁄߲ܷ௡

…
⎛ |∇݃(ܷ )| |∇݃(ܷ ∗ )| |∇݃(ܷ ∗ )| ⎞
∗
⎜ 0 1 … 0 ⎟ (3-37)
⎜ 0 0 … 0 ⎟
… … … …
⎝ 0 0 … 1 ⎠
The above matrix can be orthogonalized to obtain H, using several algorithm

such as the Gram-Schmidt algorithm [131]. Denoting ݂ଵ, ݂ଶ, … , ݂௡ for the ݅-th row
vector of the above matrix, respectively:
்
−߲݃(ܷ ∗ )⁄߲ܷଵ −߲݃(ܷ ∗ )⁄߲ܷଶ −߲݃(ܷ ∗ )⁄߲ܷ௡
݂ଵ = ቊ , ,…, ቋ,
|∇݃(ܷ ∗ )| |∇݃(ܷ ∗ )| |∇݃(ܷ ∗ )|
(3-38)
݂ଶ = {0,1,0, … ,0}் , …,
݂௡ = {0,0,0, … ,1}்
Setting:
ଵ 1
‫ܦ‬ଵ = (݂ଵ, ݂ଵ)ଶ , ݁ଵଵ = , ߛ = ݁ଵଵ, ݂ଵ,
‫ܦ‬ଵ ଵ
ଵ
‫ܦ‬ଶ = [(݂ଶ, ݂ଶ) − |(݂ଶ, ߛଵ)|ଶ]ଶ , (3-39)
(௙మ,ఊభ) ଵ
݁ଵଶ = − , ݁ଶଶ = ஽ , ߛଶ = ݁ଵଶ ∙ ߛଵ + ݁ଶଶ ∙ ݂ଶ
஽మ మ
And in a general form:
ଵ
‫ܦ‬௞ = [(݂௞, ݂௞ ) − |(݂௞, ߛଵ)|ଶ − |(݂௞, ߛଶ)|ଶ−, … , |(݂௞, ߛ௞ିଵ)|ଶ]ଶ (3-40)
72
(݂௞, ߛଵ) (݂௞, ߛଶ) (݂௞, ߛ௞ିଵ)
݁ଵ௞ = − , ݁ଶ௞ = − , … , ݁௞ିଵ,௞ = − , (3-41)
‫ܦ‬௞ ‫ܦ‬௞ ‫ܦ‬௞
Where: (݂ , ݂) and (݂ , ߛ) represent the scalar (element by element) product of

two vectors. The generated orthogonal matrix H0 is therefore:
‫ܪ‬଴் = {ߛଵఁ , ߛଶఁ , … , ߛ௡் } (3-42)
The final orthogonal matrix ‫ ܪ‬is derived by moving the first row of the
orthogonal matrix ‫ܪ‬଴ to the last row as:
‫ߛ{ = ் ܪ‬ଶఁ , ߛଷఁ , … , ߛ௡் , ߛଵ் } (3-43)
3.2.1.2.2 First-order Approximation
Parallel to what has been discussed in the presentation of the FORM, assuming
the most probable failure point (MPP) in U-space to be ܷ ∗ = {‫ݑ‬ଵ∗ , ‫ݑ‬ଶ∗ , … , ‫ݑ‬௡∗ }் , the
linear approximation of the response surface ݃(ܷ) = 0 is given by the first-order
Taylor Series expansion at the MPP:
݃෤(ܷ) ≈ ݃(ܷ ∗ ) + ߘ݃(ܷ ∗ ) ∙ (ܷ − ܷ ∗ ) = 0 (3-44)
Considering: ݃(ܷ ∗ ) = 0, on the response surface, and dividing by |ߘ݃(ܷ ∗ )|:
ߘ݃(ܷ ∗ )
݃෤(ܷ) ≈ ∙ (ܷ − ܷ ∗ ) (3-45)
|ߘ݃(ܷ ∗ )|
And therefore, as in (3-26), we obtain:
ߘ݃(ܷ ∗ ) ∙ ܷ ∗
= −ߚ (3-46)
|ߘ݃(ܷ ∗ )|
Back substituting this equation into the expanded first-order Taylor Series
expression:
ߘ݃(ܷ ∗ ) ∙ ܷ ∗
݃෤(ܷ) ≈ ∙ܷ +ߚ=0 (3-47)
|ߘ݃(ܷ ∗ )|
Applying the transformation ‫ ܪ‬into a set of mutually independent standard

normal random variables:
73
ܻ=‫ܷ∙ ܪ‬ (3-48)
And (3-46) becomes:
݃෤(ܷ) ≈ −‫ݕ‬௡ + ߚ = 0 ‫ݎ݋‬, ‫ݕ‬௡ = ߚ (3-49)
The above equation represents the first-order approximation of the response

surface in the Y-space. As in the comparison between FORM and SORM, for
linear or close to linear cases this first order approximation can provide
adequate results. In different cases, second order approximations might be
employed.
3.2.1.2.3 Second-order Approximation
Applying the second-order Taylor series expansion at the MPP for the response
surface ݃(ܷ) = 0 we obtain:
݃෤(ܷ) ≈ ݃(ܷ ∗ ) + ߘ݃(ܷ ∗ )் (ܷ − ܷ ∗ )

1 (3-50)
+ (ܷ − ܷ ∗ )் ߘଶ݃(ܷ ∗ )(ܷ − ܷ ∗ ) = 0
2
Where:
߲ଶ݃(ܷ ∗ )
ߘଶ݃(ܷ ∗ )௜௝ = (3-51)
߲‫ݑ‬௜ ∙ ߲‫ݑ‬௝
Considering: ݃(ܷ ∗ ) = 0, on the response surface, and dividing by |ߘ݃(ܷ ∗ )|:
்
ߘ݃(ܷ ∗ ) 1 ߘଶ݃(ܷ ∗ )
݃෤(ܷ) ≈ ቆ ቇ ∙ (ܷ − ܷ ∗)
+ (ܷ − ܷ ∗ )்
ቆ ቇ ∙ (ܷ − ܷ ∗ ) (3-52)
|ߘ݃(ܷ ∗ )| 2 |ߘ݃(ܷ ∗ )|
Denoting:
ߘଶ݃(ܷ ∗ )
‫=ܤ‬ቆ ቇ (3-53)
|ߘ݃(ܷ ∗ )|
Transformation from ܷ to ܻ-space, the second order approximated limit state

functions may be written as:
74
1
݃෤(ܻ) ≈ −‫ݕ‬௡ + ߚ + ∙ (‫ି ܪ‬ଵܻ − ‫ି ܪ‬ଵܻ∗ ) ∙ ‫ି ܪ( ∙ ܤ‬ଵܻ − ‫ି ܪ‬ଵܻ∗ ) (3-54)
2
Where: ܻ∗ = {0, 0, … , ߚ}் corresponding to the U-space MPP U*. Due to the
transformation in the Y-space, the yn axis is in coincidence with the β vector
geometrically approaching the calculation problem. Applying the orthogoniality
properties of ‫ ܪ‬, ‫ି ܪ‬ଵ = ‫ ் ܪ‬and substituting:
1
݃෤(ܻ) ≈ −‫ݕ‬௡ + ߚ + ∙ (ܻ − ܻ∗ )் ∙ ‫ ܻ( ∙ ் ܪ ∙ ܤ ∙ ܪ‬− ܻ∗ ) (3-55)
2
Where: (ܻ − ܻ∗ )் = (‫ݕ‬ଵ, ‫ݕ‬ଶ, … , ‫ݕ‬௡ − ߚ)்
By a series of orthogonal transformations, ‫ܪ‬ଵ, ‫ܪ‬ଶ, … ‫ܪ‬௠ for the first (n-1)
ത = {‫ݕ‬ଵ, ‫ݕ‬ଶ, … , ‫ݕ‬௡ିଵ}்:
variables, ܻ
ത′ = ‫ܪ‬ଵ‫ܪ‬ଶ, … ‫ܪ‬௠ ܻ
ܻ ത (3-56)
The resultant ‫ܪ‬ഥ‫ܤ‬ത‫ܪ‬ഥ் matrix may be written as:
݇ଵ 0 … 0
⎛ 0 ݇ଶ … 0 ⎞
ഥ ത ഥ்
‫ ⎜ = ܪܤ ܪ‬0 0 … 0 ⎟ (3-57)
… … … …
⎝0 0 … ݇௡ିଵ⎠
Where:
݇௜௝ = (‫ܪ‬ഥ‫ܤ‬ത‫ܪ‬
ഥ்)௜௝, (݅, ݆= 1,2, … , ݊ − 1) (3-58)
Having obtained the ‫ܪ‬ഥ‫ܤ‬ത‫ܪ‬ഥ் matrix, and using Breitung’s formulation [132], the
revised value of the ܲ௙, can be estimated as a correction of the value obtained
by FORM:
௡ିଵ
ܲ௙ ≈ Φ(−ߚ) ෑ (1 + ݇௜ߚ)ିଵ/ଶ (3-59)

௝ୀଵ
Based on the same logic, for the calculation of ܲ௙ different formulations can be
used, such as Tvedt’s formulation, applying two further corrective terms for the
calculation of ܲ௙, providing:
75
௡ିଵ
ଵ
ܲ௙ ≈ Φ(−ߚ) ෑ (1 + ݇௜ߚ)ି ଶ +
௝ୀଵ
௡ିଵ ௡ିଵ
ଵ ଵ
ି
+ [Φ(−ߚ) − ߮(ߚ)] ቐෑ (1 + ݇௜ߚ)ି ଶ − ෑ ൫1 + ݇௜(ߚ + 1)൯ ଶቑ + (3-60)
௝ୀଵ ௝ୀଵ
௡ିଵ ௡ିଵ
ିଵ/ଶ
+(ߚ + 1)[Φ(−ߚ) − ߮(ߚ)] ቐෑ (1 + ݇௜ߚ)ିଵ/ଶቑ − ܴ݁ቐෑ ൫1 + ݇௜(ߚ + 1)൯ ቑ
௝ୀଵ ௝ୀଵ
Although the second order approximation can in general provide more accurate
results for a wider range of limit state functions it involves a considerable
computational cost in the computation of the derivatives of ‫ܤ‬. Especially for
problems with a high number of variables where the size of matrices increases,
this problem becomes even more extensive, yielding for appropriate selection of
calculation method or scheme.
3.2.2 Simulation Methods
Simulation methods have been proposed in literature both for the representation
of statistical distributions, but also for the solution of the complicated integral of
the probability of failure using directly the results from multiple computational
experiments. In the following sections, Monte Carlo Simulation, including
importance sampling, and Latin Hypercube Simulation will be presented as
sampling methods and Design Point, Monte Carlo Simulation, Directional and
Axis-orthogonal Simulation as simulation methods that will be used in later
sections. Presentation of simulation methods will be based on [133], since
PROBAN software will be used for the verification of the codes developed.
3.2.2.1 Sampling Methods
3.2.2.1.1 Monte Carlo Simulation
Monte Carlo Simulation is based on the work of Neumann and Ulam [134]. It
refers to a simple random sampling method that generates random sampling
sets for several types of uncertain variables. During the last decades, this tool
has been developed significantly allowing approximation of the probability of an
76
event that is the output of a stochastic process. Once a distribution type has
been selected, a sampling set is generated that will later serve as input in the
simulations.
The basic concept of the method can be described by a simple problem of

calculation of the area of a specific polygon, as it is shown in Figure 14. The
area can be approximately calculated by the ratio of the number of points that
fall inside the specified area, over the total number of the sample, normalized to
the total area of sampling. The area can also be calculated geometrically, and a
simple sensitivity analysis can show that the greater the sampling number the
more accurate the results. Therefore the reference area is represented by the
probability of falling inside the area for a given number of trials.
100 100
90 90
80 80
70 70
60 60
Lenght Units
Lenght Units
50 50
40 40
30 30
20 20
10 10
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Lenght Units Lenght Units
Figure 14: Area calculation with Monte Carlo Simulation (500 and 5000
sampling points)
In problems of structural reliability, once the stochastic variables have been

identified, sampling sets are generated according to the corresponding
probability density functions. Following, simulations are executed using the
generated sampling sets in order to obtain the response of the structure. In the
area calculation example, the limit state function is represented by the
boundaries of the area to be calculated. For N trials, the probability of failure is
given as:
77
ܰ௙
ܲ௙ = (3-61)
ܰ
Generation of random variables of a stochastic distribution can be realized

based on the inverse transform method. Considering ‫ܨ‬௫ to be the cumulative
distribution function of the random variable ‫ݔ‬௜ with values in the range [0,1].
Assuming ‫ݒ‬௜ to be a generated random variable that follows a uniform
distribution, this method connects ‫ݔ‬௜ and ‫ݒ‬௜ as follows:
‫ܨ‬௫(‫ݔ‬௜) = ‫ݒ‬௜ → ‫ݔ‬௜ = ‫ܨ‬௑ିଵ(‫ݒ‬௜) (3-62)
Figure 15, illustrates the basic concept of the inverse transform method.
Figure 15: Inverse transformation method
3.2.2.1.2 Latin Hypercube Method
The Latin Hybercube Method, initially proposed in [135], is a method that can
represent multiple variables avoiding over-lapping data sets. Application of the
method initiates by dividing the distribution of each stochastic variable in n non-
overlapping intervals with equal probability. For each of the variables, one value
should be randomly selected from each interval and the analysis point obtained
from each dataset is then associated. The homogeneous allocation of intervals
78
on the probability distribution function results in relatively small variance in the
response, compared to the conventional Monte Carlo sampling. At the same
time, the analysis is much less computationally demanding to generate. Figure
16, presents the case of a two variable sampling problem.
Figure 16: Latin Hypercube Method
3.2.2.2 Simulation on Reliability Analysis
3.2.2.2.1 Monte Carlo Simulation Method
The basic formulation of the Monte Carlo Simulation for the calculation of the
probability of success of an event, introduces the indicator function I(x), in the
‫ݔ‬௜-space as follows:
1 if ‫ ≤ )࢞(ܩ‬0
‫ = )࢞(ܫ‬൜ (3-63)
0 if ‫ > )࢞(ܩ‬0
The probability of success of an event is then estimated as the fraction of the

successful realizations of the iterations over the number of samples:
ே
1
ܲ෠ா = ෍ ‫)࢏࢞(ܫ‬ (3-64)
ܰ
௜ୀଵ
79
Figure 17: Monte Carlo Simulation Method
For cases with a large number of random variables or with a very low probability
of failure, a large number of sampling sets is required, hence increasing highly
the computational time and effort. For example, the probability of failure could
be as small as 10-9; this implies that at least a billion simulation repetitions are
required to predict this behaviour. Several techniques may be employed in
order to avoid the inefficiency of direct MCS: importance sampling, subset
simulation, line sampling, etc. Further, variance reduction techniques have been
developed for a dual purpose: to reduce the computational cost and increase
accuracy using the same number of runs [136].
The importance sampling method is a modification of Monte Carlo simulation in

which the simulation is biased for greater efficiency; the sampling is done
primarily in the tail of the distribution to ensure that sufficient simulation failures
occur [137]. Starting from the basic definition of Monte Carlo Simulation for
uncorrelated variables, the mathematical formulation of the method for the
expected value of the probability of failure is described as [138]:
‫ܧ‬൫݃(ܺ)൯= ෍ ݃(‫݂)ݔ‬௑ (‫)ݔ‬ (3-65)

௫∈ఞ
80
Where: ݃(‫ )ݔ‬is the indicator function, and ݂௑ (‫ )ݔ‬the initial statistical distribution.
The Importance Sampling introduces another distribution ‫݌‬௑ (‫ )ݔ‬called ‘the
sampling distribution’ from which samples will be drawn instead of ݂௑ (‫)ݔ‬.
݂௑ (‫)ݔ‬
‫ܧ‬൫݃(ܺ)൯= ෍ ݃(‫݂)ݔ‬௑ (‫ = )ݔ‬෍ ݃(‫)ݔ‬ ‫)ݔ( ݌‬ (3-66)
‫݌‬௑ (‫ )ݔ‬௑
௫∈ఞ ௫∈ఞ
Thus,
௡ ௡
1 ݂௑ (‫ݔ‬௜) 1
‫ܧ‬෨൫݃(ܺ)൯= ෍ ݃(‫ݔ‬௜) = ෍ ݃(‫ݔ‬௜)ܾ(‫ݔ‬௜) (3-67)
݊ ‫݌‬௑ (‫ݔ‬௜) ݊
௜=1 ௜=1
Where:
݂௑ (‫)ݔ‬
ܾ(‫= )ݔ‬ (3-68)
‫݌‬௑ (‫)ݔ‬
The basic idea of importance sampling is to draw from a similar distribution and
then modify the result to correct the bias introduced by sampling the wrong
distribution. For the normal distribution, using a density function with a higher
standard deviation in Monte Carlo analysis leads in more samples being drawn
from the extremes of the distribution [139].
This importance sampling density function satisfies the four following properties
[137]:
 ‫݌‬௑ (‫ > )ݔ‬0 whenever ݂௑ (‫ ≠ )ݔ‬0

 ‫݌‬௑ (‫ )ݔ‬should be close to being proportional to |݂௑ (‫|)ݔ‬
 It should be easy to simulate values from ‫݌‬௑ (‫)ݔ‬
 It should be easy to compute the density ‫݌‬௑ (‫ )ݔ‬for any value ‫ݔ‬
Another variance reduction technique is that of the subset sampling. The main
concept behind this is to express the failure event as a sequence of partial
failure events (subsets) [140]:
‫ܨ‬ଵ ⊃ ‫ܨ‬ଶ ⊃ ⋯ ⊃ ‫ܨ‬௠ = ‫ܨ‬ (3-69)
Then:
81
௞
ሩ ‫ܨ‬௜ = ‫ܨ‬௞ ∀݇ ≤ ݉ (3-70)

௜ୀଵ
Hence the probability of failure is a product of conditional probabilities:
௠ ௠ ିଵ ௠ ିଵ
ܲ௙ = ܲ(‫ܨ‬௠ ) = ܲ ൭ሩ ‫ܨ‬௜൱ = ܲ ൭‫ܨ‬௠ | ሩ ‫ܨ‬௜൱ ܲ ൭ሩ ‫ܨ‬௜൱

௜ୀଵ ௜ୀଵ ௜ୀଵ
(3-71)
௠ ିଵ
= ܲ(‫ܨ‬௠ |‫ܨ‬௠ ିଵ)ܲ ൭ሩ ‫ܨ‬௜൱

௜ୀଵ
And finally:
௠ ିଵ
ܲ௙ = ܲ(‫ܨ‬௠ ) = ܲ(‫ܨ‬ଵ) ෑ ܲ(‫ܨ‬௜ାଵ|‫ܨ‬௜) (3-72)

௜ୀଵ
The determination of a small probability of failure using Monte Carlo simulation

requires high computational effort. The division into subsets allows transferring
the simulation of rare events to a set of simulations of more frequent events
where the probabilities of failure are easier to calculate in terms on
computational effort. The probabilities ܲ(‫ܨ‬ଵ) and ܲ(‫ܨ‬௜ାଵ|‫ܨ‬௜) can be made
sufficiently large so that their estimation can be performed efficiently by direct
Monte Carlo Simulation [141].
3.2.2.2.2 Design Point Simulation
This method was introduced in [142] and it refers to application of MC sampling

around the design point. After initial approximation of the MPP in the u-
dimensional space, Monte-Carlo simulation is executed around this point
instead of running simulations in the wider range of each distribution. As for the
Monte Carlo Simulation, a weighted indicator function for each simulation is
introduced in the sampled u-space point ‫ݑ‬௜ = ݀ + ‫ݒ‬௜ where d is the design point
u* or optionally a point shifted from the design point, and ‫ݒ‬௜ the normal
independent variables from which the Monte-Carlo simulation method samples
from. The indicator function I(u) for each simulation is:
82
1 if ݃(࢛) ≤ 0
‫ = )࢛(ܫ‬൜ (3-73)
0 if ݃(࢛) > 0
The probability of success of an event is then estimated as:
ே ௡
1 ݁‫݌ݔ‬൫∑௝ୀଵ ‫ݑ‬௜ଶ,௝൯
ܲ෠ா = ෍ ‫ݑ(ܫ‬௜) ቌෑ ߪ௝ቍ (3-74)
ܰ ௡ ‫ݑ‬௜,௝ − ݀௜,௝ ଶ
௜ୀଵ ௝ୀଵ ݁‫݌ݔ‬ቆ− ∑௝ୀଵ ൬ ߪ ൰ቇ
௝
Where: di is the design point coordinate, and σi is the standard deviation of the
sampling density. Back substituting v = u - d, the probability of success of an
event is:
ே ௡ ௡
(2ߨ)௡⁄ଶ 1
ܲ෠ா = ߮(݀) ෍ ‫ݑ(ܫ‬௜) ቌෑ ߪ௝ቍ ݁‫݌ݔ‬൮−݀ ‫ݒ‬௜ − ෍ ‫ݒ‬௜ଶ,௝ ቆ1 −
்
ቇ൲ (3-75)
ܰ ߪ௝ଶ
௜ୀଵ ௝ୀଵ ௝ୀଵ
Considering standard deviation to be equal to 1 for standardized variables and d

is the design point:
ே
(2ߨ)௡⁄ଶ
෠
ܲா = ߮(‫ݑ‬ ∗)
෍ ‫ݑ(ܫ‬௜) ݁‫(݌ݔ‬−(‫ݒ ்) ∗ݑ‬௜) (3-76)
ܰ
௜ୀଵ
Figure 18: Design Point Simulation
83
3.2.2.2.3 Directional Simulation
The Directional Simulation method, introduced in [143], samples different

directions, uniformly distributed on the ݊-dimensional surface at the origin of the
‫ݑ‬-space. Different variance reduction methods have been proposed in order to
increase efficiency of the method. The method allows unbiased and efficient
sampling of small probabilities, provided that the number of variables is not
large. The probability of success of an event is expressed as:
ܲா = න ݂௫೙మ (‫ߗ݀ݒ݀)ݒ‬ (3-77)

௚(௨ࢇ)ஸ଴
Where: v is a ‫ݔ‬௡ଶ-distributed random variable, a is a unit vector and dΩ is the

surface element of the ݊-dimensional unit sphere. Each unit direction ai is
sampled and the function ‫ݔ‬௡ଶ conditioned on ܽ is integrated. This is done by
finding the upper bound ‫ݒ‬௨,௜ and the lower bound ‫ݒ‬௟,௜ of the intervals where
݃(‫ ≤ )ࢇݑ‬0 and by adding these contributions to the integral.
ܲ(ࢇ௜) = ෍ ቀ‫ݔ‬௡ଶ൫‫ݒ‬௨,௜൯− ‫ݔ‬௡ଶ൫‫ݒ‬௟,௜൯ቁ (3-78)

௜ୀଵ
The estimated probability of occurrence of the event sampled in ܰ total

directions is:
ே
1
ܲ෠ா = ෍ ܲ(ࢇ࢏) (3-79)
ܰ
௜ୀଵ
84
Figure 19: Directional Simulation
3.2.2.2.4 Axis-Orthogonal Simulation
The method, proposed in [144] and [145] defines an axis for a small intersection
domain, and a sampling density in a plane orthogonal to this axis. The method
approximates the boundaries of an event by a set of linear surfaces obtained by
using the FORM linearization of small intersection domains. The probability of
the linearized domain is obtained by methods available for the multi-normal
distribution. The axis-orthogonal simulation method estimates an approximation
of the true probability; the quality of this approximation follows that of the quality
of the multi-normal integral. A method for correction of the approximate results
is described in [146].
Considering: ܽ௔ , to be the averaged gradient of a set of normalized gradients ܽ௜

for the single events, pointing into the interior of at the design point as:
∑௡௜ୀଵ ܽ௜
ܽ௔ = − (3-80)
ห∑௡௜ୀଵ ܽ௜ห
A new axis is defined rotating the initial coordinates as:
ܽ௨ = ‫ݑ‬
෤଴ + ‫ݑ‬௡ ݁௡ (3-81)
Where: ‫ݑ‬
෤଴ is the design point in the new coordinates and ݁௡ is the unit vector in
direction of ‫ݑ‬௡ . The probability of the event ‫ ܧ‬is calculated as follows:
85
ܲ஺ (‫ݑ‬
෤)
ܲா = ܲ௅ න ‫ܪ‬ெ (‫ݑ‬
෤)݀‫ݑ‬
෤= ܲ௅‫ܥ‬ (3-82)
ߔ ൫−‫ݑ‬௡ (‫ݑ‬෤)൯
௚(௨)ஸ଴
Where: ܲ௅ is the probability of the linearized domain, ‫ ܥ‬is the multiplicative

correction, ‫ݑ‬௡ (‫ݑ‬
෤) is the intersection of the linearized domain and
‫ݑ(ܫ‬௡ ) = ‫ݑ‬
෤+ ‫ݑ‬௡ ݁௡ (3-83)
ܲ஺ (‫ݑ‬
෤) is the true probability of the event conditioned on the line defined by I(un).
The estimator for ‫ ܥ‬is:
ே
1 ܲ஺ (‫ݑ‬
෤)
‫ܥ‬መ= ෍ (3-84)
ܰ ߔ ൫−‫ݑ‬௡ (‫ݑ‬෤௜)൯
௜ୀଵ
The integral estimator is implemented:
ܲ஺ (‫ݑ‬
෤) = න ߔ (‫ݑ‬௡ ) ݀‫ݑ‬௡ (3-85)
෥,௨೙ )ஸ଴
௚(௨
The integral is evaluated through a search for the points un(Ii) and un(ui)
denoting the lower and upper bounds of the intervals where is less than zero,
and then to sum up for ݉ intervals as:
ܲ஺ (‫ݑ‬
෤) = ෍ ቀߔ ൫‫ݑ‬௡ (‫ݑ‬௜)൯− ߔ ൫‫ݑ‬௡ (݈௜)൯ቁ (3-86)
௜ୀଵ
86
Figure 20: Axis-Orthogonal Simulation
3.3 Stochastic Response Surface Method
3.3.1 General Concept
In complex three-dimensional structures, such as a jacket structure, A

mathematical relationship between the actual loading acting on the whole
structure (eg. wave or wind loads) and the actions that each member is
subjected to (eg. axial force and bending moments) is difficult to be explicitly
expressed. For such cases of complicated failure processes, simulation
techniques can deal with the complexity of the problem; however, they are often
inefficient for the calculation of small values of probability of failure, since a
great number of iterations is required until sufficient results are derived. In [147]
an analytical procedure from Rubinstein is presented. Alternative procedures for
the calculation of the optimum sampling number can also be found in [148],
[149], but comprehensive presentation is beyond the scope of this Thesis.
For such cases, where simulation techniques are computationally intensive, the
stochastic response surface method (SRFM) [150], [151], can provide an
accurate estimation of structural reliability, regardless of the complexity of the
system under consideration. The concept of this method is the approximation of
the actual limit state function, which in some cases can be unknown, using
87
simple and explicit mathematical functions of the random (stochastic) variables
affecting the response of the structural member or system. Those functions can
be simple polynomials (eg. second or of higher order) with coefficients that can
be calculated by fitting the response surface function to a number of sample
points from calculation of the response of the member. In this explicit
expression of the limit state function First and Second Order Reliability Methods
can be applied for the estimation of the reliability index and therefore the
probability of failure. Further, although the number of variables is the same in
the response surface function and the initial limit state function, simulation
techniques are more computationally efficient since this expression is less
complicated than matrix manipulation.
Limitations of the Stochastic Response Surface Method arise in cases where

the initial limit state includes non linearities or in cases where very low
probabilities of failure should be accurately calculated. The above comments
are highlighted in [152] and [153], and are caused due to the improper
representation of the response surface based on arbitrary sample points that
might be relatively far from the MPP. In order to overcome these restrictions,
several methods have been proposed based on the adaptation of the response
surface function to the location close to the design point, as this will be indicated
by the FORM, are included in [154], [155], [156], [157], [158], and will be
presented later in this chapter. In [159], the accuracy of a highly non-linear limit
state depends on the initial selection of sampling points.
In most cases, the order of polynomials that is selected for the approximation of
the response surface function is 2 (quadratic terms) since it demands few
sample points - (2݊ + 1) - for the approximation of the coefficients of the
function. In [159], the use of higher order polynomials is investigated in depth.
The disadvantage of this practice is that it needs more sampling points,
requirement that is not always feasible, due to difficulty in the computational
process and the fact that ill conditioned matrices are structured for the
derivation of the coefficients of the polynomials through regression [160], [161].
The latter problem can be overcome through use of Chabyshev polynomials
88
and statistical analysis of the high-order response surface [160]. In Figure 21, a
chart representing the regression coefficient based on the Least Square Method
(LSM) and the R2 criterion is presented for the estimation of the stress (v.
Mises) response of a structural member under combined loading, modelled with
three stochastic variables and using 25 sampling points from the original limit
state function. The oscillation of the regression coefficient due to the quality of
the regression design matrix can be observed.
0.99
0.985
0.98
0.975
R2
0.97
0.965
0.96
0.955
0 1 2 3 4 5
Order of Polynomial
Figure 21: Regression vs. Order of Polynomial
Figure 22: Different Sampling Approaches [160]
In [162], an algorithm is presented that uses quadratic response surface

obtained from central composite designs (ARERSA - Adaptive Reliability
Estimation Response Surface Algorithm). After a global search and once the
89
most likely failure point domain is identified, an updated response surface is
fitted locally, in order to apply the reliability calculation routines. Complementary
checks for valid solutions might be executed in order to ensure validity of the
results. In [163], an algorithm is proposed (RSAED - Response Surface with
Adaptive Experimental Design) which approximates the response surface in the
standardized u-space. This algorithm uses previous results and calculations
and reduced numerical procedures. In cases of highly non linear target limit
state surfaces, the ‘gradient projected technique’ [164] would approximate the
actual limit state surface with a linear one. Further to this work, a ‘cumulative
response surface method’ has been proposed [165] according to which a
design point is calculated based on FORM and an initial linear response
surface. Second order terms are then employed to more accurately
approximate the limit state function and SORM should be used to find the new
design points. Further refinement of the response surface is done in the area
around the current design point.
Most of the methods and algorithms that are available in literature, concentrate
of the use of polynomials with quadratic terms due to the simplicity of the linear
systems and the limited sample points required. In [166], a method is proposed
that aims to reduce the number of samples based on statistical properties of
polynomials to account for the dependence of the stochastic variables with the
response of the system.
3.3.2 Review and Notation
Considering a vector ܺ = [‫ݔ‬ଵ, ‫ݔ‬ଶ, … , ‫ݔ‬௡ ] to be the vector containing ݊ stochastic

variables of a system, and ݃(‫ )ݔ‬to be the limit state function that represents the
critical failure surface, the probability of failure of the system is described as
ܲ௙ = ܲ[݃(‫ < )ݔ‬0]. The approach of the response surface method, introduces a
new polynomial function ݃෤(‫)ݔ‬, of ݇-th order, that will use adequate sample
points to determine the coefficients of the polynomial. The most common
notation of the function, as it has been initially proposed [150] uses second
order – quadratic – polynomials:
90
௡ ௡
݃෤(‫ ܽ = )ݔ‬+ ෍ ܾ௜ܺ௜ + ෍ ܿ௜ܺ௜ଶ (3-87)

௜ୀଵ ௜ୀଵ
In order to calculate the (2n+1), ܽ, ܾ௜ and ܿ௜ coefficients, an equal or a greater

number of sample points is required. Determination of sampling points can be
done in a way that the measured response is mapped in a better way.
According to different methods, the number of samples can be from (2n+1) to
3n, commonly combining ߤ௜ and ߤ௜ ± ݂ߪ௜ where ߤ௜ and ߪ௜ are the mean value
and standard deviation of stochastic variable ܺ௜, and ݂ is a coefficient (typical
value is 3). Figure 22, presents different patterns of samples combinations for a
two variables problem [160].
In cases of linear limit states, selection of parameter ݂ is less significant than for
those with a non linear performance. In the later cases, selection of sampling
points is significant for the approximation of the initial limit state functions.
Another expression of the generic polynomial approximation of quadratic limit
states is presented in [167], and includes mixed terms of the stochastic
variables. Although this expression increases the complexity of the design
matrix for the calculation of the polynomial coefficients, it may capture non-
linearity of a limit state. The required number of sample points among the ݊-
௡(௡ିଵ)
dimensional space is between ቀ + 2݊ + 1ቁ and (3௞), and the
ଶ
corresponding mathematical formulation:
௡ ௡ ௡ିଵ ௡
݃෤(‫ ܽ = )ݔ‬+ ෍ ܾ௜ܺ௜ + ෍ ܿ௜ܺ௜ଶ +෍ ෍ ݀௜௝ܺ௜ܺ௝ (3-88)

௜ୀଵ ௜ୀଵ ௜ୀଵ ௝ୀ௜ାଵ
Table 5, presents an example of approximation of a quadratic limit state

function of 3 stochastic variables, approached by (2n+1)=7 and 3n=27
neglecting mixed terms since, for the problem considered (response of a
member), linear performance was expected. Reliability index was calculated
with FORM and SORM Methods. From those results it can be observed that the
smaller sample can approximate the limit state for reliability index calculation
introducing an error in estimation.
91
β Pf
N FORM SORM FORM SORM

n
3 4.6661 4.6488 1.54E-06 1.67E-06
3n+1 4.2853 4.2853 9.13E-06 9.13E-06
Table 5: Reliability Index and Probability of Failure vs. Sampling Number
3.3.3 Adaptive Response Surface Method
In [168], a procedure of weighted regression is been presented and in [169]

examples are examined in order to illustrate the benefits of the method. An
initial sample, as it was been described in Figure 22, ‘maps’ the whole range of
each variable, and the sample points contribute equally in the formation of the
design matrix for the calculation of the regression coefficients. Conventional
formulation of the least square method, as it can be found in [170] or other
regression analysis handbooks, denotes that for quadratic approximation and ‫݌‬
sampling points, the design matrix ‫ ܯ‬௨ of the independent variables and the
vector ݃෤with values of the response (dependent variables) as:
1 ‫ݑ‬ଵଵ . ‫ݑ‬௡ଵ ଶ ଶ
‫ݑ‬ଵଵ . ‫ݑ‬௡ଵ ‫ݑ‬௜ଵ ∙ ‫ݑ‬௝ଵ
‫ ܯ‬௨ = ቎. . . . . . . . ቏
1 ‫ݑ‬ଵ௣ . ‫ݑ‬௡௣ ଶ ଶ
‫ݑ‬ଵ௣ . ‫ݑ‬௡௣ ‫ݑ‬௜௣ ∙ ‫ݑ‬௝௣
(3-89)
݃̅ = [݃ଵ . ݃௣ ]்
Calculation of coefficients vector ܽ

ത, based on normal regression, is obtained by
solving the following linear system:
ത= (‫ ܯ‬௎் ‫ ܯ‬௎ )ିଵ‫ ܯ‬௎் ݃̅

ܽ (3-90)
The weighted regression introduces a diagonal weighted matrix ܹ ீ that gives

greater weights in the points closer to the limit state function. Therefore, the
equation in a matrix form, that includes the weighted regression would become:
ത= (‫ ܯ‬௎் ܹ ீ ‫ ܯ‬௎ )ିଵ‫ ܯ‬௎் ܹ ீ ݃̅

ܽ (3-91)
92
According to [168], for the construction of ܹ ீ , after an initial set of simulations,
a limit state function is obtained and the best design value ‫ݕ‬
ො௕௘௦௧ (or ݃௠ ௜௡ ) is
selected as the one that minimizes the initial response surface:
‫ݕ‬
ො௕௘௦௧ = ݉ ݅݊ห݃′௫ห (3-92)
The ‫ ݓ‬௜ element in the diagonal weight matrix ܹ , would be obtained as:
݃′(‫)ݔ‬௜ − ‫ݕ‬
ො௕௘௦௧
‫ ݓ‬௜ = ݁‫݌ݔ‬ቆ− ቇ (3-93)
‫ݕ‬
ො௕௘௦௧
Based on this weight matrix, weighted regression formulas can be executed,

correcting the initial approximation. This formulation of the weight coefficients
has the drawback that if one of the sample points lies on or very close to the
limit state function, ‫ݕ‬
ො௕௘௦௧ ≈ 0 and therefore the regression system becomes ill
conditioned, providing inconsistent results. In order to overcome this problem,
two different expressions for the weight coefficients are proposed in [169].
Initially, a different reference point is selected rather than ‫ݕ‬
ො௕௘௦௧. This is the value
of the limit state function at the origin of the standardized space ݃(‫ݑ‬଴), which for
commonly measured probabilities of failure is different than zero. Therefore:
݃′(‫)ݔ‬௜ − ݃(‫ݑ‬଴)
‫ீ ݓ‬௞ = ݁‫݌ݔ‬ቆ− ቇ (3-94)
݃(‫ݑ‬଴)
Further optimization of the procedure, based on CQ2RS (Complete Quadratic

Response Surface with Resampling) Method [171], imports an additional
‘penalty factor’. For each iteration of the reliability calculation procedure, the
relative distance ‫ܦ‬௞ between the sampling points and the current design point
ܲ is used to form the weight factors as:
‫ܦ‬௞ଶ
‫ ݓ‬஽௞ = ݁‫݌ݔ‬ቆ− ቇ (3-95)
2
The resulting weight matrix for the linear regression system is:
ܹ ீ = ‫ீ ݓ‬௞ ∙ ‫ ݓ‬஽௞ (3-96)
93
3.3.4 Algorithms of the Stochastic Response Surface Method
Based on the work that has been carried out so far, algorithms that combine
FORM/SORM with linear and quadratic response surfaces approximation are
developed. The normal Response Surface Method is initiated by formulation of
combinations of different sets of variables, as presented in Figure 22. From the
responses that are obtained, quadratic regression can be executed and the limit
state surface can be formulated. This algorithm provides sufficient results for
cases where the response of the system under investigation behaves linearly as
a function of the basic stochastic variables. Especially for cases such as the
complex jacket structure that will be investigated in the next chapters, where
simulations are obtained by an external software package, this method provides
efficient results both in accuracy and in computational effort considering that the
number of simulations can be obtained by (2݊ + 1) combinations Figure 22 (a).
In cases where the simulation procedure can be programmed to automatically

execute new simulations, the adaptive Response Surface Method can be
formed, which can perform better in cases of non-linear limit state functions and
accurate calculation of small probabilities. The first step of application of the
method suggests characterization of main variables as either favourable
(resistance variable) or unfavourable (load variables). In order to achieve this, a
reference value ݃(ܺത) should be obtained by calculating the value of the limit
state function for the mean values of the stochastic variables ܺത. For each of the
variables a value of ‫ݔ‬௜ = ‫ݔ‬
ഥప − ݂ߪ௜ (value for ݂ can be selected as 3 in order to
represent a point at the tail of the distribution) is calculated, and the comparison
|݃(ܺ)| < |݃(ܺത)| will mark for the variable an indicator -1 else wise +1. The
above procedure will provide (n+1) points that will allow linear regression to be
executed.
Having obtained an initial, linear response surface function based on samples

that lie within the whole range of each variable, a first iteration of FORM will run,
based on the mean values of the stochastic variables. An initial value for ߚ and
a new design point can then be obtained as:
94
‫ݔ‬ଵ = ‫ݔ‬
ഥప + ߚߪ௫೔ܽ௜
߲݃(ܷ ∗ ) ߲݃(ܺ ∗ )
݃(ܷ ∗ ) − ∑௡௜ୀଵ ∙ ߪ௫೔ ∙ ‫ݑ‬௜∗ ∙ ߪ௫೔
߲‫ݔ‬௜ ߲‫ݔ‬௜ (3-97)
=‫ݔ‬
ഥప + ߪ௫೔ ଵൗ
ଶ ଶ ଶ
߲݃(ܺ ∗ )
ඨ ∑௡௜ୀଵ ൬߲݃(ܷ ) ∙ ߪ௫ ൰
∗
ቈ∑௡௜ୀଵ ൬ ∙ ߪ௫೔൰ ቉
߲‫ݔ‬௜ ೔ ߲‫ݔ‬௜
Based on this new series of n points combined with the mean value point and
including the previous (n+1) points, quadratic regression can be executed. A
weight matrix will account for the contribution of each of the sample points to
the actual response surface giving a higher weight to the points closer to the
design point.
݃′(‫)ݔ‬௜ − ݃(‫ݔ‬ଵ)
‫ீ ݓ‬௞ = ݁‫݌ݔ‬ቆ− ቇ (3-98)
݃(‫ݔ‬ଵ)
Following this procedure of a dynamically constructed response surfaces and

considering that linear limit state functions converge after only a few iterations,
this procedure can provide adequate results in the calculation of FORM/SORM
reliability index.
3.4 Regression methods
3.4.1 Linear Regression
In the problem where two (or more) variables are required to be expressed as a
function, linear regression is the fundamental concept to follow. In an
experimental procedure this can represent the problem of correlating
measurements to properties. Linear regression refers to approaching the
dependent variable as a linear function of some parameters (independent
variables); otherwise regression should be characterized non-linear.
Graphically, this approach assumes that the plotted sets of dependent and
independent variables can be represented efficiently with one straight line. The
earlier method proposed by Gauss and Legendere is referred to as the Least
Squares Method (LSM), and provides a solution by minimizing the absolute
95
distance between the data provided and the potential function (residuals) to find
the optimum fit. Following a mathematical notation it can be expressed as:
‫ܽ = )ݔ(ݕ‬௢ + ܽଵ ∙ ݂ଵ(‫ )ݔ‬+ ܽଶ ∙ ݂ଶ(‫ )ݔ‬+ ⋯ + ܽ௩ ∙ ݂௩(‫ )ݔ‬+ ݁ (3-99)
Where, ܽ௜ ݅= 1,2, . . . , ‫ ݒ‬is the regression coefficient vector and ݁ the error of the
model equation. Forming the above equation in a matrix form:
ܻ= ܺ∙ߙ+݁ (3-100)
Where,
‫ݕ‬ଵ 1 ݂ଵ(‫ݔ‬ଵ) ݂ଶ(‫ݔ‬ଵ) … ݂௠ (‫ݔ‬ଵ) ܽଵ ݁ଵ

‫ݕ‬ 1 ݂ଵ(‫ݔ‬ଶ) ݂ଶ(‫ݔ‬ଶ) … ݂௠ (‫ݔ‬ଶ) ܽଶ ݁ଶ
ܻ = ቎ ଶ቏, ܺ = ൦ ൪, ܽ = ቎ ቏, ݁ = ቎ ቏
⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
‫ݕ‬௡ ܽ ݁
1 ݂ଵ(‫ݔ‬௡ ) ݂ଶ(‫ݔ‬௡ ) … ݂௠ (‫ݔ‬௡ ) ௡ ௡
The least squared method, expressed in a matrix form, is expressed as follows

in order to derive the regression coefficients vector ܽ:
ܽ = (ܺ ் ∙ ܺ)ିଵ ∙ ܺ ் ∙ ܻ (3-101)
Having calculated the regression coefficients, the values of the dependent

variables for the sampled dependent ones and the error for each of them is:
ത = ܺ ∙ ܽ and ݁ = ܻ − ܻ
ܻ ത (3-102)
The total sum of squares SST, regression sum of squares SSR and error sum
of squares SSE are calculated as:
ܵܵܶ = ்ܻ ∙ ܻ
ܴܵܵ = ത
ܻതത்ത∙ ܻ
ത = ்ܽ ∙ ܺ ் ∙ ܻ (3-103)
ܵܵ‫ ܶܵܵ = ܧ‬− ܴܵܵ
In order to evaluate the level of accuracy of the modelled equation, a coefficient

of determination (ܴଶ) can be calculated as follows. The practical meaning of this
equation, implies that when the regression sum of errors equals zero, and
96
therefore ܴଶ = 1 the modelled function satisfies all of the sets of (‫ݕ‬, ‫ݔ‬௜) and
therefore absolute regression has been achieved.
ܵܵ‫ܧ‬
ܴଶ = 1 − (3-104)
ܵܵܶ
3.4.2 Multivariate Regression
For the case where more than one independent or dependent variables are
present, the fundamental equation can be solved providing adequate sets of
(‫ݕ‬, ‫ݔ‬௜). The general problem can be described as:
‫ = )ݔ(ݕ‬෍ ࢇ࢏ ∙ ‫݌‬௜(‫ݔ‬ଵ, ‫ݔ‬ଶ, … , ‫ݔ‬௡ ) + ݁ (3-105)

௜
Considering monomials, this can also be described as:
‫ = )ݔ(ݕ‬෍ ࢇ࢏ ∙ ‫ݔ‬ଵఈ೔ ∙ ‫ݔ‬ଶఉ೔ … ‫ݔ‬௡ ఠ ೔ + ݁ (3-106)

௜
Where: ࢇ࢏ are the regression coefficients and ߙ௜,ߚ௜ ,…, ߱ ௜ are the power
coefficients for the independent variables.
For a case where the maximum monomial degree is 2, with 2 independent

variables, the expression can be rewritten as:
‫ܽ = )ݔ(ݕ‬௢ + ܽଵ ∙ ‫ݔ‬ଵ + ܽଶ ∙ ‫ݔ‬ଶ + ܽଷ ∙ ‫ݔ‬ଵଶ + ܽସ ∙ ‫ݔ‬ଶଶ + ܽହ ∙ ‫ݔ‬ଵ ∙ ‫ݔ‬ଶ + ݁ (3-107)
Considering ܻ to be a (݊ × ‫ )ݍ‬data matrix containing the dependent variables, ܺ

to be a (݊ × ‫ )݌‬data matrix containing the independent variables, ‫ ܣ‬a (‫)ݍ × ݌‬
data matrix with the regression coefficients and ‫ ܧ‬is (݊ × ‫ )ݍ‬matrix with the error
terms. It forms the above equation in a matrix form:
ܻ = ܺ෨∙ ‫ ܣ‬+ ‫ܧ‬ (3-108)
Where ܺ෨ denotes a matrix formed from X, containing the different powered

values of X.
97
The above dimensions of the participating matrices imply that in order for the
system to have a solution, (‫ )ݍ × ݌‬sets of data should be available. An
important observation that can ensure accuracy in the regression coefficients
results is the level of how well conditioned the matrix ܺ ் ∙ ܺ is.
3.4.3 Alternative Regression Methods
For a higher quality of response surface, schemes such as the central

composite design method [172] might be employed allowing redundancy in
some of the data sets. Drawback of this scheme is that it requires a large
number of data (N=2n+2n+1) in order to approach the exact limit state function
with a quadratic equation by regression. Increase in the number of variables
exponentially increases the required computational effort. A different approach,
suitable for asymptotic behaviour of the structural behaviour is the selection of
inverse polynomials as interpolating functions [173] providing good regression
results but increasing the complexity of the problem.
A different method for regression analysis is ‘kriging’, which is a procedure for

constructing a minimum error variance linear estimate at a location where the
true value is unknown. Initially used for geological applications its use has been
extended to a wide variety of different applications including aerodynamics,
structures, and multi objective problems [174]. Extensive background theory of
the method can be found in [175] and [176] while an initial application in the
response surface method is presented in [177] for the solution of the problem of
the reliability analysis of damaged steel structures using finite element analysis.
Corresponding to the common regression expressions, the predicted results
from the kriging model may be obtained from:
ො= ߚመ+ ‫ିܴ)ݔ( ்ݎ‬ଵ൫‫ ݕ‬− ݂ߚመ൯

‫ݕ‬ (3-109)
Where ’‫ ’ݕ‬is predicted response value at x, ’x’ are the sample points, ’ ߚ’ is the
constant underlying global portion of the kriging model, ’ ‫ ’ݎ‬is the correlation
vector of length n, ’ ܴ’ is an n×n symmetric matrix with ones along the diagonal
98
and ’ ݂’ is column vector of length n. The method can in general approach
complicated surfaces with multiple variables.
3.5 Numerical model Developed
3.5.1 Description
The algorithm that has been implemented for the reliability calculation in the
application sections of this Thesis is based on the methodology that has been
presented in paragraph 3.3 and 3.4. MATLAB was selected as the proper
programming language for this scope due to its simplicity and the wide range of
tools that are available in order to simplify and optimize the programming
procedure.
In the case when a methodology is proposed and a new code is conducted,

verification is a very important step towards establishing its accuracy and
efficiency. In the next sections, the algorithm that has been proposed using a
normal and an adaptive response surface and FORM/SORM reliability methods,
will be compared to results obtained by direct simulation for the problem of a
simple 3-dimensional truss under stochastic loading. For this to be achieved an
FEA code will be initially conducted and the corresponding sequence of steps
will be followed. The Matlab FEA code has been verified for a deterministic case
using the commercial software ABAQUS (and DNV SESAM Genie), while the
reliability estimation routine has been verified according to DNV SESAM
PROBAN software for different cases of estimation of the probability of failure
within different range of values and using different numerical techniques. In all
cases the algorithms and the codes have been found to provide adequately
good results. Figure 23 to Figure 25 present in a block diagram form the
reliability analysis procedure based on direct simulation as well as the normal
and adaptive Response Surface Method.
99
DEFINITION OF VARIABLES AND
Chapter 4
FEA MODEL DIMENSIONS
DEFINITION OF SAMPLING NUMBER

Par. 3.2.2.1
GENERATION OF SAMPLING SETS
FROM INDEPENDENT VARIABLES
SOLUTION OF THE FEA PROBLEM

FOR EACH DATA SET AND
Par. 3.5.2.1
CALCULATION OF THE LIMIT STATE
FUNCTION
FROM THE VALUES OBTAINED

DERIVE THE PROBABILITY OF Par. 3.2.1
FAILURE AND ANALYTICALLY β
Figure 23: Calculation of reliability based on Direct Simulation
DEFINITION OF VARIABLES AND

Chapter 4
CREATE (2n+1) SETS OF

Par. 3.3.1
VARIABLES (xi±fσi), xi
SOLUTION OF THE FEA PROBLEM

FOR EACH DATA SET AND
Par. 3.5.2.1
CALCULATION OF THE LIMIT STATE
FUNCTION
APPLY REGRESSION FOR (2n+1) DESIGN

Par. 3.3.1
POINTS
APPLY ITERATIVE FORM/SORM AND Par. 3.2.1

CALCULATE β
Figure 24: Calculation of reliability based on Normal Response Surface

Method
100
DEFINITION OF VARIABLES AND Chapter 4
Par. 3.3.1-
OBTAIN RESPONSE FOR INITIAL DESIGN MATRIX
3.5.2.1
APPLY LINEAR REGRESSION IN (n+1) POINTS Par. 3.3.1
CALCULATION OF β AND NEW

Par. 3.2.1
DESIGN POINTS
OBTAIN RESULTS FOR (N) NEW POINTS

Par. 3.3.3
APPLY WEIGHTED REGRESSION FOR
(2n+1) DESIGN POINTS
APPLY ITERATIVE FORM/SORM AND

Par. 3.2.1
CALCULATE β
Figure 25: Calculation of reliability based on Adaptive Response Surface

Method
101
3.5.2 Verification Process
3.5.2.1 FEA Code
3.5.2.1.1 Basic Theory of FEA
The FEA code that is conducted is based on the fundamental idea of the Finite
Element Analysis method to numerically find approximate solutions of the partial
differential equations (or integral equations) that model the system. Basic
literature can be found in [178], [179] and [180]. Here the case of a 3
dimensional frame structure will be considered. Based on linear interpolation
and matrix calculations, programming of this method is feasible and obtaining of
a solution with a low computational cost. The most significant operation in
computational effort is the derivation of the inverse of a square matrix. The
background theory modelled is mainly based on [181].
For a 3D frame two-node element, each node has 6 global degrees of freedom:
3 in translation, 3 in rotation. A vector of displacements in the local coordinates
system can be formulated as:
்࢛ = [‫ݑ‬௜௫ ‫ݑ‬௜௬ ‫ݑ‬௜௭ ߠ௜௫ ߠ௜௬ ߠ௜௭ ‫ݑ‬௝௫ ‫ݑ‬௝௬ ‫ݑ‬௝௭ ߠ௝௫ ߠ௝௬ ߠ௝௭] (3-110)
Following an analytical method starting from single beams, moving to 2D

frames and extending to 3D frames, the derived stiffness matrix is defined in
local coordinates as:
102
ࡷ
‫ܣܧ‬ ‫ܣܧ‬
⎡ 0 0 0 0 0 − 0 0 0 0 0 ⎤
⎢ ‫ܮ‬ ‫ܫܧ‬ ௭
‫ܮ‬ ‫ܫܧ‬௭ ⎥
‫ܫܧ‬௭ 0 0 6 ଶ ‫ܫܧ‬௭ 0 0 6 ଶ ⎥
⎢ 12 ଷ 0 ‫ܮ‬ 0 −12 ଷ 0 ‫ܮ‬
⎢ ‫ܮ‬ ‫ܫܧ‬௬ ‫ܮ‬ ‫ܫܧ‬௬ ⎥
⎢ ‫ܫܧ‬௬ 0 −6 ଶ 0 ‫ܫܧ‬ ௬ 0 −6 ଶ 0 ⎥
12 ଷ ‫ܮ‬ 0 0 −12 ଷ ‫ܮ‬
⎢ ‫ܮ‬ ‫ܮ‬ ⎥
⎢ ‫ܬܩ‬ ‫ܬܩ‬ ⎥
⎢ 0 0 0 0 0 − 0 0 ⎥
‫ܮ‬ ‫ܫܧ‬௬ ‫ܮ‬
⎢ ‫ܫܧ‬௬ 0 0 6 ଶ ‫ܫܧ‬௬ ⎥
⎢ 4 0 ‫ܮ‬ 0 2 0 ⎥
⎢ ‫ܮ‬ ‫ܫܧ‬௭ ‫ܮ‬ ⎥
‫ܫܧ‬௭ 0 −6 ଶ 0 ‫ܫܧ‬௭
⎢ 4 ‫ܮ‬ 0 0 2 ⎥ (3-111)
=⎢ ‫ܮ‬ ‫⎥ܮ‬
⎢ ‫ܣܧ‬ ⎥
0 0 0 0 0
⎢ ‫ܮ‬ ‫ܫܧ‬௭ ⎥
⎢ ‫ܫܧ‬௭ 0 0 −6 ଶ ⎥
⎢ 12 ଷ 0 ‫ܮ‬ ⎥
‫ܮ‬ ‫ܫܧ‬௬
⎢ ‫ܫܧ‬௬ 0 6 ଶ 0 ⎥
⎢ 12 ଷ ‫ܮ‬ ⎥
‫ݕݎݐ݁ ݉ ݉ݕݏ‬ ‫ܮ‬
⎢ ‫ܬܩ‬ ⎥
⎢ 0 0 ⎥
⎢ ‫ܮ‬ ⎥
⎢ ‫ܫܧ‬௬ ⎥
4 0
⎢ ‫ܮ‬ ⎥
⎢ ‫ܫܧ‬௭ ⎥
⎣ 4 ⎦
‫ܮ‬
Where: ‫ ܩ‬is the shear modulus and ‫ܬ‬the polar moment of inertia. Adapting the
above to the global coordinates system, the following basis for transformation
may be used:
࢛࢒࢕ࢉࢇ࢒ = ࡹ ࢛ࢍ࢒࢕࢈ࢇ࢒ (3-112)
݉ 0 0 0
ࡹ = ቎0 ݉ 0 0቏ (3-113)
0 0 ݉ 0
0 0 0 ݉
‫ܥ‬௑௫ ‫ܥ‬௒௫ ‫ܥ‬௓௫

݉ = ቎‫ܥ‬௑௬ ‫ܥ‬௒௬ ‫ܥ‬௓௬ ቏
‫ܥ‬௑௭ ‫ܥ‬௒௭ ‫ܥ‬௓௭ (3-114)
‫ܥ‬௑௫ = cos (ߠ௑௫)
Where: angles ߠ௑௫ , ߠ௒௫ and ߠ௓௫ are measured from global axis ܺ, ܻ and ܼ with
respect to the local axis ‫ݔ‬, respectively. Hence the stiffness matrix in global
coordinates system is defined by:
ࡷ ࢍ࢒࢕࢈ࢇ࢒ = ࡹ ் ࡷ ࢒࢕ࢉࢇ࢒ࡹ (3-115)
103
For a system composed of bars and beams, once the stiffness matrix ‫ ܭ‬is
derived, the vector of nodal forces ݂ to statically represent the system is:
ࡷ ࢛=ࢌ (3-116)
where is ‫ ݑ‬is the displacement vector.
This system can be solved to obtain the displacement vector u:
࢛ = ࡷ ି૚ࢌ (3-117)
Many processes can be used to solve easily and quickly this system: Cholesky
resolution, QR or LU decomposition.
Further post processing can transform displacements and rotations to axial

forces and bending moments, following a procedure that can be widely found in
literature.
3.5.2.1.2 Verification Results
Figure 26 presents the hypothetical truss structure that has been considered for
the verification process. This is a 3D structure, fixed at the bottom, with lateral
(e.g. wind forces) and vertical forces (e.g. weight loads). Here the structure is
considered only in tension-compression. Table 6, presents the four stochastic
variables, normally distributed, considered in this application; ࣨ (ߤ, ߪ)
represents a normal distribution of mean ߤ and standard deviation ߪ.
The first step is to verify the ‘core’ of the whole procedure: the Finite Element
Analysis. The code has been verified using the software ABAQUS on this
simplified structure. The purpose was to compare the axial stresses obtained
with both methods for the mean (deterministic) values of the stochastic
variables. The stresses are listed in Table 7 to compare the results between
ABAQUS and the MATLAB code, while Figure 27 illustrates the relative
deformations of the two cases.
104
Figure 26: Reference structure
Parameters Probability’s law
Loads F × N(1, 0.2)
Elasticity N(21×1010, 1×1010)
Area A × N(1, 0.01)
Allowable stresses N(100000, 10000)
Table 6: Stochastic Loads Consideration
105
12
10
0 4
2
6 4 0
2 0
Figure 27: ABAQUS and Matlab code deformed model
Element Matlab Abaqus Element Matlab Abaqus

1 0 -1.6676E-11 21 0 1.1117E-11
2 0 2.2235E-11 22 64031.2 64031.2
3 40000 40000 23 1E-10 -2.7793E-11
4 40000 40000 24 64031.2 64031.2
5 0 4.4469E-12 25 0 1.6676E-11
6 -64031.2 -64031.2 26 -50000 -50000
7 -64031.2 -64031.2 27 -40000 -40000
8 80000 80000 28 -64031.2 -64031.2
9 0 2.2235E-11 29 -120000 -120000
10 -50000 -50000 30 0 0
11 -3E-10 0 31 50000 50000
12 50000 50000 32 -80000 -80000
13 -80000 -80000 33 0 2.2235E-11
14 0 1.4369E-37 34 0 1.6676E-11
15 -50000 -50000 35 -40000 -40000
16 64031.2 64031.2 36 0 -4.4469E-11
17 0 -2.2235E-11 37 0 -2.2235E-11
18 80000 80000 38 0 0
19 0 2.7793E-11 39 120000 120000
20 0 -5.5587E-12 40 50000 50000
Table 7: Verification of FEA code: results (kPa)
106
From the above data, the coincidence obtained for the results of the two
different methods provide adequate verification of the code.
3.5.2.2 Reliability calculation codes
After execution of Direct Simulation as well as the normal and the Adaptive
FORM/SORM procedure, focus is been concentrated on the members of the
reference structure that have a probability of failure different from 0 in every
procedure. This is due to the fact that, as it has been described so far, Direct
Simulations cannot describe low probabilities of failure, while FORM/SORM
methods do. Table 8, presents a sensitivity analysis of Direct Simulations with
different sampling numbers.
(sample size) 10000 100000 10000000

Elapsed time 20s 4min 4h
ELEMENT PF Beta PF Beta PF Beta
18 0,001500 2.97 0,001720 2.93 0,001790 2.91
21 0.00E+00 Inf 0 Inf 5.70E-03 4.39
22 0,021900 2.02 0,020580 2.04 0,020568 2.04
23 0.00E+00 Inf 0 Inf 5.70E-03 4.39
25 0,003700 2.68 0,004480 2.61 0,004489 2.61
26 0,025400 1.95 0,023980 1.98 0,024213 1.97
27 0,003700 2.68 0,004480 2.61 0,004489 2.61
28 0,000300 3.43 0,000270 3.46 0,000351 3.39
Table 8: Sensitivity Analysis of MCS Probability of failure results
(sample size) DS (10000000) Normal-RSM ARSM

Elapsed time 4h 18s 12m
ELEMENT PF Beta PF Beta PF Beta
18 0.001790 2.91 0.000860 3.13 0.000692 3.19
21 5.70E-03 4.39 8.03E-08 5.24 3.850E-08 5.37
22 0.020568 2.04 0.018430 2.09 0.016752 2.12
23 5.70E-03 4.39 8.03E-08 5.24 3.850E-08 5.37
25 0.004489 2.61 0.002897 2.76 0.002454 2.81
26 0.024213 1.97 0.022259 2.01 0.020378 2.04
27 0.004489 2.61 0.002897 2.76 0.002454 2.81
28 0.000351 3.39 0.000084 3.76 0.000060 3.84
Table 9: MCS vs ARSM Probability of failure results
107
In Table 9, comparative results between the direct Simulation and the
application of the normal and the adaptive Response Surface Method (ARSM)
are presented for the meaningful members as before. It can be concluded that
the difference between the deterministic and the simulation methods are
acceptable for the validation of the method selected for the numerical
application of this Thesis. Especially for the members with greater probability of
failure (lower reliability index), such as members 22 and 26, the results appear
to coincide significantly. Among the normal and the adaptive Response Surface
Methods, the latter seems to provide a correction to the results; however due to
the fact that the normal method is more conservative and the small difference in
the results is small in the expense of computational cost, the normal SRSM
seems to perform adequately and can be applied in the application section.
However, in different cases with non-linear performance, the adaptive
Stochastic Response Surface Method is expected to produce significantly more
accurate results than the normal Response Surface Method.
3.5.3 Validation of the FORM/SORM code
For a typical jacket structure, where multiple members exist and their reliability
should be assessed initially in a local level, a MATLAB code was conducted
based on the theory that has been developed earlier in this Chapter. MATLAB
has been selected as the most appropriate programming language due to the
convenience in handling partial derivatives, a fact that minimizes the required
computational time. This code, that was conducted and has been verified using
the PROBAN commercial software provided by DNV, will allow combination with
FEA codes and facilitate the procedure of multiple iterations required due to the
large number of element comprising the structure. A simple case of a complex
structure has been considered with four variables stochastically modelled.
Figure 28, illustrates the FEA model created in DNV SESAM Genie software
from which corresponding von Mises stresses were exported in order to
formulate the limit state function that was later solved with both the MATLAB
code and the commercial software. Table 10, presents the stochastic loads
considered.
108
Figure 28: FEA model in DNV Genie software
Load Description Distribution Characteristics Unit
Force on top of the structure ܽ = 2129, b = 7.1867

1 acting on x direction
Weibull (kN)
Force on top of the structure

2 acting on y direction
Normal ߤ = 2200, ߪ = 200 (kN)
3 Modulus of Elasticity Normal ߤ = 210, ߪ = 10 (GPa)
4 Material Yield Log-Normal ߤ = 275, ߪ = 25 (MPa)
Table 10: Stochastic loads on verification model
The MATLAB code that was conducted, applies the First and Second Order
Reliability Methods (FORM/SORM), with the ability to handle different types of
limit state functions (with some small alterations). Input to the code is a set of
stresses obtained from a series of simulations in the DNV Genie FEA software
109
that are automatically converted in utilization ratios that will be used to derive
the limit state functions, and a matrix of basic variables corresponding to the
variable combinations of the cases that were executed. A separate function for
multivariate data regression will return the regression coefficients for the
quadratic variables of the linearized limit state function. This function has been
verified with corresponding functions included in Microsoft Excel and was found
to provide sufficient results. The scaling that is included in the code, prevents
from ill conditioned results in cases where data with significant difference in
order of magnitude exists. Those intermediate results from the code will be
used both in the MATLAB reliability estimation routine and in the DNV PROBAN
software. DNV PROBAN software can solve limit state integrals using both
deterministic and probabilistic methods. From the later, the Axis Orthogonal
simulation, the Directional simulation and the Design Point simulation will be
applied in conjunction to FORM and SORM results and will be compared to the
results obtained from the MATLAB code. Comparison of the results obtained is
listed in Table 11. The test structure has been designed in a way that a wide
range of reliability indices can be obtained, in order to better verify the validity of
the comparison results.
From the results that have been obtained, it can be observed that for almost all
cases, the results of the two methods used coincide with excellent accuracy.
Therefore the Matlab code that is derived can be used with confidence for the
analytical part of this Thesis. Among First and Second order deterministic
methods, the results coincide significantly, which should be expected due to the
high linearity of the limit state function. Between the three simulation
techniques, the Axis Orthogonal simulation was found to be the most accurate
one for the input parameters examined (50 simulations, standard normal
density). Sufficient results, were also provided by the other two methods, with
the Design Point simulation (1000 simulations) providing better results than the
Directional simulation (100 simulations, 0.5 step, probability limit: 1.0e-080)
which presents some small variation for the cases of large values of beta.
110
MATLAB PROBAN
FROM SORM FROM SORM AXIS ORT. DIRECT DES.P.
1 Bm46 13.18 13.15 13.17 13.13 13.13 13.12 13.15
2 Bm49 N/A N/A N/A N/A N/A N/A N/A
3 Bm50 0.40 0.40 0.39 0.39 0.39 0.38 0.42
4 Bm51 9.22 9.22 9.24 9.24 9.24 9.26 9.25
5 Bm53 22.24 22.21 22.22 22.19 22.19 N/A 22.20
6 Bm54 26.75 26.83 26.72 26.81 26.72 N/A 26.71
7 Bm5 9.33 9.33 9.36 9.36 9.36 9.43 9.36
8 Bm6 8.22 8.22 8.17 8.17 8.17 8.15 8.18
9 Bm7 8.06 8.06 8.09 8.09 8.09 8.09 8.10
10 Bm8 9.51 9.51 9.46 9.46 9.46 9.45 9.44
11 Bm9 14.32 14.32 14.32 14.32 14.32 14.89 14.36
12 Bm10 13.62 13.55 13.63 13.56 13.63 13.93 13.65
13 Bm11 13.56 13.56 13.56 13.56 13.56 13.56 13.56
14 Bm12 14.41 14.41 14.40 14.40 14.40 14.67 14.40
15 Bm16 5.48 5.48 5.49 5.49 5.49 5.44 5.47
16 Bm13 6.30 6.30 6.32 6.32 6.33 6.31 6.31
17 Bm48 25.24 25.22 25.24 25.21 25.21 N/A 25.16
18 Bm14 5.58 5.58 5.55 5.55 5.55 5.55 5.55
19 Bm15 6.39 6.39 6.40 6.40 6.40 6.49 6.43
20 Bm17 13.41 13.41 13.41 13.41 13.41 13.45 13.40
21 Bm18 1.80 1.80 1.76 1.76 1.75 1.76 1.76
22 Bm19 16.77 16.77 16.76 16.76 16.76 16.75 16.78
23 Bm20 2.55 2.55 2.52 2.52 2.52 2.50 2.49
24 Bm21 1.84 1.84 1.81 1.81 1.81 1.83 1.78
25 Bm22 14.32 14.32 14.32 14.32 14.32 14.33 14.34
26 Bm23 15.66 15.66 15.65 15.65 15.65 15.67 15.68
27 Bm27 10.44 10.44 10.44 10.44 10.44 10.43 10.46
28 Bm24 2.50 2.50 2.46 2.46 2.46 2.47 2.46
29 Bm25 4.72 4.72 4.68 4.67 4.67 4.70 4.66
30 Bm26 11.08 11.08 11.08 11.08 11.08 11.18 11.08
31 Bm29 9.99 9.99 9.99 9.99 9.99 10.28 10.00
32 Bm28 4.57 4.57 4.53 4.53 4.53 4.62 4.54
33 Bm30 4.47 4.47 4.51 4.51 4.51 4.48 4.48
34 Bm31 11.51 11.51 11.51 11.51 11.51 11.44 11.51
35 Bm32 4.82 4.82 4.80 4.80 4.80 4.74 4.79
36 Bm33 6.86 6.86 6.88 6.88 6.88 6.89 6.89
37 Bm34 8.92 8.92 8.91 8.91 8.91 8.89 8.91
38 Bm35 8.95 8.95 8.90 8.90 8.90 9.00 8.90
39 Bm36 6.45 6.45 6.42 6.42 6.42 6.48 6.41
40 Bm37 6.65 6.65 6.65 6.65 6.65 6.70 6.66
41 Bm38 9.29 9.29 9.33 9.33 9.32 9.42 9.32
42 Bm39 8.60 8.60 8.61 8.61 8.61 8.58 8.60
43 Bm40 6.67 6.67 6.63 6.63 6.63 6.73 6.62
45 Bm42 12.63 12.61 12.58 12.51 12.49 12.58 12.51
47 Bm44 1.24 1.24 1.28 1.28 1.28 1.28 1.29
48 Bm45 10.57 10.57 10.57 10.57 10.57 10.56 10.60
49 Bm47 16.12 16.11 16.12 16.10 16.10 16.16 16.13
50 Bm52 15.88 15.87 15.88 15.87 15.87 15.91 15.87
Table 11: Results of comparative analysis of test case structure
111
The conclusions of this verification analysis is that for this type of linear limit
state function derived by linear elastic analysis, deterministic methods provide
sufficient results compared to the more complicated simulation methods. This
can reduce required computational effort significantly allowing more simulations
to be executed and more parameters to be examined. Between the MATLAB
code and the DNV PROBAN code, both provide results that coincide, allowing
the first to be used in the analysis that will be presented later in this Thesis. For
the simple test case, the computational time required for the regression analysis
and the computational of the reliability index with the FORM and SORM is less
than 30 sec, while the manual input of coefficients in the less user friendly
commercial software required more than 15 minutes; considering that this is a
reduced time that has been achieved through an intermediate code conducted
in Visual Basic.
3.5.4 Codes Included
In Appendix D of this Thesis the following codes are included:
 Response Surface Method code for FEA
 Adaptive Response Surface Method code for FEA
 Standalone code for FORM/SORM (limited to 4 variables)
 Multivariate Regression Analysis code with scaling
All the above codes have been modelled in MATLAB 2010a version, and use
the MATLAB Statistics and Symbolic Math Toolboxes.
3.6 Summary
In this Chapter, the numerical methods for structural reliability assessment have
been discussed. Deterministic Methods, including First and Second order
reliability methods have been presented according to the procedure that has
112
been followed for the formulation of codes that have been developed.
Simulation methods were also briefly discussed. The Stochastic Response
Surface Method (SRSM) has been extensively presented, and the procedures
for the normal and the adaptive SRSM were derived. Polynomial Regression
analysis was also included as it is required for the transformation of the
simulation output to a multivariate quadratic limit state surface. The Chapter
closes with an extensive verification of the MATLAB codes that were derived,
based on established tools/techniques, in order to obtain confidence for their
application in the later chapters.
113
4 STOCHASTIC MODELING OF ENVIRONMENTAL
LOADING AND LOAD CAPACITY OF OFFSHORE
STRUCTURES
4.1 Introduction
In this Chapter, the issue of appropriate stochastic modelling of environmental

loads as well as the load bearing capacity variables will be discussed.
Treatment of variables is a very important decision that should be based on well
informed data and experienced engineering judgement, as it should account for
the extreme conditions the structure may suffer from throughout its service life.
As far as the environmental loads are concerned wave is the most important
due to its volatile magnitude. Complementary to this, current wind and
operational loads form a combination of loads that should be considered for the
design of offshore structures. On the other hand, the load capacity of the
structure should consider variables such as the material properties of members
(yield strength, Young’s Modulus and Poisson’s Ratio, etc), geometrical
properties and geotechnical conditions. Deterioration of capacity is also a
significant issue to be considered due to its increasing effect throughout the
service life of the structure. Incorporation of the above in the design will be
discussed in the next sections.
4.2 Classification of Loads
In reliability analysis, appropriate consideration of loading is one of the most

critical issues. During its operation, an offshore structure is subjected to several
different types of loading depending on its service and the environmental
conditions in the location of deployment. Those loads may be categorized as:
dead loads, live loads, environmental loads, construction loads, removal and
reinstallation loads, and dynamic loads [2].
115
Dead loads are the weight of the entire structure and any permanent equipment
mounted on the platform which does not change with the mode of operation.
Dead loads should include the hydrostatic forces acting on the structure below
the waterline, including external pressure and buoyancy.
Live loads are the loads imposed on the structure during its operation. They
include the weight of any drilling and production equipment, the weight of living
quarters, heliport and other life support equipment, life saving equipment, diving
equipment and utilities equipment, the weight of consumable supplies and
liquids in storage tanks, the forces exerted on the structure from operations and
the forces exerted on the structure from deck crane usage. Live loads are
usually idealized as uniform distributed loads [182].
Environmental loads are loads imposed on the platform by the natural

phenomena including wind, current, wave, earthquake, snow, ice and earth
movement. Environmental loads also include the variation in hydrostatic
pressure and buoyancy on members caused by changes in the water level due
to waves and tides. Those loads are random in nature and can be accurately
quantified combining sufficient meteocean data with probabilistic properties.
Construction loads are loads resulting from fabrication, load out, transportation
and installation. Removal and reinstallation loads are those arising from
removal, on loading, transportation, upgrading and reinstallation particularly for
platforms which are to be relocated to new sites.
Dynamic loads are the loads imposed on the platform due to response to any
excitation of cyclic nature caused by waves, wind, earthquake or machinery; or
due to reaction to impact. Impact may be caused by a barge or boat
approaching the platform or by drilling operations.
From the above loads, the environmental loads are the most important in the
design of offshore structures due to their high level of randomness; therefore in
this chapter they will be presented analytically. Especially for offshore
116
structures, wave is found to dominate the total loading of the structure [183],
therefore different methods for wave modelling will be presented extensively.
4.3 Environmental Loading
Hydrodynamic forces play the most significant role in the design of offshore
structures that are subjected to both steady and time dependent forces due to
the action of winds, currents and waves. Parts of the structure that exceed the
sea water level suffer from both steady wind forces but also from significant
gusts which induce high unsteady local forces on structural components. In the
part of the structure that stands below the sea surface, steady forces in
conjunction with localized forces of vortex shedding induce substantial unsteady
forces on structural members. For both wind and current loads a steady or
unsteady flow will exert a correspondingly steady or unsteady (line) force
parallel to the incident flow direction; however, the localized interaction between
flows corresponding to a structural member will also cause flow irregularities
and will induce unsteady transverse forces perpendicular to the incident flow
direction.
Although several models are proposed for the modelling of wind and current
loads, waves are found to induce the largest force on most offshore structures.
For efficient structural design, it is essential to establish and appropriately select
methods for the transformation of environmental loads into the resultant steady
and time dependent forces acting on the structure. In this chapter the most
important of those methods will be presented. This procedure of moving from a
global level-environment to a structural level could categorize the following
three steps [184]:
 Definition of design environmental conditions, i.e. design parameters to

define waves, currents and winds
117
 Evaluation of water particle motions (kinematics) resulting from the
waves and currents, assuming that the structure does not affect the
gross kinematics
 Derivation of external forces on the individual members of the structure
resulting from the water particle motions and wind velocity
4.3.1 Design Environmental Conditions
As discussed earlier in the definition of the parameters that should be

considered in a reliability assessment, the reference return period should be set
in order to relate the environmental conditions which are exceeded on average
of ߋ years. For offshore structures, a reference return period of 100-years is
proposed by API in [2]. Different organizations, such as the UK Department of
Energy, propose a different reference return period of 50-years. Use of widely
applied statistical techniques can expand observations of a limited time period,
1-5 year time scale, to project the probability of failure of the 50 or 100 years
that should be considered by standards.
Since the contribution of different environmental loads to the response of the

structure cannot be defined, each of those loads is supposed to act at the same
time and towards the same direction to account for the worst case scenario.
Although this assumption is conservative, it can provide realistic results for the
design against structural failure; however it cannot consider fatigue damage on
the design. While in the North Sea sites, design values for the reference period
of 100-years are assumed to act simultaneously on the structure, for the sites of
the Gulf of Mexico, current is neglected completely or accounted in a smaller
scale.
The most important load parameters for the evaluation of the extreme
environmental loading are: significant wave height, mean zero-crossing period,
wind speed averaged over a suitable time interval, and current speed and
profile. Those parameters will be presented extensively in the following
sections.
118
4.3.2 Environmental Loads
4.3.2.1 Sea State Condition
The sea state conditions for a selected site can be effectively described by the
wave and current parameters. As it has already been mentioned, the design
wave is the most important element in the determination of the forces acting on
the structure. Design wave heights are expressed in terms of wave heights that
have a low probability of occurrence within the prescribed reference design
period (50 or 100 years) and achieve adequate safety considering the shorter
life period of offshore structures (usually 20 years).
4.3.2.1.1 Wave Load
Prediction of wave conditions is based on collections of observed data and

interpolation to the reference (return) period of investigation. Correlation
between significant wave height and wave period should be based on joint
probability distribution models, as it will be later described in greater detail.
Different directions of wave loads should be considered for the offshore
structure since in several cases different directions yield for different load
distributions on the structure; therefore the dominant direction should be
identified. In some cases, different combination of wave heights and
corresponding periods should be also investigated, in order to determine the
critical wave loading on a structure.
4.3.2.1.2 Current Load
Current refers to the motion of water due to reasons different than surface
waves. Depending on the deployment sites, this load can contribute significantly
to the total force applied in the submerged part of the structure. Common
categories of currents are: tidal currents, circulation currents and storm-
generated currents. The vector sum of these three currents is the total current,
and the speed and direction specified elevations constitute the current profile.
Appropriate consideration of the current properties is an important aspect in the
design of the structure as it might affect the location and orientation of
119
deployment as well as that of boats approaching the structure. Forces due to
currents are added in those due to the wave, therefore magnitude and relative
direction of the two should be specified in a way that realistically represents the
current’s profile from sea water level to seabed.
Determination of the force induced by current is based on the maximum design

velocity. Selection of this value should be based on actual measurements
and/or suitable data tables. For cases of shallow waters, where not sufficient
data are available, the vertical distribution of current velocity can be described
as [185]:
ଵ
‫଻ݖ‬ (4-1)
்ܷ௓ = ்ܷௌ ቀ ቁ
݀
Where: ்ܷௌ is the speed of current in the sea water surface, ‫ ݖ‬is the distance
above the seabed and ݀ is the total water depth.
For the case of deep waters that in general correspond to more slender
structures, wave and current interaction should be taken into account. In such
cases the drag loading is proportional to the square of the wave and current
velocity, therefore the response of the structure will be significantly affected by
any change in the current velocity. The simplest model for the consideration of
this interaction is the superposition of current and wave velocity. In [186], three
methods to account for the combination of wave and current are described:
 Current profile stretching to the instantaneous surface

 Mass continuity with profile stretching
 Cut off in through, uniform current addition in crest
The above methods are pictorially illustrated in Figure 29. Reviewing the above
methods, the first one seems the most appropriate since it represents the
convection of water particles in a wave considering its horizontal velocity with
an added component to account for the presence of the current [187].
120
Figure 29: Methods for combining current and wave [186]
In cases where the wave ‘rides’ upon current, a current in the wave direction
tends to stretch the wave length, while an opposing current shortens it [188].
Therefore, the frequency of the wave varies along the wave. This phenomenon
is called “Doppler Effect” on wave frequency; a detailed mathematical
explanation of this can be found in [189].
121
Further theoretical models on the description of wave-current interaction may be
found [190], where two wave models are developed (the linear and bilinear
shear current models) for the representation of symmetric finite amplitude water
waves of arbitrary order, propagating on currents with velocity profiles which are
described by two or three straight lines.
4.3.2.2 Wind Load
Wind load on the structure refers to the forces induced in the part that exerts
above the sea water level and contains equipment, decks and irregularities that
cause drag from the air particles motion. Wind speed is classified as either
sustained or gust wind speed with direction varying in time and space. Data for
values of wind speed should refer to specific elevation and duration of
measurements.
The sustained wind speed is defined as the average wind speed over a time
interval of 1 minute measured at an elevation of 10 m above still water level
(SWL). The wind velocity varies significantly with height due to the boundary
layer induced by viscosity. The speed at a height ‫ ݖ‬above sea water level could
be estimated by the following semi-empirical relationship [191]:
‫ ݖ‬଴,ଵଵଷ
ܷ௪ (1݉ , ‫ܷ = )ݖ‬௪ (1݉ , 10) ∙ ቀ ቁ (4-2)
10
Where: ܷ௪ (1 ݉ , 10) is the 1 minute mean sustained wind speed at 10 m above

sea water level. A similar expression, proposed in [2], connects the 1 hour
mean speed to different elevations as:
‫ ݖ‬଴,ଵଶହ
ܷ(1ℎ‫ݎ‬, ‫ܷ = )ݖ‬௪ (ℎ‫ݎ‬, ‫ݖ‬ோ ) ∙ ൬ ൰ (4-3)
‫ݖ‬ோ
The gust wind speed is defined as the average wind speed over a time interval
of 3 seconds measured at an elevation of 10 meters above SWL. Following the
same description adjustments for elevation are given by the following equation
[191]:
122
‫ ݖ‬଴,ଵ଴଴
ܷீ (‫)ݖ‬ = ܷீ (10) ∙ቀ ቁ (4-4)
10
Where: ܷீ (10) is the gust wind speed at 10m above SWL.
The gust factor is defined as:
‫ݐ(ܩ‬, ‫ = )ݖ‬1 + ݃(‫)ݖ(ܫ ∙ )ݐ‬ (4-5)
Where: ‫ )ݖ(ܫ‬is the turbulence intensity and ‫ݐ‬is the gust duration in seconds and
can be estimated as follows:
ି଴.ଵଶହ
0.15 ∙ ൫‫ݖ‬ൗ‫ݖ‬௦൯ , ‫ݖ ≤ ݖ‬௦
‫ = )ݖ(ܫ‬ቐ (4-6)
‫ݖ‬ ି଴.ଶ଻ହ
0.15 ∙ ൫ ൗ‫ݖ‬௦൯ , ‫ݖ > ݖ‬௦
Where: ‫ݖ‬௦ is measured in elevation of 20 m. The factor ݃(‫ )ݐ‬is calculated as:
݃(‫ = )ݐ‬3.0 + ݈݊[(3/‫)ݐ‬଴.଺] (4-7)
The wind force acting on a structure is the sum of the wind forces acting on
individual members. Loads from wind and waves are assumed to act
simultaneously in the structure, since in most cases in the presence of high
winds, severe wave phenomena are observed. The equation for the drag force
of an object within a flow is applied in the case of a member exposed to a wind
of uniform velocity ܷ:
1
‫ܥ = ܨ‬ௌߩܷ ଶ‫ܣ‬ (4-8)
2
Where: ߩ is the air density, ‫ ܣ‬is the characteristic area of the body facing the
wind and depends on the shape of the exposed member and ‫ܥ‬ௌ is a shape
coefficient. The values of this parameter will be extensively discussed in a later
section, however typical values are provided by design standards.
123
4.3.3 Wave Modelling
4.3.3.1 Fundamentals of Wave Modelling
In [192], the most probable maximum wave height is approached by a Reyleigh

distribution, based on the significant wave height, the mean zero-crossing
period and the reference storm duration. This wave can be efficiently
approached by a deterministic two dimensional wave with an associated period
in order to be considered in the evaluation of water particle kinematics. A
current profile is combined with the wave consideration to produce the total
water particle motion. Several wave theories have been proposed for the
combination of waves and currents.
Figure 30: Ranges of appropriate wave theories [1]
Among different environments, different wave theories are applicable, based on

different environmental parameters such as, water depth, wave height and
wave period. Figure 30 [1] and Figure 31 [193], present the different areas of
applicability of the different wave theories based on the above parameters.
124
The coordinate system employed in wave theories is a three dimensional
(‫ݔ‬, ‫ݕ‬, ‫ )ݖ‬with ‫ ݔ‬following the direction of wave propagation, and ‫ ݖ‬is measured in
the vertical depth direction. Those two dimensions are sufficient to describe a
normal wave as it is described in Figure 32. Waves are assumed to be periodic,
with a period ܶ, and uniform with a height ‫ ܪ‬. Further, fluid is assumed to be
incompressible, the flow to be irrotational, and that the free surface
uncontaminated [194].
For a periodic wave, the speed of a given crest or trough is called celerity. This
parameter links the length and period of the wave as:
‫ܮ‬
ܿ= (4-9)
ܶ
Alternative applicable definitions are that of the angular frequency, ߱ = 2ߨ/ߒ

and the wave number, ݇ = 2ߨ/‫ܮ‬.
Figure 31: Ranges of appropriate wave theories [193]
125
Figure 32: Coordinate system of wave propagation [194]
4.3.3.2 Formulation of Wave Theories
The existing different wave theories, attempt to determine the velocity potential
ߔ or equivalently, the stream function, ߖ pertaining to the fluid region. This
should satisfy the Laplace equation:
߲ଶߔ ߲ଶߔ
+ =0 (4-10)
߲‫ݔ‬ଶ ߲‫ݖ‬ଶ
Considering ࣁ(‫ݔ‬, ‫ )ݐ‬to be the free surface elevation measured above the still
water level ‫ = ݖ‬0, the following boundary conditions should be satisfied:
߲ߔ
= 0 ܽ‫ = ݖ ݐ‬−݀
߲‫ݖ‬
߲ߟ ߲ߔ ߲ߟ ߲ߔ
+ ∙ − = 0 ܽ‫ߟ = ݖݐ‬ (4-11)
߲‫ݖ߲ ݔ߲ ݔ߲ ݐ‬
߲ߔ 1 ߲ߔ ଶ ߲ߔ ଶ
+ ∙ ቈ൬ ൰ + ൬ ൰ ቉+ ݃ ∙ ߟ = 0 ܽ‫ߟ = ݖ ݐ‬
߲‫ ݐ‬2 ߲‫ݔ‬ ߲‫ݖ‬
From the above boundary conditions, the first one corresponds to the fact that
at the seabed a zero vertical component on the fluid particle velocity is
assumed, while the other two represent the kinematic and dynamic free surface
boundary conditions respectively. Solution of this complicated problem is a
difficult task because the free surface boundary conditions are nonlinear and
must be satisfied at the free surface which is constantly changing. Analytical
126
methods for the wave modelling, corresponding to available wave theories, can
be found to Appendix A of this Thesis.
4.3.3.3 Comparison and Application of Wave Theories
Selection of the most appropriate wave theory for each particular application is
a significant decision as it will be shown later in this Thesis. Based on values of
‫ ܪ‬, ܶ and ݀, different wave theories are expected to describe differently the
water particle motion. As mentioned in the introduction of this Chapter, empirical
charts will prescribe the areas of applicability of each one of the methods,
Figure 30, and Figure 31.
Figure 33: Applicability regions of wave theories based on the relative

error on the fit of the two nonlinear free surface boundary conditions [195]
In [195], a comprehensive comparison of several available wave theories is

presented. The wave theories examined include linear wave theory, Stokes
third and fifth order theories, cnoidal, solitary and the stream function theories;
the ranges of each theory is illustrated in Figure 33. Considering that the
Laplace and bottom boundary condition are satisfied in all the theories included,
the relative error of fit to the two nonlinear free surface boundary conditions was
used in order to benchmark the performance of the methods. In the cumulative
graphs that are available, the basic variables are combined in terms of ߅ /݃ܶଶ
127
and ݀/݃ܶଶ. It can be observed that all those charts relatively agree on the
regions of applicability they indicate.
Once the water particles motion has been derived by applying the appropriate
wave theory, hydrodynamic forces can then be calculated. The non-linear
relation between water particle kinematics and water surface displacement, the
turbulent flow process about a structural member, the natural variability of wave
forces and the possibility of resonance between waves and structures are some
of the difficulties imposed in the calculation of hydrodynamic forces.
4.3.3.4 Hydrodynamic Wave Forces
Based on the type and size of structural members on an offshore structure,

different formulations for wave forces may be applicable. Mainly three different
ways are applied for this calculation and a classification of those methods can
be found, and are presented in [196]:
 Morison Equation
 Froude-Krylov theory
 Diffraction theory
Morison equation theory, as was presented in [197], describes the forces acting
on a vertical pile subjected in a viscous, unsteady flow; this method is applied in
cases when the structure is small compared to the water wave length and when
the drag force is significant. In different cases, with small drag forces or inertia
dominant, but still referring to relatively small structures, the Froud-Krylov theory
is applicable. Finally, the diffraction theory is applied in cases when the size of
the structure can be compared to that of the waves.
Morison proposed that the forces exerted on a vertical cylindrical pile which
extends from the bottom to the free surface is composed of two components;
that of drag and inertia. Combination of the two provides the total force as
follows:
128
߲࢛
݂ = ‫ܥ‬ெ ‫ܣ‬ூ + ‫ܥ‬஽ ‫ܣ‬஽ |‫ݑ|ݑ‬ (4-12)
߲‫ݐ‬
Where: ‫ܥ‬ெ and ‫ܥ‬஽ are the inertia and drag coefficients, ‫ ݑ‬is the velocity of the
డ࢛
wave particles, is the local water particle acceleration at the centre line of the
డ௧
cylinder and ‫ܣ‬ூ, ‫ܣ‬஽ are the characteristic areas for each members calculated
as:
గ ଵ
‫ܣ‬ூ = ߩ ସ ‫ ܦ‬ଶ and ‫ܣ‬஽ = ଶ ߩ‫ܦ‬ (4-13)
Where: ߩ is the mass density of water and ‫ ܦ‬is the member’s diameter. In the
above equation, the inertia term is proportional to the acceleration of the water
particles while the drag term is proportional to the square of the velocity. The
absolute term is applied to ensure that the drag force will coincide with the
direction of the flow. The wave velocity and acceleration of water particles
decay with depth; therefore the force distribution is expected to follow the same
pattern [198]. The previous expression provides the force for the unit length of a
vertical cylinder; the total force will be derived by integration as follows:
ௗ ௗ
߲࢛
‫ = ܨ‬න ݂݀‫ = ݏ‬න ൤‫ܥ‬ெ ‫ܣ‬ூ + ‫ܥ‬஽ ‫ܣ‬஽ |‫ݑ|ݑ‬൨݀‫ݏ‬ (4-14)
߲‫ݐ‬
଴ ଴
In the above formulation of the total wave force, hydrodynamic coefficients ‫ܥ‬ெ ,
‫ ܦܥ‬should be appropriately selected. Based on experimental research, those
coefficients are found to be correlated to the Reynolds and Kuelegan-Carpenter
(KC) numbers [194]; ‫ܥ‬ெ increases with Re while K is related to the diameter D of
the member and the amplitude of the wave A. Figure 34 and Figure 35, adopted
from [199], show the correlation between coefficients and characteristic values.
The analysis shows that at higher values of Re (Re > 105), ‫ ܦܥ‬approaches 0.65
and CM approaches 1.8. The coefficients ‫ܥ‬ெ and ‫ ܦܥ‬were also found to depend
on both Re and a frequency parameter (ߚ = ܴ݁/‫ )ܥܭ‬when the coefficients
were plotted against KC. Further extensive results based on field tests may be
found at [200] showing extensive scatter of results due to the complexity of
129
wake structures in oscillatory flows which cannot be fully covered by Morison’s
equation.
Figure 34: CD-KC number for different values of Re and β=Re/KC [199]
Selection of hydrodynamic coefficients in practice is driven by the provisions of

different standards and societies such as the American Petroleum Industry
(API), Det Norske Veritas (DNV), International Organization for Standardization
(ISO). A typical example is the classification of coefficients based on different
roughness of the surface of the member; for a smooth cylinder, API suggests
‫ = ܦܥ‬0.65 and ‫ܥ‬ெ = 1.6 while for rough cylinders ‫ = ܦܥ‬1.05 and ‫ܥ‬ெ = 1.2.
Figure 35: CM-KC number for different values of Re and β=Re/KC [199]
130
A comparative study for the different proposed values for the hydrodynamic
coefficients can be found in [201]. Design practice for global deterministic
analysis combines a low drag factor with conservative estimates of the wave
kinematics and current. API combines the wave kinematics factor with a low
current value, current blockage and shielding.
Further literature, available in [184], considers that for tubular members of fixed
offshore structures under extreme storm conditions, the Reynolds number lies
in the post-critical regime (ܴ݁ > 2 ‫ ݔ‬10଺) and ‫ ܥܭ‬exceeds 10. Large scale
laboratory experiments are presented in [202], for ܴ݁ numbers exceeding this
value, ‫ ܥܭ‬numbers up to 90 and including rough and smooth cylinders
undergoing random oscillations in an attempt to realistically approximate tubular
members of fixed offshore structures under extreme storm conditions. Table 12,
summarizes the values of the hydrodynamic coefficients for smooth and rough
cylinders as a function of the ‫ ܥܭ‬number. For the cases of rough cylinders, ܴ݁
was found to have a low dependency on the value of the hydrodynamic
coefficients.
KC Smooth Force Coefficients Rough Force Coefficients

Number Cd Cm Cd Cm
0.0 0.7 2.0 1.5 2.0
6.0 0.7 2.0 1.5 2.0

30.0 0.7 1.7 1.3 1.5
60.0 0.7 1.6 1.2 1.3
90.0 0.7 1.6 1.2 1.3
high 0.6 1.6 1.1 1.3
Table 12: Values of hydrodynamic coefficient for circular cylinders [202]
Extensive research on oscillatory flows around cylinders have studied

separation from the surface of the cylinder and formation of wake behind it,
correlating vortex shedding behaviour at different values of the ‫ ܥܭ‬number
[203]. An empirical formulation of the vortices formed around tubular sections is
given in [191], correlating the Strouhal number valid for a range of Reynolds
number between 60 − 2 × 10ହ.
131
The majority of fixed offshore structures are drag dominated, considering
ultimate limit state, and therefore the inertia coefficient has a lower influence.
However, this becomes important for structures with large buoyancy legs and
for other purposes such as fatigue design.
4.3.4 Joint Probability of Environmental Parameters
4.3.4.1 Wave Current Interaction
The first step in order to obtain the environmental design value is to consider
the oceanographic data of the site where the structure is due to be deployed.
The design wave method is a widely used method for the extrapolation of the
data to obtain the extreme value for the environmental parameters; this has
been an active research area, especially for the characterization of the extreme
wave heights. Published research in [204], [205] and [206], have used real
measurements to derive values from the meteocean data. The same method is
also applicable for the estimation of extreme values for wind. The total loading
experienced by the structure is based upon a set of design values of the
individual environmental parameters such as the summation of the forces of
wind, wave and current for the 50 or 100 year return period.
Drawback of the method is that it considers all the extreme environmental loads
to occur at the same time, causing the worst case scenario to the structure. For
example, occurrence of extreme wave current is not directly correlated to the
instance of the extreme wind speed or extreme current speed. This implies that
the initial assumption of simultaneous occurrence of all three extreme events,
leads to over conservative consideration of forces.
Extensive studies have been carried out in order to systematically model the
effects of joint probability of environmental loading on the structural response
and are presented in [207] and [208]. Unlike the design wave method, a joint
probability approach, assumes non correlated occurrence of the extreme
events. The concept, widely used in the reliability analysis to quantify the design
condition, has received increasing attention as the measured data set of the
132
metocean conditions becomes available and confidence in the hindcast model
is constantly growing.
Over the years several techniques of joint probability have been developed. The
joint probability of winds, waves and currents are mainly studied since these
loads dominate the total loads applied on the structure. Combination of
parameters in [208] have verified the model’s ability to provide acceptably
accurate results in cases of wave dominant structures, and describe
complicated loading patterns for a wide range of environmental conditions.
Application of the method has led to the reduction of loads which are exerted on
the structure compared to the individual ܰ -year wind, wave and current; in
specific cases allowing reduction of even 20% [207]. Parallel studies for the
Norwegian environments have concluded that wave and current loads during a
storm rarely occur simultaneously at the same direction and therefore a more
systematic consideration might substantially reduce the calculated
hydrodynamic loads on drag dominated offshore structures [209]. Further
studies [210] have verified this tendency for overestimation.
Another important issue that will be discussed in depth in the following sections
is the incorporation of the design period in the extreme design conditions.
4.3.4.2 Wave Height- Period Joint Distribution
A systematic recording of environmental conditions for several locations in the

North Sea is available in [211]. Most of the oceanographic data available are
measured for typical time intervals of 10-20 minutes [212]. In cases of poor
data, hind casting is the common method; different approximation methods are
presented in [212]. In [213] and [214] where storm data have been studied,
problems regarding limitations in computation of using hind casting and use of
poor data are addressed and a threshold is proposed in order to obtain the joint
probability of wave height and its period for the extreme waves.
For a probabilistic design of an offshore structure a joint distribution of

significant wave height and wave period is required to describe the long term
133
wave climate. Those two variables are used as primary parameters to specify a
set of stationary wave spectra, where the structural response is easily
computed. [215], presents a review between the joint distributions of wave
height and wave period, summarizing the work of several researchers. Among
different available joint distribution models which may be applied, the
applicability of each model should be checked for each specific location.
Evaluation of different bi-variate distribution functions should be done
empirically in order to ensure that the selected distribution provides a sufficient
basis for the prediction of sea states.
Collected data for the significant wave height ߅ and the spectral period ܶ௉ are
classified in scatter diagrams which provide a convenient and simple to interpret
way of summarizing pairs of random variables. The spectral peak period is
chosen due to the fact that it is less correlated to the significant wave height
[216]. The significant wave height and peak spectral period can adequately
describe the sea state, and are important parameters in characterization of a
wave spectrum [217].
Once the data are binned appropriately, the probability of occurrence of each is
calculated. A three dimensional surface can be then constructed and smooth
contour lines of same probability can be distinguished, giving an estimate of the
bivariate probability distribution. Similar scatter diagrams and contour plots may
be constructed from a fitted distribution and comparison between the two will
show which bivariate probabilistic model has the best fit. Among several
available bivariate probability density functions of the joint distribution available,
three will be discussed here and will be applied in a later chapter.
4.3.4.2.1 Bivariate Log Normal Distribution
This model, proposed in [218] forms the joint probability density function based
on five basic parameters (μx,σx,μy,σy and Ρxy), where x=ln Χ, y=ln Υ and the μx, μy
σx, σy Ρxy. Therefore the density function is formulated as:
134
1
݂(‫ݔ‬, ‫= )ݕ‬
ଶ
2࣊ߪ௫ ∙ ߪ௬ ∙ ܻܺ ∙ ඥ 1 − ߩ௫௬
−1 ‫ ݔ‬− ߤ௫ ଶ 2ߩ௫௬ ∙ (‫ ݔ‬− ߤ௫)൫‫ ݕ‬− ߤ௬ ൯

∙ ݁‫݌ݔ‬൝ ൥൬ ൰ − (4-15)
ଶ ൯
2൫1 − ߩ௫௬ ߪ௫ ߪ௫ ∙ ߪ௬
ଶ
‫ ݕ‬− ߤ௬
+ቆ ቇ ቉ൡ
ߪ௬
As would be considered for a bivariate normal probability density function, it is

implied here that if the characteristic of wave height and wave period follow the
log normal probability law, then the combined statistic properties should follow
the bivariate log normal probability law.
4.3.4.2.2 Bivariate Weibull Distribution
Weibull distribution is widely used for the description and prediction of the
extreme value of significant wave height. The form of the bivariate Weibull
distribution is formulated as follows considering ߙ, ߚ, ߛ and ߣ are the parameters
of the two distributions:
ߛ௬ംషభ ‫ ݕ‬ఊ ߚ௫ഁషభ ‫ ݔ‬ఉ

݂௫௬ (‫ݔ‬, ‫ߙ ;ݕ‬, ߚ, ߛ, ߣ) = ∙ ݁‫݌ݔ‬൜− ൬ ൰ ൠ∙ ∙ ݁‫݌ݔ‬ቊ− ቀ ቁቋ (4-16)
ߟఊ ߟ ߙఉ ߙ
This biνariate distribution consists of a marginal Weibull distribution for the

spectral peak period with parameters ߛ and ߟ and the conditional Weibull
distribution for significant wave height ߙ, ߚ; these parameters may be estimated
by the methods of moments.
4.3.4.2.3 Marginal Weibull and Conditional Log-Norma Distribution.
Combination of the above distributions derives the marginal Weibull and

conditional log-normal distribution. This is one of the joint distributions which are
used to describe the significant wave height and the spectral peak period, in
cases where the 2-parameter Weibull distribution does not provide adequate
results within the full range of wave heights. In [216] a combination between the
2-parameter Weibull distribution for large heights with a log-normal distribution
for low heights is presented. An abbreviation of the method names it as the
'Lonowe' distribution (from LOgNOrmal and WEibull) and has shown to provide
good fit to many sets of wave data. The transition point between the two
135
distributions is a decision that has some effect on the calculated probability of
extreme wave heights. The joint probability density function is conveniently
written:
ܶ௣
݂ு௦்௣ ൫‫ܪ‬௦, ܶ௣ ൯= ்݂೛ ⁄ுೞ ൬ ൰݂ு௦(‫)ݏܪ‬ (4-17)
‫ܪ‬௦
Where ݂ு௦(‫ )ݏܪ‬and ்݂೛ ⁄ுೞ൫ܶ௣⁄‫ܪ‬௦൯ are the marginal probability density function
and the conditional probability density function for the significant wave height
and the spectral peak period for a given wave height respectively. The
probability density functions are fitted to the observations separately. Values of
those probabilities are calculated as follows:
1 (ln ‫ܪ‬௦ − ߠ)ଶ

݂ு௦(‫= )ݏܪ‬ ∙ ݁‫݌ݔ‬ቊ− ቋ ‫ߟ ≤ ݏܪ‬௛
ඥ2ߨ݇‫ܪ‬௦ 2݇ଶ
(4-18)
ఉିଵ
ℎ௦ ‫ܪ‬௦
݂ு௦(‫∙ ߚ = )ݏܪ‬ ∙ ݁‫݌ݔ‬൜− ൠ ‫ߟ ≤ ݏܪ‬௛
ߩఉ ߩ
Where: ߠ and ݇ଶ are the mean and variance of ݈݊(‫ܪ‬௦) and where additionally,
continuity is required for ݂ு௦(‫ )ݏܪ‬at ‫ߟ = ݏܪ‬௛ , and ߚ and ߩ are the shape and
scale parameter for the Lonowe model. Altematively, the 3-parameter Weibull
distribution may be adopted.
The conditional distribution of ܶ௣ given ‫ܪ‬௦ is herein modeled by the lognormal

distribution as follows:
ଶ
1 ൫ln ܶ௣ − ߤ൯
݂ு௦்௣ ൫ܶ௣⁄‫ܪ‬௦൯= ∙ ݁‫݌ݔ‬൝− ൡ (4-19)
ඥ 2ߨߪߒ௣ 2ߪଶ
Where ߤ = ‫ܧ‬ൣln ܶ௣ ൧and ߪଶ = ܸܽ‫ݎ‬ൣln ܶ௣ ൧
4.3.5 Estimation of Extreme and Design Values
Estimation of extreme values of environmental variables is defined by the

marginal probabilities of exceedance which are determined by statistical
136
processing of an extreme value probability distribution of available data. Based
on appropriate fitting, extrapolation of the data to small exceedance probabilities
can be realized. Long term values of environmental variables can be obtained
based on empirical procedures. In [219], the procedure of estimation of extreme
values through statistical extrapolation distinguishes the following steps:
 Obtain a dataset through hind cast or measured data

 Statistical model fit
 Derive the required return value
In the following Sections, the above topics will be briefly discussed.
4.3.5.1 Data sampling
The most common methods to produce sub-sets of data for the derivation of
extreme values are:
 Initial Distribution (ID) method

 Annual Maxima (ΑΜ) method
 Peaks Over Threshold (ΡΟΤ) method.
In the Initial Distribution (ID) method, as is presented in [212], all available data
are considered for extrapolation, including data that might include multiple
values generated by the same storm event. The estimation will be executed
using an appropriate statistical model to a distribution of data that does not
necessarily describe properly the distribution of extremes. This practice might
impose an error in the derived extreme design values.
In the case of the Annual Maxima (ΑΜ) method, extremes are derived as the
single most severe observation within a year period, providing a series of
uncorrelated observations. Some issues raised from the definition of a year in
temperate climates, such as the North Sea, and for significant rare events that
do not occur in a location every year, in tropical climates such as the Gulf of
Mexico [220].
137
In the Peaks Over Threshold (ΡΟΤ) method, the extremes are assumed to be
generated differently than the other methods. Following this approach, an
extreme is considered each time a value exceeds a given limit. The major issue
in this method is an appropriate selection of this limit. Adequate time intervals
can allow for several measurements to be taken, providing independent and
uncorrelated observations. For the most important environmental variable which
is the wave height in regions such as the Gulf of Mexico, this method is
particularly applicable [205]. Further this method can distinguish severity of
environmental phenomena.
4.3.5.2 Statistical model fit
Following the selection of the dataset, extreme values will be derived from
statistical fitting based on an appropriate theoretical model. Selection of the
most suitable fit should be done based on the best fit of available distributions
[221]. Through the available probability distributions for the calculation of the
extreme values the most widely used are the following:
 Log-Normal distribution
 Fisher – Tippett Type I (FT-I), Type II (FT-II), Type III (FT-III)
 Weibull 2 and 3 Parameter
Log-Normal Distribution is widely used in reliability analysis due to the fact that
it can take only positive values and is the first to be used for extreme value
calculation. Together with the Weibull distribution, those are the two
representative distributions to describe the long-term behavior of the significant
wave height. Apart from the selection of the selection of distribution functions,
the following methods are used for the derivation of the parameters of the
extreme distributions:
 Method of moments
 Least squares method
 Maximum likelihood method
138
The method of moments is simple and directly applicable and is based on the
equation of the first two or three moments of the distribution to those of the data
and therefore establishes relationships between estimated parameters and the
sample mean, variance and skewness.
The least squares method is applicable to the Log-Normal, FT-I and FT-II
distributions. This method initially transforms variables in a way that will
linearize the cumulative density in a graph with a new set of variables; fitting a
line to the transformed set will determine the parameters of the distribution.
Although this method is widely used, it has the drawback of a bias [221].
The method of maximum likelihood may overcome this problem providing

estimated parameters with small variance, based on the concept of constructing
a likelihood function which depends on the probability density function of the
observations, which are expressed in terms of the unknown parameters of the
distribution; maximization of the logarithm of this function determines the value
of the parameters. According to [222], advantage of this method is in its good
asymptotic properties, since as the number of observations increases, the
parameters estimators converge to the true values [223].
Further to the above, commonly used methods for the determination of the
distribution parameters, combinations of the least squares method, least
absolute deviation method and optimization method may be used. This method,
proposed in [224], evaluates distribution parameters by automated search of
optimal curve type based on the least absolute deviation between the empirical
frequency point and distribution curve on ordinate.
The selection of an appropriate distribution for the estimation of extreme values

is based on the level that data are represented by a line when variation of this
line can be estimated by fitness tests. The advantage of using such methods is
based on empirical distribution function (EDF). These EDF tests are distribution
free and therefore could be useful for distinguishing between the various
possible distributions [221]. Statistic tests that are used for this purpose are the
Kolmogorov-Smirnov, the Crammer-Von Mises and the Anderson-Darling
139
statistic tests. Further reference to those tests is beyond the scope of this
Thesis. An important note is the consistent selection of a single plotting position
throughout the testing procedure [221].
4.3.5.3 Design Value
Once a particular probability distribution has been selected following the

procedure that has been described above, the design value may be determined
based on a return period, ܶோ or encounter probability, ‫ܧ‬௣ . The return period, ܶோ ,
is defined as the average time interval between successive events being
exceeded, and is directly related to the probability of exceedence as:
‫ݎ‬ ‫ݎ‬
ܶோ = = (4-20)
ܳ(ܺ) 1 − ܲ(ܺ)
Where: r is the recorded interval associated with each data point and Q(X) is the
probability of exceedence. The probability ‫ܧ‬௣ is defined as the probability that
the successive event is exceeded during a prescribed period ‫ܮ‬௙, which will
describe the design lifetime of a structure, and may be preferable instead of
using ܶோ to correspond to the prescribed values of ‫ܧ‬௣ and ‫ܮ‬௙ and is defined with
the following formula [225]:
‫ܧ‬௣ = 1 − ݁‫݌ݔ‬൫− ‫ܮ‬௙⁄ܶோ ൯ (4-21)
4.3.6 Fluid Loading on Offshore Structures
In reliability analysis, loading is the most important concern to the response of

the structure. During the operational life of a jacket structure, it is subjected to
different loads depending on the type of service and the environmental
conditions in the location of its deployment. The previous sections have covered
the methods for the derivation of the environmental parameters. Some
difficulties though arise from application of those parameters to structural
analysis, setting the determination of analysis as an important step in the study
of an offshore structure. Good engineering judgment combined with sound
140
scientific knowledge is essential for the validity of the results obtained in order
to minimize errors.
The methods existing for the analysis of fluids can distinguish three main
categories: deterministic, probabilistic and spectral. The deterministic approach
[226], which is the simplest and most widely applied so far, analyses the fluid
load by taking into account nonlinearities, but having the drawback of the
restriction to incorporate uncertainties related to the sea state conditions [196].
Studies have shown that a deterministic model may produce unrealistic force
distributions among the height of the structures, underestimating actual loads by
10% compared to a more advanced probabilistic model [227].
The probabilistic model is an extensively studied technique in the analysis of

fluid loading, incorporating randomness of sea and wave forces with their
statistical properties and enabling elaboration of the period’s statistical
properties prior to evaluation of the 50 or 100-year design wave. In the
probabilistic method, a distribution is usually formed to describe the significant
wave height, zero mean crossing period and the extreme conditions. In order to
transfer from a deterministic to a probabilistic mode [228], some of the normal
wave theories should be employed by selection of an appropriate wave height
(eg. ‫ ܪ‬ଵൗ , ‫ܪ‬ଵ଴, ‫ܪ‬௠ ௔௫). Having selected the appropriate design wave, a
ଷ
corresponding period can be allocated, as was described in the relevant

section, based in studies such as [229], [230] and [231].
141
Figure 36: Frequency domain analysis [232]
Finally, the spectral method is the more comprehensive approach to the

representation of the fluid loading, as it fully describes the loads and response
of the structure statistically. It incorporates directly the variability of sea surface
associated kinematics. Application of the method is based on linear super
position or linearization of non-linear processes. The method is often used for
the dynamic and fatigue response assessment of deepwater structures in
severe environmental conditions [233]. Figure 36, adopted from [232], illustrates
142
the sequence of spectral analysis; time discritization is applied to account for
variation in time, and transfer functions are established to calculate the wave
force response from the sea surface spectrum. The sea surface spectrum
ܵௐ (߱) is the input to the wave force transfer function while the ܵி (߱) is the
output. The wave force spectrum will be used in the structural analysis in order
to obtain displacements by multiplying the wave force spectrum with the transfer
function of the structure.
4.3.7 Response of Structure under Environmental Loading
In [194] and [232], the issue of structural loading is studied. Starting from
definition of environmental conditions through elaboration of meteorological
data, the wave conditions can be established. The appropriate wave theory will
analyze the kinematics properties of the fluid, deriving the hydrodynamic forces
and therefore calculating the structural response. It has been discussed in
Chapter Two that static or dynamic approach can be followed for the structural
analysis assessment and criteria for the selection of each approach, based on
the natural frequency, have been established. In [196], the dynamic problem
can be categorized as follows:
 Lowest natural frequency, which refers to the case where the natural
frequency of the structure is lower than the resonance frequency. In
such cases static analysis is sufficient; in different cases where the
dynamic part of the load is significant then dynamic analysis might be
required.
 One, or more, of the natural frequencies of the structure is in the range
of the frequency of resonance. In such cases, dynamic analysis is
definitely required.
A useful graph, presented in Figure 37, illustrates the areas of applicability of

static and dynamic analysis. In cases where the response is controlled by
stiffness, the analysis might follow the quasi-static method, while for the case
where the response is controlled by mass and damping, the dynamic analysis
should be employed. As far as dynamic analysis is concerned, it can be
143
classified in either frequency or time domain. The first method is more time and
computationally efficient; however if the amount of data is restricted, the time
domain analysis might be used, since it allows consideration of the drag term in
the force equation without linearization, eliminating errors due to non linearity in
sea states.
Figure 37: Static and Dynamic analysis response curve
4.4 Capacity of Offshore Structures
4.4.1 Resistance Model
Derivation of the resistance part of a limit state function defines the performance
of a structure under specific load conditions, as a function of a number of
different basic variables. For a realistic model of resistance, all the relevant
variables should be identified and included, while the variables that have a
lower significance on the model may be omitted.
Theoretical and empirical methods are available to derive a resistance model;

however for a comprehensive selection, validation based on comparisons of the
above techniques is essential. Comparison between the two sets of obtained
data can be done using appropriate plots that will allow correlation between
144
variables. If the calculation method accounts adequately for the respective
variables, correlation between theoretical and experimental values can be
established. The types of plots that can be drawn are:
 Plot of the observed resistance values against the calculated values

using the measured properties
 Plot of the observed resistance against each of the basic variables
Significant deviations in the plot of the second type should lead to

reconsideration of the theoretical model. Insufficient consideration of all of the
important variables and incompleteness of the resistance equation, without
considering every design parameter, may impose a significant source of
uncertainty on the reliability estimation.
4.4.2 Material Data
Material properties, for each grade of materials, are derived from relevant
experiments on samples of sufficient number, following appropriate
standardized procedures. For the reliability assessment, the properties of
interest are the yield strength, Young’s modulus, Poisson’s ratio, and fracture
toughness. Those properties are presented in this section.
4.4.2.1 Yield Strength of Steel Specimens
As it has been referred in the description of the Limit State Design Method in
Chapter Two, the characteristic yield strength is defined as the 5 percent
quantile of the test data [234] and [25]. In Table 13, the results of a study
presented in [235] are presented. In this study, different grades of steel
specimens have been tested and the Coefficient of variation has been derived
showing that as the nominal value increases, this value decreases. Another
interesting result of the experiments is that the yield strength follows a Log-
Normal distribution; a fact that is countersigned by the DNV standards’
provisions for reliability assessment of offshore structures [14].
145
Material Distribution Yield Strength CoV
(MPa) (%)
<350 8.0
Steel Lognormal 350-400 6.0
>400 5.0
Table 13: Yield Strength Properties [235]
4.4.2.2 Young’s Modulus and Poisson’s Ratio
The property of Young’s modulus for steel specimens is found to follow a

Normal distribution while for different materials, such as aluminium and
concrete, a fixed value might be selected. In [236], a mean value of 2.1x1011
(Pa) with a coefficient of variation of 5% is proposed, while in [237] a value of 6
%. Poison’s ratio can be represented by a fixed value of 0.3 for steel. Table 14
presents cumulatively data for Young’s modulus and Poisson’s ratio.
Poisson’s ratio
Material Distribution Young’s modulus
Mean CoV Mean CoV
(MPa) (%) (%)
Steel 210x10
3
5.0 0.3 -
Aluminium Normal 7.0x10

4
- 0.3 -
Concrete 3.0x10
4
- 0.2 -
Table 14: Young’s modulus and Poisson’s ratio
4.4.2.3 Fracture Toughness
Fracture toughness in steel specimens is characterized by stress intensity

factors ‫ܭ‬ூ஼ and Crack Tip Opening Displacements (CTOD). Although initially
considered to follow lognormal distributions, a two-parameter Weibull
distribution was found to provide a better fit for (CTOD), as is documented in
[238]. Due to the fact that fracture analysis will not be covered in this Thesis,
further details for the consideration of fracture is beyond this scope and will not
be included.
146
4.4.3 Geometry Data
Geometrical data refer to the physical quantities that describe the shape, size,
cross sections and tolerances of members and structures. Compared to the
material and loading properties, uncertainties in geometrical data are
significantly less important since they depend on the production procedure of
the elements and the quality of manufacturing and also the accuracy of the
post-manufacture measurements; factors that can be efficiently controlled by
means of quality assurance etc. Geometrical properties may be assumed to be
deterministic variables, however in cases where their deviation has a significant
effect on the structural response of the structure they can be considered as
stochastic variables or as parameters of variables that describe actions or
structural properties.
Eccentricities, misalignments, inclinations and curvatures are geometrical

anomalies due to poor fabrication of members or inefficient assembly. Further
than affecting the actual capacity of the members and as a result of the
structure, eccentricities, misalignments and inclinations might amplify dynamic
loads and therefore might cause fatigue issues on the structure while curvatures
may have important influence in the buckling capacity of members.
Material Distribution Thickness CoV

(mm) (%)
12.7 1.8
Steel Normal 25.4 1.0
50.8 0.7
Table 15: Variation in thickness of steel plates [239]
In the fabrication process of steel members, physical wearing in the shaping

rollers of the steel mill will cause variation in the corss sectional dimensions of
the members. Further, selection of the acceptable levels of tolerance will
determine the accuracy of the manufacturing process. An extensive study made
in [239] in I shaped beam elements has shown that the relative variation in
height and flange width of the specimens is small, compared to the variation in
thickness. Further, it is observed that flanges tend to be thinner while the web to
147
be thicker than its nominal values. Table 15 presents statistical properties of
experimental data for specimens of different thicknesses [239].
4.4.4 Capacity of Members and Joints
This section deals with the buckling performance of structural members.

Buckling curves illustrate the capacity performance of members to plastic
deformation; extensive work is available in [239]. Design standards of steel
structures extensively describe the consideration of buckling in members, while
an important background reference is [240]. Further background literature in
buckling curves is available in [241] and [242]. Although elastic analysis is used
in traditional design, consideration of non-linear plastic analysis might be used
in order to take advantage of post-buckling plastic performance of structural
members. Due to the complicated requirements for the consideration of a non-
linear performance several items should be considered in the implementation in
conjunction to design standards [8]:
 Procedure for the identification of failure modes should be appropriately

included taking into consideration the non-linear analysis.
 Modelling of joints should incorporate the effect of local details for end
restraints or force-deformation relationships.
 Fabrication tolerances (member straightness and joint eccentricities)
should be effectively included.
 Residual stresses on buckling capacity should also be considered.
 Failure criteria in terms of maximum strain at failure for components
containing relevant imperfections from, e.g., welding and at regions
containing notches.
 Repeated yielding in case of reversed loading due to, e.g., wave action.
 Sensitivity of input parameters and analysis assumption for evaluation of
acceptance criteria.
Application of non-linear analysis definitely requires more effort and skill from
the engineer than that of traditional deterministic analysis of working within the
elastic region of materials; however if appropriately executed it might produce
148
more optimized structures. It should be noted that a systematic procedure for
the use of non-linear analysis of offshore structures does not presently exist.
Regarding the axial capacity of stiffened plates a comparative study presented

in [243] summarizes the provisions of different design codes. Data for those
properties are essential for the design of floating structures and therefore will
not be discussed further here. As far as the design of shell members and shell
frame structures are concerned, several standards have been developed, while
uncertainties have been studied in [244] and [245]. Most updated studies should
be considered and design standards should be updated.
In [246] and [247], the design equations for static strength of grouted joints are
derived. Test data for static capacity of tubular joints are available in [248].
Further work on the formulation of capacity equations is available in [244] and
[249]. In a reliability assessment, background of the databases and equations
used should illustrate the failure criterion used for each analysis. Therefore,
definition of the characteristic design equations in various design standards
varies since they are associated with different bias. Background literature for
the derivation of those equations in standards can be found in [250] and [251].
4.4.5 Geotechnical Data
Modelling of geotechnical data may include aleatory and epistemic uncertainty;

therefore stochastic representation can be applied through appropriate
probability distribution functions. Statistical properties of soil depend on a
number of parameters and can be obtained by several methods including a
Monte-Carlo simulation scheme. Different statistical tools may be more suitable
for problems of different nature and depending on the number of data points
and simulations that can be realized. Although variables that are related to
geotechnical data will not be considered in the application part of this Thesis, a
brief reference will follow for reasons of completeness.
149
4.4.5.1 Soil Characteristics
An initial statistical approach is that of short-cut estimates, that can be used in

cases of little data available. The method can obtain the mean and variance and
gives a bound for standard deviation, and has been proven useful for cases of
‘symmetrical’ data. Benefit of the method is that it can be applied relatively
quickly. The stochastic interpolation approach, proposed in [252] and [253] can
be also applied to describe and present the properties of a site. The method can
derive several soil properties but demands adequate data to be available. A
more empirical method suggests use of histograms or fitted probability
distribution functions. This method is maybe the most preferable one and
should be used whenever the data sample allows so, deriving the distribution
function by plotting in probability paper or by using statistical fit-tests.
Soil characteristics are summarized in Table 16 together with proposed

appropriate probability distribution functions. Summary of statistical properties
can be found in [254]. It can be observed that most of the characteristics follow
a Normal or Lognormal distribution, however in specific cases may be described
by uniform distributions. In the presence of adequate data, validation of the
stochastic representation of each characteristic should be executed through
appropriate distribution fit tests.
In cases of large soil volumes, in order to avoid local fluctuations of soil

properties, the averaged property is used. In order to avoid negative values in a
distribution, the lognormal distribution may be prefered. Therefore,
characterization of the soil volume under consideration is an essential decision
before proceeding to statistical analysis. In cases of large volumes, the site
description strategy might be employed [252] and [253]; two steps are
distinguished:
 Identification of the structure of the soil data.

 Use of the ‘kriging’ stochastic interpolation technique to estimate the soil
property at the location of interest.
150
Probability Mean
Soil Characteristic Soil Type CoV
distribution function Values
Sand Lognormal * *
Cone Resistance
Clay Normal/Lognormal * *
Marine
Lognormal * 5-20%
Clay
Undrained Shear
Strength su Clay Lognormal * 10-35%
Clay silt Normal * 10-30%
su normalised w.r.t.
Clay Normal/Lognormal ** 5-15%
Vertical effective stress
Plastic Limit Clay Normal 0.13-0.23 3-20%
Liquid Limit Clay Normal 0.30-0.80 3-20%
Submerged unit weight All soils Normal 4.5-11 0-10%
Friction angle Sand Normal * 2-5%
Void ratio and porosity,

including intial void All soils Normal * 7-30%
ration
Over consolidation ratio Clay Normal/Lognormal 1.2-40 10-35%
*: Site and Soil type-dependent
**: Function of over consolidation ratio
Table 16: Probability Density Function for Soil Characteristics [254]
Those methods can increase the quality of data and overcome the natural
heterogeneity of the soil characteristics. The uncertainties involved in the soil
characteristics, represent values for the parameters that are used for the
analysis of bearing capacity of gravity structures, jack-up platforms and suction
anchors, where "shallow" type of foundation failure is modelled. For piled
structures, relevant uncertainties are studied in the following section.
151
4.4.5.2 Piled Foundations
Empirical methods derived from onshore tests on piles of small scale for the
axial pile capacity have derived design parameters applicable to piles for
offshore applications. The design variables considered for clay volumes are the
skin friction factor along the pile, the undrained shear strength, the end bearing
factor and appropriate factors to account for the correlation between site
specific effects. As far as sand volumes are concerned, the design variables
considered include coefficient of lateral soil stress, soil-pile friction angle,
limiting skin friction, bearing capacity factor and limiting end bearing. API
suggests values for pile design parameters in clay.
As far as sand characterization is considered, differences among available

methods exist due to the following factors [8]:
 The value of the earth pressure coefficient, and thus the value of the
effective stress, allowed in the calculations of axial pile capacity for
compression and tension loading.
 The limiting side friction value, and the extent to which it depends on
relative density.
 The limiting end bearing value and its dependence on relative density.
4.4.5.3 Uncertainty in Calculation Model and Load Effects
Uncertainties in the calculation model depend on the problem under

consideration. The statistical properties of the soil model are difficult to
determine, however a useful approximation can be based on literature review,
comparisons of model test results with calculations, survey of expert opinions or
relevant case studies and back-calculation. Further, discussion on studies of the
uncertainty in different calculation models can be found in [255] while extensive
studies exist for the modelling uncertainty for shallow types of foundations and
quantification of model uncertainties [256] and [257]. Modelled uncertainty is
characterised through the parameter of bias; value of bias greater than unity
implies a conservative model, under predicting the actual capacity while a value
152
less than unity refers to a calculation model that over predicts the actual
capacity.
Uncertainties in loads and load effects, either aleatory or epistemic, should be

considered in a systematic way as they might significantly affect the
geotechnical analysis. Engineering practice shows that load uncertainties are
dominant in the structural response when compared to geotechnical
uncertainties. Since influence of load effects depends on soil properties,
parametric and sensitivity studies should be included to illustrate the actual
effect of an incremental change on the deterministically modelled soil properties
to the load effects.
4.4.6 Fatigue Data
Although fatigue limit states are not covered within this Thesis, this section will
include some information regarding consideration of capacity of structures
under fatigue loading. [258], gives S-N data for welded connections for different
classes of structures in different environments (air, seawater and corrosion free)
in the region between 105-107 cycles to failure. This is a common number for
offshore structures. In cases where this range is exceeded, S-N data might be
un-proportionally larger than that of the standard. Further data for crack growth
parameters, such as ‫ ܥ‬and ݉ for Paris’ crack growth formula are also included.
From a stochastic perspective, ݉ is modelled as a fixed parameter while ݈݊‫ܥ‬
follows a normal distribution, implying that ‫ ܥ‬follows a lognormal distribution.
Further literature on the stochastic modelling of those variables can be found in
[259] and [260].
It is common practice in offshore and further large scale structures to avoid

using steel of normal grade and instead use high strength steel with yield
strength of more than 450 MPa. Initiation period of this kind of steel will be
longer and this should be taken into account since most of the available S-N
data are derived from experiments in normal graded steel specimens. This fact
may impose a level of extra conservatism.
153
Based on experimental work and finite element analysis models, starting from
simpler geometries and moving to more complicated configurations, parametric
formulas for the derivation of stress concentration factors are available in
literature [261], [262], [263]. Stochastic properties of the ratio between
measured and predicted values are summarized in [264]. This ratio indicates
the bias of the experimental procedure. The derived mean values are in the
range 0.81 − 1.01 and the corresponding COV's in the range 0.13 − 0.25.
Finally for the cumulative damage on a structure, the Miner-Palmgren

Hypothesis can be applied for calculation of reliability. For steel structures, the
Miner’s damage ratio is proposed to follow a lognormal distribution with mean
value 1 and CoV of 0.3 [265].
4.4.7 Corrosion Modelling
4.4.7.1 Definition of corrosion
Offshore and marine structures are designed to survive harsh and very
corrosive environments. The level of capacity deterioration due to this effect
should be considered throughout its service life, in order for the structure to
sustain adequate safety levels. Systematic maintenance and corrosion
protection systems, that refer to surface coating and cathodic protection
arrangements are often employed to restrict the effect of corrosion on the
structures. Extensive research is available regarding the performance of
vessels or pipelines deployed offshore.
In [266], a classification of marine corrosion may be divided into the following

categories, based on the deterioration mechanisms that are developed on the
surface of plate materials:
 Immersion
 Splash/tidal zone
 Atmospheric
 Semi-enclosed space
154
Characterization of corrosion may be also based on the fact that the depth
profile of a corrosion instance might distinguish a single defect or a cluster of
defects through the thickness of the material. A definition based on this
performance distinguishes [267]:
 Pitting corrosion, which refers to corrosion with length and width less
than three times that of the un-corroded wall thickness.
 General corrosion (wastage), which refers to corrosion with length and
width more than three times the un-corroded wall thickness.
From a structural reliability perspective, corrosion together with fatigue have

time-dependent effects on the structure. In particular, there is a monotonic
decreasing performance in the deterioration of Resistance as a function of time
accompanied by an increase in uncertainty in the remaining strength. Figure 38
[268] and Figure 39 [269] present those two facts. In Figure 39, it can be
observed that the distribution that describes the probability density function of
the structural resistance ‘flattens’ with time and the mean value decreases,
increasing the overlapping between the load (effect) and resistance curves.
There are a number of parameters that influence the corrosion performance of a

structure. Environmental factors, such as biological, chemical factors and
physical parameters might significantly determine the corrosion effect on
structural members. From this category, the physical factors are the most
important ones, since they account for temperature, pressure (significant in pipe
lines), water velocity, suspended solids and percentage wetting. In general it
has been observed that increased temperature will increase the corrosion rate,
as it is observed in Figure 40 [269]. Water particles velocity has the same effect,
as presented in Figure 41 [268], as increased velocity will increase corrosion
losses. Further, factors such as pollution and water salinity may influence
corrosion on the structure.
155
Figure 38: Continuous load process and potential exceedance of the
deteriorating structural Resistance [268]
Figure 39: Strength Deterioration Structural Reliably Problem [269]
156
Figure 40: Corrosion depth as a function of sea water temperature [269]
Figure 41: Effect of water velocity on early loss corrosion [268]
157
4.4.7.2 Corrosion Models
Several models have been proposed for consideration of thickness deterioration

due to corrosion, both considering different periods of the life of the structure as
well as for its whole service life. Most of the models are initially based in
experimental work followed by scientific foundation of findings. The simplest
way to account for corrosion is through constant annual rate deterioration.
Based on the work of [270] and [271], typical values of annual deterioration can
be found, and are illustrated in Table 17. Although those values can be reliable
for initial design of structures, in case of reassessment of existing structures a
more advanced model should be employed.
Environmental conditions Corrosion rate (mm/year)
Immersed steel specimens, ocean conditions 0.05 – 0.20
Steel sheet piling – immersion zone 0.08
Tidal zone 0.10 – 0.25
Atmospheric zone 0.05 – 0.10
Ship deck plating (uncoated) 0.10 – 0.50
Table 17: Typical annual corrosion rates [270], [271]
Although in most of the studies executed so far the effect of corrosion is

represented by a constant corrosion rate, which corresponds to a linear
decrease of the plate thickness throughout its service period, there is
experimental evidence that a non linear model would be more appropriate.
Before proceeding to description of those models, a reference to the
background of corrosion should be made. General corrosion is of main interest
in this aspect since pitting corrosion cannot affect the main in plane stress
distribution in a steel plate [272].
As it has already been mentioned, corrosion is affected by a number of

parameters therefore a probabilistic model such as the one presented in [269]
including a mean value expression and an expression for randomness and
uncertainty can more systematically describe the expected corrosion:
158
‫ݐ(ܥ‬, ܲ, ‫݂ = )ܧ‬௡ (‫ݐ‬, ܲ, ‫ )ܧ‬+ ߝ(‫ݐ‬, ܲ, ‫)ܧ‬ (4-22)
Where: ‫ݐ(ܥ‬, ܲ, ‫ )ܧ‬is the total weight loss of material, ݂௡ (‫ݐ‬, ܲ, ‫ )ܧ‬the mean function
and ߝ(‫ݐ‬, ܲ, ‫ )ܧ‬the zero mean error function, ‫ݐ‬is the time parameter, ܲ is a vector
with parameters for the corrosion protection systems included, and ‫ ܧ‬is a vector
of environmental conditions. Throughout the service life of the structure,
different corrosion processes can be distinguished and are illustrated in Figure
42 [269]:
 initial corrosion
 oxygen diffusion controlled by corrosion products and micro-organic
growth
 limitation on food supply for aerobic activity
 anaerobic activity
Figure 42: Different corrosion mechanism as a function of time [269]
For general corrosion several models exist for a more systematic description of
corrosion deterioration. In [273], Southwell proposes a linear and a bilinear
model while Melchers [269] extends this model, calculating coefficients based
on available experimental data. As presented in [272], the extended Southwell
linear model may be stochastically formulated as:
ߤௗ (‫ = )ݐ‬0.076 + 0.038‫ݐ‬ (4-23)
159
ߪௗ (‫ = )ݐ‬0.051 + 0.025‫ݐ‬
Correspondingly, the extended Southwell bilinear model may be written as:
0.09‫ݐ‬, 0 < ‫ <ݐ‬1.46 ‫ݏݎܽ݁ݕ‬

ߤௗ (‫ = )ݐ‬൜
0.076 + 0.038‫ݐ‬, 1.46 < ‫ <ݐ‬16 ‫ݏݎܽ݁ݕ‬
(4-24)
0.062‫ݐ‬, 0 < ‫ <ݐ‬1.46 ‫ݏݎܽ݁ݕ‬
ߪௗ (‫ = )ݐ‬൜
0.035 + 0.017‫ݐ‬, 1.46 < ‫ <ݐ‬16 ‫ݏݎܽ݁ݕ‬
The exponential model proposed by Melchers, based on the Southwell linear

formulation, is the Melchers-Southwell non linear model, and is expressed as
follows:
ߤௗ (‫ = )ݐ‬0.084‫ݐ‬଴.଼ଶଷ
(4-25)
ߪௗ (‫ = )ݐ‬0.056‫ݐ‬଴.଼ଶଷ
Finally, in [274], a tri-linear and a different power approximation model are

suggested for the corrosion wastage thickness as:
0.170‫ݐ‬, 0 < ‫ <ݐ‬1

݀(‫ = )ݐ‬൝ 0.152 + 0.0186‫ݐ‬, 1 < ‫ <ݐ‬8 (4-26)
−0.364 + 0.083‫ݐ‬, 8 < ‫ <ݐ‬16
A different nonlinear model, based on observation data, has been proposed in

[275]. According to this, the corrosion process is divided in three phases; the
first where no corrosion is present due to effectiveness of corrosion protection
means, the second which is initiated when the corrosion protection becomes
damaged and the third which corresponds to the phase where the corrosion
process stops. The practical interpretation of the last period implies that due to
the corroded external surface of the plate, protection is provided. This should be
taken into account since partial or full removal of this layer will initiate a new
corrosion cycle in the corroded thickness of the plate. Appropriate formulations
of the model can be described as the solution of a differential equation of the
corrosion wastage:
̇ ) = ݀ஶ
݀ஶ ݀(‫ )ݐ‬+ ݀(‫ݐ‬
160
Where: ݀ஶ is the long-term thickness of the corrosion wastage, ݀(‫ )ݐ‬is the
thickness of the corrosion wastage at time ‫ݐ‬, and ݀(‫ )ݐ‬is the corrosion rate. The
solution of the above equation leads to:
݀ ൫1 − ݁(ି(௧ିఛ೎)/ఛ೟) ൯, ‫߬ >ݐ‬௖
݀(‫ = )ݐ‬ቊ ஶ
0, ‫߬ <ݐ‬௖
(4-27)
݀ஶ
߬௧ =
‫ߙ݃ݐ‬
In the above equations, ߬௖ is the coating life (1.5-5.5 years), equal to the time
interval between the painting of the surface and the time when its effectiveness
is lost, and ߬௧ is the transition time (5-10 years) time and ߙ is the angle as
formed in Figure 43.
Figure 43: Thickness of corrosion wastage as a function of time [275]
Interesting work presented in [272], develops a parametric approach to the

corrosion modelling introducing the Weibull function. The advantage of this
distribution is that based on the values of specific parameters, different shapes
of distributions can be derived. Further, by setting specific values in some of the
parameters of the model, the more straight forward models that were described
above can be obtained. Due to the strong mathematic formulation of this
method, it will not be presented in this Thesis; however a comparative example
that was executed among methods based on experimental data and using
161
statistical-fit methods to derive the best approach illustrates that this new
method has a very good potential.
As far as pitting corrosion is considered, the problem that might be cause on the
structure is deep holes that could lead to penetration. The local character of
pitting corrosion makes it less significant than that of wastage corrosion. For
reasons of completeness, following similar logic as for the general corrosion, a
refined model is presented in Figure 44 [276], [277]. The parameters of this
figure are explained in Table 18 together with a calculation model.
Figure 44: Model for maximum pit depth as a function of exposure period
[276]
Phase Phase description Governing parameters

and corrosion as function of ࢀ
controlling mechanism
0 Initial pit growth
Pit growth under overall ‫ݐ‬௔ = 6.61 ∙ ݁‫( ∙ ݌ݔ‬−0.088 ∙ ܶ)
1 and 2
aerobic conditions under rust cover ‫ܥ‬௔௣ = 0.99 ∙ ݁‫( ∙ ݌ݔ‬−0.052 ∙ ܶ)
Rapid pit growth under overall
3 ‫ݎ‬௔௣ = 0.596 ∙ ݁‫( ∙ ݌ݔ‬0.0526 ∙ ܶ)
anaerobic conditions under rust cover
Steady-state pit growth under ‫ܥ‬௔௣ = 0.641 ∙ ݁‫( ∙ ݌ݔ‬0.0613 ∙ ܶ)
4 overall anaerobic conditions
‫ݎ‬௔௣ = 0.353 ∙ ݁‫( ∙ ݌ݔ‬−0.0436 ∙ ܶ)
under rust cover
Table 18: Phases in pitting corrosion and calibrated functions for model
parameters [276]
162
4.4.7.3 Corrosion Protection Systems (CPS)
Painting coatings and cathodic protection schemes are the main techniques
applied for the protection of steel members against corrosion, and are included
in Appendix B of this Thesis. Those measures, together with appropriate
maintenance, should be able to efficiently treat corrosion. However, practice
from field observations shows that maintenance in offshore structures might not
always be effective and CPS might be ineffective. Further, some parts of an
offshore or marine structure are difficult to access and therefore apply any
maintenance action.
4.5 Summary
This Chapter has presented the stochastic modelling of environmental loads

and load bearing capacity of offshore structures. Wind, wave and current loads
have been considered analytically. Especially for wave loads, after a thorough
discussion of the transformation of sea state conditions to loads acting on the
structural members, the joint probability modelling of environmental variables
has been discussed. The capacity of offshore structures is also included in this
Chapter. Stochastic modelling of material data has been presented in detail
while geotechnical data are also briefly included. Finally, this chapter presents
analytically the mechanisms of corrosion that lead to capacity deterioration of
structures, and the available models to account for general corrosion during the
design process.
163
5 APPLICATION OF THE RESPONSE SURFACE
METHOD IN A TYPICAL JACKET STRUCTURE
5.1 Introduction
Based on the theory and methodology that has been presented previously, this
chapter will present the application of structural reliability assessment for a
typical reference offshore jacket structure. This analysis will start with a
discussion on the analytical formulation of the limit state functions based on
widely acceptable and used failure criteria. Following that, an application of the
joint probability distribution of wave height and peak spectral period will be
presented for a typical North Sea site, in order to determine an analytical
relationship between the two correlated variables. Having obtained this, and
after description of the reference structure that has been selected, its reliability
will be derived in a local and later global-system level considering stochastic
modelling of variables, based on the Stochastic Response Surface Method.
Material properties, sea state conditions, selection of appropriate wave theory,
effect of different statistical distributions in the modelling of the variables will be
examined in addition to a thorough investigation of the effect of corrosion on
structural members based on different thickness deterioration models.
5.2 Limit State Formulation
Appropriate formulation of limit states should link the performance of the

structure with appropriate failure mechanisms that might occur on each
member. For the scope of this PhD, limit state functions will be formulated
based on the failure criteria commonly applied in structural design. For the
reference structure that will be used, static analysis will determine its ultimate
strength; therefore static strength will be of interest here.
Strength of a structural member is a property that is determined by its identity,

the treatment and processing during and after manufacturing, and the loading at
165
which it will be subjected during its service life. For cases of complex structures,
like the ones that will be studied here, the position of each element will also
determine its criticality as a result of the effect of the loads acting on the whole
structure.
Static loads refer to the case where a stationary force or a couple (moment) is
applied to a member and remains unchanged in magnitude, point and direction
of application. Result of a static load might be axial tension or compression, a
shear load, a bending load, a torsional load, or any combination of these. In this
section, the relation between static loading and strength is discussed, in order
to later focus the attention on the predictability of potential failures that might
result in degradation of reliability and eventually lead to failure.
In Chapter 2, reference has been made to brittle and ductile failure performance
of structures. The same classification can be followed for members’ failure,
characterizing them in a way that will identify the member’s loading
performance and allocate appropriate applicable failure mechanisms in order to
efficiently design against failure. Although structural members belong to one of
those categories, a ductile material can fail in a brittle manner under specific
loading conditions. The basic limit for the characterization of a material as
ductile is based on the maximum strain (elongation) before failure ߝ௙ and
suggests [278]:
ߝ௙ ≥ 0.05 (5-1)
Ductile materials have a yield strength which is considered the same for
compressive and tensile loads ܵ௬௧ = ܵ௬௖ = ܵ௬ . Contrary to that performance,
brittle materials do not have an identifiable yield strength, constituting
engineering decision making more difficult, and are classified by individual
consideration of ultimate tensile and compressive stresses ܵ௨௧ and ܵ௨௖
respectively. After years of engineering hypothesis, accepted practices have
been formally developed to theories that are currently applied in modern
engineering. Different theories are distinguished for different types of materials
resulting to the following been proposed [278]:
166
Ductile materials (yield criteria):
 Maximum shear stress (MSS). Commonly referred to as the Tresca or

Guest theory, it predicts that ‘yielding begins whenever the maximum
shear stress in any element equals or exceeds the maximum shear
stress in a tension test specimen of the same material when that
specimen begins to yield’.
 Distortion energy (DE). Alternatively called the von Mises or von Mises–
Hencky theory, shear-energy theory or octahedral-shear-stress theory, it
predicts that ‘yielding occurs when the distortion strain energy per unit
volume reaches or exceeds the distortion strain energy per unit volume
for yield in simple tension or compression of the same material’.
 Ductile Coulomb-Mohr (DCM). Also known as internal-friction theory, it is
applicable for materials with different tensile and compressive strength
and provides a geometrical representation of stress states.
Brittle materials (fracture criteria):
 Maximum normal stress (MNS)

 Brittle Coulomb-Mohr (BCM)
 Modified Mohr (MM)
A universally applicable theory for any material is not currently available

therefore different theories are more suitable for different problems. In the
following sections, the above mentioned yield criteria will be briefly presented
based on [278], [279] and [280]. After presentation of the criteria a critical
comparison between them, will distinguish the applicability region of each of
them. Presentation of the fracture criteria for brittle materials is beyond the
scope of this work and therefore will not be included.
167
5.2.1 Maximum shear stress (MSS) – Tresca or Guest Theory
Following the observation that on an element that undergoes a tensile load, slip
(Lüder) lines are formulated at approximately 45° with the axis of the strip
declaring initiation of yield, and in parallel observations in cases of loading to
fracture where shear stress is maximized at 45° from the axis of tension, a
failure criterion is suggested to describe this failure mechanism. For the
fundamental case of a simple tensile test, the corresponding stress might be
given as:
ܲ
ߪ= (5-2)
‫ܣ‬
Considering the fact that the maximum shear stress occurs on a surface 45°
from the tensile stress, the maximum shear stress should be described and
correlated to yield strength ܵ௬ as follows:
ߪ ܵ௬
߬௠ ௔௫ = = (5-3)
2 2
Following connotation from Mohr’s circles for three dimensional stress, for a
general stress state three principle stresses might be determined as ߪଵ ≥ ߪଶ ≥
ߪଷ. Expressing the shear stress using principal stresses, the above equation
might be expressed as:
ߪଵ − ߪଷ
߬௠ ௔௫ = (5-4)
2
And the corresponding failure criterion:
ௌ೤
߬௠ ௔௫ ≥ or ߪଵ − ߪଷ ≥ ܵ௬ (5-5)
ଶ
An application of safety factor ݊ might be considered for design purposes as:
ௌ೤ ௌ೤
߬௠ ௔௫ ≥ or ߪଵ − ߪଷ ≥ (5-6)
ଶ௡ ௡
168
Considering formulation of limit state functions in the form of the difference
between supply (allowable load) and demand (actual load) the corresponding
LSF is derived denoting failure for the MSS criterion:
ௌ೤
‫ܩ‬ெ ௌௌ = − ߬௠ ௔௫ ≤ 0
ଶ௡
(5-7)
ܵ௬
‫ܩ‬ெ ௌௌ = − (ߪଵ − ߪଷ) ≤ 0
݊
Plane stress (or membrane state stress) are problems which refer to cases
where all loads applied, are symmetric to the mid-plane direction and at the
same time in-plane displacements, strains and stresses can be considered to
be uniform through the thickness, and the normal and shear stress components
in the ‫ ݖ‬direction are negligible. In such problems, one of the principal stresses
is zero therefore only two components can be distinguished (ߪ஺ and ߪ஻ ). Table
19, summarizes the formulation of the criterion for each case of principle
stresses while Figure 45 represents the MSS theory in plane stress problems.
Figure 45: The MSS theory for plane stress problems (two nonzero
principal stresses)
169
Principle Principle Stresses Failure
Case
Stresses Correlation Criterion
1 ߪ஺ ≥ ߪ஻ ≥ 0 ߪଵ = ߪ஺ ߪଷ = 0 ߪ஺ ≥ ܵ௬
2 ߪ஺ ≥ 0 ≥ ߪ஻ ߪଵ = ߪ஺ ߪଷ = ߪ஻ ߪ஺ − ߪ஻ ≥ ܵ௬
3 0 ≥ ߪ‫ܤߪ ≥ ܣ‬ ߪଵ = 0 ߪଷ = ߪ஻ ߪ஻ ≤ −ܵ௬
Table 19: MSS Failure Criterion for Plane stress problems
5.2.2 Distortion energy (DE) – von Mises Stress Theory
This theory was developed through the observation that ductile materials
stressed hydrostatically exhibited greater yield strength compared to the values
given by a simple tensile test. This phenomenon integrates the problem to a
more complicated one, considering the yielding that is related to the angular
distortion of the stressed element rather than been described as a simple
tensile or compressive test. Introducing the term of hydrostatic stress ߪ௔௩ in
each principal direction of the three dimensional stress states, the correlation
between principal stresses is given as:
ߪଵ + ߪଶ + ߪଷ
ߪ௔௩ = (5-8)
3
Considering the above, in a member under tri-axial stress, that undergoes both
volume change and energy distortion, each stress component can be described
by a hydrostatic ߪ௔௩ and a distortional component (ߪ௜ − ߪ௔௩). For the three
dimensional element, the strain energy per unit volume can be described as:
1 1
‫=ݑ‬ (ߝଵߪଵ + ߝଶߪଶ + ߝଷߪଷ) = [ߪଶ + ߪଶଶ + ߪଷଶ − 2‫ߪ(ݒ‬ଵߪଶ + ߪଶߪଷ + ߪଷߪଵ)] (5-9)
2 2‫ ܧ‬ଵ
Substituting the above two equations, the strain energy for producing volume
change ‫ݑ‬௩ is given as:
ଶ
3ߪ௔௩ 1 − 2‫ ݒ‬ଶ
‫ݑ‬௩ = (1 − 2‫= )ݒ‬ [ߪଵ + ߪଶଶ + ߪଷଶ + 2ߪଵߪଶ + 2ߪଶߪଷ + 2ߪଷߪଵ] (5-10)
2‫ܧ‬ 6‫ܧ‬
The distortion energy is obtained as:
170
1 + ‫ߪ( ݒ‬ଵ − ߪଶ)ଶ + (ߪଶ − ߪଷ)ଶ + (ߪଷ − ߪଵ)ଶ
‫ݑ‬ௗ = ‫ ݑ‬− ‫ݑ‬௩ = ቈ ቉ (5-11)
3‫ܧ‬ 2
Considering that for a simple tensile test, the distortion energy is given as:
1+‫ ݒ‬ଶ
‫ݑ‬ௗ = ܵ (5-12)
3‫ ܧ‬௬
The general state of stress the failure criterion is then formulated as:
ଵൗ
(ߪଵ − ߪଶ)ଶ + (ߪଶ − ߪଷ)ଶ + (ߪଷ − ߪଵ)ଶ ଶ
ߪᇱ = ቈ ቉ ≥ ܵ௬ (5-13)
2
Where: ߪᇱ is the von Mises stress.
For a plane stress problem with two non zero principal stresses, as it was
presented before, the expression for the von Mises stress is written as follows,
representing a rotated ellipse as shown in Figure 46, including both DE and
MSS theories:
ଵൗ
ߪᇱ = (ߪ஺ଶ − ߪ஺ ߪ஻ + ߪ௕ଶ) ଶ (5-14)
Moving to a Cartecian coordinate system for the three dimensional stress state,
the von Mises stress can be rewritten as:
1 ଶ ଶ
ߪᇱ௫௬௭ = ቂ൫ߪ௫ − ߪ௬ ൯ + ൫ߪ௬ − ߪ௭൯ + (ߪ௭ − ߪ௫)ଶ
√2
(5-15)
ଵൗ
ଶ ଶ ଶ ൧ ଶ
+ 6൫߬௫௬ + ߬௬௭ + ߬௭௫ ൯
And for plane stress problems:
ଵൗ
ߪᇱ௫௬௭ = ൫ߪ௫ଶ − ߪ௫ߪ௬ + ߪ௬ଶ + 3߬௫௬
ଶ
൯ ଶ (5-16)
Following the same principles for the formulations of the limit states for the
failure region in reliability analysis:
ܵ௬
‫ܩ‬஽ா = − ߪᇱ ≤ 0 (5-17)
݊
171
Figure 46: Distortion Energy Theory for plane stress problems
5.2.3 Ductile Coulomb-Mohr (DCM)
The ductile Coulomb-Mohr Theory is applicable for materials where strength

against compression is different than that due to tension. Based on the Mohr
theory of failure, it is a simplified procedure with results easy to be geometrically
interpreted. The concept of the theory is based on combination of tension,
compression, and shear tests observing performance to yield or fracture. Based
on results of such tests, a graph such as the one in Figure 47 may be drawn
and a closed region may be distinguished as the (non-straight) line ABCDE in
the figure, above the ߪ axis in order to describe the stress state of a body.
This theory also known as the internal-friction theory, assumes that the BCD
part of the envelop boundary is straight, implying that only tensile and
compressive strengths are necessary. Considering a circle with centre ‫ܥ‬ଶ, as
illustrated in Figure 47, that connects the maximum and minimum principal
stresses, geometrical relationships of similar triangles denote:
172
ߪଵ − ߪଷ ܵ௧ ܵ௖ ܵ௧
‫ܤ‬ଶ‫ܥ‬ଶ − ‫ܤ‬ଵ‫ܥ‬ଵ ‫ܤ‬ଷ‫ܥ‬ଷ − ‫ܤ‬ଵ‫ܥ‬ଵ −2 −2
= → 2 = 2 →
ܱ‫ܥ‬ଶ − ܱ‫ܥ‬ଵ ܱ‫ܥ‬ଷ − ܱ‫ܥ‬ଵ ܵ௧ ߪଵ + ߪଷ ܵ௖ ܵ௧
− +
2 2 2 2
(5-18)
ߪଵ ߪଷ
→ − =1
ܵ௧ ܵ௖
Where: ܵ௧ and ܵ௖ denote the tensile and compressive strength respectively. As

for the case of the MSS theory, Table 20 and Figure 48, summarize the stress
state of the plane stress problem.
Figure 47: Mohr Cycles and Coulomb-Mohr Failure criterion
Principle Principle Stresses Failure

Case
Stresses Correlation Criterion
1 ߪ஺ ≥ ߪ஻ ≥ 0 ߪଵ = ߪ஺ ߪଷ = 0 ߪ஺ ≥ ܵ௧
ߪ஺ ߪ஻
2 ߪ஺ ≥ 0 ≥ ߪ஻ ߪଵ = ߪ஺ ߪଷ = ߪ஻ − ≥1
ܵ௧ ܵ௖
3 0 ≥ ߪ‫ܤߪ ≥ ܣ‬ ߪଵ = 0 ߪଷ = ߪ஻ ߪ஻ ≤ −ܵ௖
Table 20: CM Failure Criterion for Plane stress problems
173
Figure 48: Coulomb-Mohr Failure criterion for Plane Stress problems
The corresponding limit state function ‫ܩ‬, will be formulated as follows for the
failure region:
1 ߪଵ ߪଷ
‫ܩ‬஼ெ = − ൬ − ൰≤ 0 (5-19)
݊ ܵ௧ ܵ௖
5.2.4 Comments on Failure Criteria
Based mainly on work presented in [281], performance of different ductile

materials has been studied in order to evaluate the applicability of each failure
theory and corresponding design criteria. Elaboration of those experimental
results illustrates that use of both the maximum shear stress theory and the
distortion energy theory is applicable to the design and analysis of ductile
materials. Selection of one theory rather than another is subjected to the
engineer’s decision and available tools. In general, the MSS theory is easier to
implement but it is more conservative leading to overdesigned structures (by a
percentage of approximately 15 %, which can be derived analytically for the
simple case of a tensile specimen). Coulomb-Mohr’s theory is applicable in
cases of ductile materials with unequal tensile and compressive strengths,
providing a simple equation where only those strengths participate and are
compared with principal stresses.
174
A useful comment on the different failure criteria that have been presented
above is that for the case of loaded members where stresses are present in one
of the axis, and one principal stress is zero due to the two dimensional nature of
the problem. The formulation of the principal stresses from the corresponding x-
y-z directions can be expressed as:
ߪ௫ + ߪ௬ ߪ௫ + ߪ௬ ଶ
ߪଵ,ଶ = ඨ
± ൬ ଶ
൰ + ߬௫௬ (5-20)
2 2
Where for the axially loaded member:
ଶ ଶ
ߪ௫ = ߪ௔,௫௫ + ටߪ௕,௫௬ + ߪ௕,௫௭
(5-21)
ߪ௬ = 0
From (5-20), and neglecting shear stresses:
ଶ ଶ
ߪଵ = ߪ௫ = ߪ௔,௫௫ + ටߪ௕,௫௬ + ߪ௕,௫௭
(5-22)
ߪଶ = 0
Applying the above in the limit state functions that have been derived earlier,
the following expression can be derived, corresponding and verifying all of
them:
ܵ௬ ܵ௬ ଶ ଶ
‫ܩ‬௘௤ = − ߪଵ = − ߪ௔,௫௫ + ටߪ௕,௫௬ + ߪ௕,௫௭ ≤0 (5-23)
݊ ݊
Or, alternatively expressed as:
ଶ ଶ
ߪ௔,௫௫ + ටߪ௕,௫௬ + ߪ௕,௫௭
‫ܩ‬௘௤ = ≤1 (5-24)
ܵ௬ൗ
݊
The above equation is adopted with minor alterations, mainly in the

consideration of the safety factor of the material yield and the vectorial sum of
175
the two stresses due to bending moments, by the standards’ provisions for
ultimate limit state. The provisions of different standards will be presented in the
next chapter. For cases where torsion and shear loads can be neglected, and
only axial load and bending moment are present, as in the general case of
frame structures, the above limit state corresponds to both the Tresca and the
von Mises yield criteria.
Commercial FEA packages can provide values of principal stresses that will
allow use of any of the criteria described above. Table 21, Summarizes the limit
state functions derived, based on each of the failure criteria.
ௌ೤
‫ܩ‬ெ ௌௌ = − ߬௠ ௔௫ ≤ 0
Maximum shear stress (MSS) - Tresca or Guest ଶ௡
theory ܵ௬
‫ܩ‬ெ ௌௌ = − (ߪଵ − ߪଷ) ≤ 0
݊
Distortion energy (DE) - von Mises–Hencky ܵ௬

‫ܩ‬஽ா = − ߪᇱ ≤ 0
theory ݊
Ductile Coulomb-Mohr (DCM)- Internal-friction 1 ߪଵ ߪଷ

‫ܩ‬஼ெ = − ൬ − ൰≤ 0
theory ݊ ܵ௧ ܵ௖
Table 21: Limit states for different Failure Criteria
5.2.5 Buckling Limit state
Further to failure criteria for the ultimate strength of structural members, an

additional limit state should be introduced for the analysis of members under
compression in order to avoid buckling phenomenon. For the appropriate
consideration of each case, members undergoing compressive loads should be
categorized as either long or intermediate long, with central or eccentric loads.
The term column is used for members where the principle stress derives mainly
from axial forces that usually fall along the centreline of the member.
For a vertical column that undergoes an axial force ܲ along its centroidal axis,
simple compression will occur for relatively low values of this force. Increasing
the value of ܲ, and under specific conditions, the column might become
176
unstable, leading to extensive bending. The critical force between those two
states, can be derived following the bending deflection equation of the member,
providing a differential equation where appropriate boundary conditions may be
applied for each problem and finally resulting in the critical load for unstable
bending (Euler column formula):
‫ߨܥ‬ଶ‫ܫܧ‬
ܲ௖௥ = (5-25)
݈ଶ
Where: ‫ ܧ‬is the Young modulus of the material, ‫ ܫ‬the moment of inertia and ݈
the free length of the member. The parameter ‫ܥ‬, accounts for the end
conditions of the member: ‫ = ܥ‬1 for both ends rounded or pivoted; ‫ = ܥ‬4 both
ends fixed; ‫ = ܥ‬0.25 one end free and one end fixed; ‫ = ܥ‬2 one end rounded
and pivoted, and one end fixed. An alternative useful expression for the critical
load can be obtained as:
ܲ௖௥ ‫ߨܥ‬ଶ‫ܧ‬
= (5-26)
‫ܣ‬ (݈/݇)ଶ
Considering: ‫ ܭܣ =ܫ‬ଶ, where: ‫ ܣ‬is the area and ݇ the radius of gyration, and
(݈/݇) is called the slenderness ratio which classifies the length of a column. The
ratio ܲ௖௥/‫ ܣ‬is called ‘crucial unit load’; it has the same units as strength and it
represents the necessary load per unit area that should be applied on the
column in order to exceed the stability equilibrium. It depends on Young
modulus and slenderness ratio, implying that it is independent of the material
yield strength.
Based on diagrams between the slenderness ratios and the unit load, also
called PQR ratios, a critical value of the slenderness ratio is proposed, in order
to distinguish different classes of beams (long, intermediate). A typical value
selected for this scope denotes that the Euler’s formula is applicable for cases
where slenderness ratio is greater than that:
݈ 2‫ߨܥ‬ଶ‫ܧ‬
൬ ൰ = (5-27)
݇ ଵ ܵ௬
177
Several formulas have been proposed for cases of slenderness ratio below
(݈/݇)ଵ, most based on a linear relationship between slenderness ratio and unit
load. Applying appropriate boundary conditions, the corresponding formula
suggests:
ܲ௖௥ ܵ௬ ݈ ଶ 1
= ܵ௬ − ൬ ൰ (5-28)
‫ܣ‬ 2ߨ ݇ ‫ܧܥ‬
Formulation of the corresponding limit states, will be derived by comparison of

the critical load to the actual axial force of each member. Therefore the
following LS functions represent failure:
஺஼గ మா ௟ ௟
‫ܩ‬௕ଵ = ܲ௖௥ଵ − ܰ௫௫ = (௟/௞)మ
− ܰ௫௫, for ቀ௞ቁ > ቀ௞ቁ
ଵ
(5-29)
ௌ೤ ௟ ଶ ଵ ௟ ௟
‫ܩ‬௕ଶ = ܲ௖௥ଶ − ܰ௫௫ = ‫ ܣ‬൤ܵ௬ − ቀଶగ ௞ቁ ൨− ܰ௫௫, for ቀ௞ቁ ≤ ቀ௞ቁ
஼ா ଵ
It should be noted that in most design standards, as it will be pointed out in the
next chapter, distinguish between different classes of members is presented,
following the above theory. This PhD has focused on the yield, buckling and
corrosion which are the main failure mechanisms in the design of structural
members. Further from that, for the comprehensive consideration of the integrity
of structural members, additional mechanisms such as the following should be
also considered:
 Creep, is a mechanism that forms gradually over time and can be

described as the tendency of solid materials to deform permanently
under the effect of stresses below the yield strength of materials. The
magnitude and rate of this mechanism is influenced by exposure to high
temperature and can lead to failure (ULS) or extensive deformation
(SLS).
 Fatigue, is another progressive, localized effect occurring at members
subjected in cyclic loading. The formation of the mechanism is initiated
by microscopic cracks that are formed in the surface of a member, and it
gradually propagates resulting to failure. Fatigue corrosion is another
178
mechanism that combines corrosion and cyclic loading. As it has already
been mentioned, this failure mechanism is of great importance for
offshore structures that undergo dynamic loads.
 Fracture, is described as the separation of a member subjected to stress
in to two or more pieces. Ductile and brittle materials have a different
fracture strength which is determined by tensile tests and stress-strain
curves.
 Impact, is considered as a sudden force or acceleration applied on a
member over a short period of time which forces the member to react in
an unforeseen, and often non linear way. The load bearing capacity of a
damaged member should be considered, in order to assess the
performance of the structure throughout the rest of its service life.
 Wear, is the erosion of a member due to the action of another surface
acting on it, due to the interaction of particles between the two surfaces.
This mechanism might result to cross sectional loss of the material which
might affect its load bearing capacity.
 In a global level de-pilling is another mechanism that might lead to global
failure, when the extensive lateral loads or seabed movement, tends to
dislocate the structure.
5.3 Application of Joint Probability Distribution
Before proceeding to the reliability assessment of the selected offshore jacket

structure, an example of the application of the joint probability distribution
between significant wave height and peak spectral period will be included, as it
is required for the next steps of this application, in the formulation of the variable
input matrix. Data for this analysis are obtained by [211] for a typical site (Grid
Point 14212 - 61.507°N, 0.942°W) in a binned data form for ‫ܪ‬௦ and ܶ௉ . For this
analysis, different Matlab codes have been conducted and have been verified
for the case presented in [282]. The matrix of the bivariate scatter diagram ܼ
and the mean values of the classes are the input of the initial code that
179
processes the real data. Linear and logarithmic contours of equal probability are
then drawn for this data; the later type deals with the fact that peaks that
correspond to more observations that occur at regions of lower interest gather
more of the contour lines underestimating the significance of lower probabilities
regions. Figure 49 and Figure 50 present those plots.
Plot of generated 2D data - Linear plot of Real data
5
Hs
0
0 2 4 6 8 10 12 14 16 18
Tp
.
Figure 49: Linear Plot of real data (Hs (m), Tp (s))
Plot of generated 2D data - Logarithmic plot of Real data
5
Hs
0
0 2 4 6 8 10 12 14 16 18
Tp
Figure 50: Logarithmic Plot of real data (Hs (m), Tp (s))
180
The next code that has been conducted, applies statistical fit in order to
generate the coefficients of distribution for ܶ௉ and ‫ܪ‬௦ for lognormal and Weibull
distributions. Before proceeding to approximation of the joint distribution
function, further analysis has been carried out in order to estimate the
correlation between the two variables. In order to achieve this, the following
procedure is introduced: starting from the scatter diagram, and taking the
coordinates of the centre of each bin, Cartesian points with a number equalling
to that of the observations were generated in order to apply the calculation
formula. For the generation of those points, a routine was followed that
produces points in a circular or elliptical pattern of radius equalling the half width
of each bin dimension. The results of this approach, after verification, have
been found to approximate the analytical solution sufficiently. Figure 51,
presents a contour generated by those reconstructed data. Table 22, presents
the coefficients estimated for this case.
Contour plot of generated Regenerated 2D data
5
Hs
0
0 2 4 6 8 10 12 14 16 18
Tp
Figure 51: Linear Plot of regenerated data (Hs (m), Tp (s))
181
Lognormal Approximation Weibull Approximation Correlation coefficient
TP μ = 2.1805 σ = 0.2709 λ = 10.1157 κ = 3.9406 ρ 0.5624
HS μ = 0.9792 σ =0.5027 λ = 3.4227 κ = 2.0934
Table 22: Distribution coefficients
Once the distributions’ coefficients have been estimated, three different codes
have been conducted in order to produce different contours that will allow
selection of the most appropriate distribution. Each code will generate bivariate
correlated random numbers. The procedure followed, starts with generation of a
sequence of sets of values following normal distributions ܰ(0,1) taking into
account a correlation coefficient. After that, for each set, the value of the Normal
cumulative distribution function ‫ܨ‬௡ is calculated and the final values of each
distribution will be derived as the inverse cumulative distribution function of
‫ܨ‬௡ିଵ(݅) which corresponds accordingly to any statistical distribution. This is an
easy procedure to program. The code will later bin data in a new scatter matrix
and finally linear and logarithmic contours will be plotted. Using Lagrange
interpolation, smoothened contours are produced. For the three different joint
distribution functions, the results are presented in Figure 52 through Figure 54.
Although linear plots are generated as well, the logarithmic ones will be
presented since they include more information for the contours of lower
probability.
182
Smoothened Logarithmic plot of LogN bivariate joint distribution
5
Hs
0
0 2 4 6 8 10 12 14 16 18
Tp
Figure 52: Logarithmic plot of LogN bivariate joint distribution

(smoothened) (Hs (m), Tp (s))
Smoothened Logarithmic plot of Weibull bivariate joint distribution
5
Hs
0
0 2 4 6 8 10 12 14 16 18
Tp
Figure 53: Logarithmic plot of Weibull bivariate joint distribution

183
Smoothened Logarithmic plot of LogN-Weibull bivariate joint distribution
5
Hs
0
0 2 4 6 8 10 12 14 16 18
Tp
Figure 54: Logarithmic plot of LogN-Weibull bivariate joint distribution

From the above plots, and based on visual observation [282], the bivariate
Weibull distribution provides poor approximation of the original data
underestimating the extreme values of ܶ௉ and allocating more points in ‫ܪ‬ௌ.
Further, the area of the higher contour is larger, providing a less steep peak.
The bivariate Lognormal and the marginal Lognormal-Weibull joint distributions
provide a better fit on the approximation of the original data with the later having
a close shape of the outer contour that is the one of interest, since it represents
the area of low probability. The codes that have been conducted also provide
the relative error contours between the original and the generated data in a
logarithmic plot, distinguishing regions of greater error concentration. According
to [283], the Lognormal distribution can provide good fit for cumulative
probabilities below 0.99, while it deteriorates above this threshold. Figure 55,
shows the comparison between the (reconstructed) original data, the Lognormal
and the marginal Lognormal-Weibull cumulative distribution functions for the
region of higher probabilities. It can be observed that the latter distribution
function approximates better the curve of the original data.
184
Cdf of Lognormal-Weibull distributions
1
0.95
0.9
cdf
0.85
0.8
LONOWE
0.75 LogN
5 6 7 8 9 10 11 12
Hs
Figure 55: Comparison of best fit on the tail region of the CDF for Hs (m)
Having determined the most appropriate joint distribution model, the long term
response of wave climate should be determined. There are two main
approaches for the prediction of the design value of waves [284]; the 100 years
storm and the 100 years design curve. The design curve approach refers to the
use of the joint distribution function defining the region that the return period,
which corresponds to the average period between exceedance of one value,
probability is exceeded. This method will be followed here and the return period
of 100 years will be selected since this is considered by most researchers and
modern standards.
From the marginal Lognormal-Weibull distribution function, for the tail region of
the cumulative distribution function, the extreme value corresponding to the 100
year return period can be evaluated, once the cumulative probability [ܲ(‫ܪ‬௦)] ்ೝof
a given return period ܶ௥ is calculated as [285]:
1
[ܲ(‫ܪ‬௦)] ்ೝ = 1 − (5-30)
8 × 365 × ܶ௥
Applying the inverse Weibull probability distribution function for the above
probability, will derive the corresponding value for the significant wave height as
185
‫ܪ‬௦ = 11.47 m. The conditional probability of the spectral peak period for a given
significant wave height is calculated as [283]:
ଶ
ܶ௉ = ݁‫݌ݔ‬ቈߤ்|௉ − ට1 − ߩு் ߪ் ቉
Where: (5-31)
ߪ்
ߤ்|௉ = ߤ் + ߩு் (݈݊‫ܪ‬௦ − ߤு )
ߪு
For the distributions that have been derived, and for the specific value of ‫ܪ‬௦ the
corresponding ܶ௉ = 11.0 sec. Based on literature [191], the results derived
above, are reasonable. The extreme value with the return period of 100 years is
a probabilistic value, so in order to derive a deterministic value of this value
several approaches have been proposed and are presented in Table 23, [286].
From this table, that summarizes the work of several researchers [287], [288],
[289], [290], [291], [292], [293], [294], Wiegel’s work [295], the following formula
correlates significant wave height to the maximum design wave height:
‫ܪ‬௠ ௔௫ = 1.87‫ܪ‬௦ (5-32)
The procedure described and followed above has been verified with the
example case presented in [230].
Reference Data Type ࡴ ࡿ⁄ࡴ ࡻ ࡴ ૚⁄૚૙⁄ࡴ ࡿ ࡴ ࢓ ࢇ࢞⁄ࡴ ࡿ
Munk [287] Field data 1.53 --- ---

Seiwell [288] Field data 1.57 --- ---
Wiegel [290] Field data --- 1.29 1.87
Barber [289] Theoretical 1.61 --- 1.50
Putz [292] Field data 1.63 --- ---
Longuet-Higgins [291] Theoretical 1.60 1.27 1.77
Putz [292] Theoretical 1.57 1.29 1.80
Darbyshire [293] Field data 1.60 --- 1.50
Hamada et al. [294] Experimental 1.35 --- ---
Table 23: Wave Height Statistical Correlations [286]
186
The values that are presented in Table 23, refer to a Reyleigh distribution fit of
observed data, which stands as a special case of the Weibull distribution. The
statistical values of interest of the Reyleigh distribution as presented in a
parametric plot in Figure 56, are the following:
 The average wave height ‫ܪ‬௢

 The significant wave height ‫ܪ‬௦ or ‫ܪ‬ଵ/ଷ
 The 1/10 highest have height ‫ܪ‬ଵ/ଵ଴
The Reyleigh distribution of wave heights

0.8
Height of Most Probable Wave
0.7 Height of Average Ho Wave
0.6 Height of Significant Hs Wave

Probability Density
0.5
0.4
0.3
Height of H1/10 Wave
0.2
0.1
0
0 0.5 1 1.5 2 2.5 3
Wave Height Ratio H/Hmax
Figure 56: The Reyleigh distribution of wave heights
5.4 Reference Structure
Based on experience from offshore oil and gas platforms, the jacket-type
support structure is a commonly selected configuration of subsea structures for
medium and high water depths. Current interest for such locations considering
187
massive deployment of wind turbines, sets a reference depth between 40-50 m.
Benefits of the jacket configuration compared to others, are discussed in [296]
and [297]. Selection and implementation of a jacket support structure for the
Beatrix wind farm, which has been deployed in a depth of 45 m, and stands as
the deepest location so far, countersigns its applicability for this water depth
[60].
For the application of the reliability assessment procedure that has been
developed in the previous chapters, a typical structure will be examined for the
reference depth of 50 m. The design depth in conjunction with the operational
loads, will determine the general layout of the structure, in aspects of number of
required legs, general layout and consideration of loading conditions. The depth
of interest allows a four-legged configuration located in the corners of a square
cross section at each elevation.
The structure is assumed to be constructed of tubular steel members of

common 355 MPa steel with Young modulus of 200 GPa. Each of the legs is
supported with a pile driven through the legs and extended to the seabed. The
four legs are battered to achieve better stability against toppling, with a common
bat angle with ratio 1:10. It has 5 elevations of horizontal and 4 of inclined full X
bracing. The base elevation, which is positioned on the seabed, has dimensions
of 25.0 m x 25.0 m and the structure extends above the water surface by 12 m,
resulting to a total height of 62 m.
On top of the jacket support structure, an additional load will be considered to

account for any operational loads acting on top of the support structure eg. the
loads due to machinery on the top of an oil and gas platform, including the drag
force imposed by the complicated geometry of the top side, or the aerodynamic
loads induced by the operation of a wind turbine including the drag of the
turbine tower. For this scope, an additional load will be applied on the top of the
support structure proportional to the square of wind speed. The technique of
ultra-stiff elements was employed in order to transfer the point loads to the legs,
avoiding extensive stresses and deflection to the members of the top elevation.
188
The structure has been designed with the commercial software DNV SESAM
[298], specialized for the design of offshore steel structures. This software
allows efficient modelling of environmental loads, including wave loads and
piling, providing the static and dynamic response of structural members in the
form of, bending moments, axial forces and principal stresses. The basis for the
design of this structure was selected to be API RP 2A WSD, since this is the
most conservative existing design standard, as it will be discussed in the next
chapter. Design according to the provisions of this standard, has ensured that
both buckling and ultimate strength capacity of members have been achieved.
Table 24, presents the design load input parameters for dimensioning of the
structure. For the numbering of members the following common notation has
been followed: for legs, the string ‫ܣ‬0‫ ܤ‬characterizes the member, for horizontal
braces, ‫ܣ‬1‫ܤ‬, while for vertical braces ‫ܣ‬5‫ܤ‬, where ‫ ܣ‬represents the elevation
and ‫ ܤ‬the orientation of the member.
Although selection of different loading conditions describing the wave sea state
is common in offshore structures taking into account deployment along
preferential directions, having a greater significance for larger scale and more
complicated structures, the symmetry of the reference structure allows
consideration of two directions of ‘wave attack’ (0 and 45 degrees), and
application of the same derived cross sections to the corresponding members.
Having ensured that the operational loads have been considered in the worst
case direction combined to the environmental loads, this simplification
diminishes significantly the number of load cases that need to be examined,
and in the present study will allow investigation of more parameters.
Figure 57, illustrates the structure of reference, while in Appendix C, details on

the cross sections of members are included.
189
Parameter Value Unit
Significant Wave Height 11.47 m
Design Wave Height 21.46 m
Associated Wave Period 13.3 sec
Drag Coefficient 1.05
Morison Coefficient 1.2
Wind Speed 25 m/sec
MWL: 1
Current Profile
-25 m: 0.5 m/sec
Table 24: Design load input parameters for dimensioning of the structure
Figure 57: FEA model of a jacket structure developed in DNV GeniE

software
190
5.5 Component Structural Reliability Assessment
5.5.1 Base case
The case that will stand as the basis for comparison in the later sections,
accounts for four stochastic variables that will be considered in this analysis.
Incorporation of more variables is feasible, however as it will be derived from
the analysis, fewer variables should be modelled stochastically that have a
greater effect on the structural response. For the derivation of the reliability
indices of each member, several simulations have been executed in FEA
software and the results are imported in the MATLAB code that has been
developed for the data regression and later the calculation of the FORM and
SORM reliability index. The four variables that are considered stochastically are
summarized in Table 25.
Variable Distribution Type Coefficients Units
‫ݔ‬1 Wave height Reyleigh ‫ = ܣ‬8.08 m
‫ݔ‬2 Wind Force Normal (400,40) kN
‫ݔ‬3 Current Normal (0.8,0.15) m/sec
‫ݔ‬4 Yield LogNormal (2.55,1.398) MPa
Table 25: Properties of stochastic variables
The structure has been modelled according to 5th order Stokes wave theory,
considering drag and Morison coefficients for rough members’ surfaces
(‫ܥ‬ௗ = 1.05 and ‫ܥ‬௠ = 1.2). The effect of those assumptions, as well as the
contributions of each of the variables will be examined in the later sections.
Table 26 to Table 28, summarize the results of reliability indexes of members,
for directions of 0 and 45 degrees for this base case.
191
ࢼ ࡼࢌ ࢼ ࡼࢌ
Member ID Member ID
0 degs 45 degs 0 degs 45 degs 0 degs 45 degs 0 degs 45 degs
b116 35.01 35.01 ∞ ∞ b316 20.87 9.05 4.83E-97 7.08E-20
b115 35.01 35.01 ∞ ∞ b315 20.86 14.8 6.64E-97 7.03E-50
b114 54.83 54.83 ∞ ∞ b314 6.94 7.68 1.99E-12 8.14E-15
b113 54.83 54.83 ∞ ∞ b313 10.48 7.52 5.40E-26 2.69E-14
b112 54.83 54.83 ∞ ∞ b312 6.96 7.5 1.65E-12 3.16E-14
b111 54.83 54.83 ∞ ∞ b311 10.46 7.76 6.55E-26 4.23E-15
b216 18.71 26.7 2.04E-78 2.30E-157 b416 8.37 6.02 2.84E-17 8.92E-10
b215 18.72 16.38 1.70E-78 1.43E-60 b415 8.36 9.64 3.03E-17 2.80E-22
b214 9.33 11.43 5.45E-21 1.45E-30 b414 5.76 6.91 4.24E-09 2.42E-12
b213 12.33 9.52 3.14E-35 8.36E-22 b413 8.98 6.7 1.31E-19 1.05E-11
b212 9.70 9.48 1.53E-22 1.27E-21 b412 6.00 6.72 9.61E-10 9.01E-12
b211 12.34 10.62 2.60E-35 1.14E-26 b411 8.98 7.04 1.36E-19 9.50E-13
Table 26: Reliability Index of Horizontal Members
Member ID Member ID
0 degs 45 degs 0 degs 45 degs 0 degs 45 degs 0 degs 45 degs
b158 6.74 6.08 7.67E-12 5.96E-10 b358 6.16 5.37 3.62E-10 3.95E-08
b157 6.76 6.81 6.89E-12 4.93E-12 b357 6.16 5.24 3.59E-10 8.14E-08
b156 5.7 6.77 5.95E-09 6.48E-12 b356 5.31 5.12 5.46E-08 1.56E-07
b155 5.55 5.92 1.44E-08 1.64E-09 b355 5.33 5.19 5.04E-08 1.05E-07
b154 6.94 5.93 1.99E-12 1.48E-09 b354 6.09 5.21 5.69E-10 9.37E-08
b153 6.93 6.63 2.10E-12 1.64E-11 b353 6.08 5.14 6.04E-10 1.38E-07
b152 5.54 6.78 1.50E-08 6.10E-12 b352 5.33 5.2 5.02E-08 9.75E-08
b151 5.7 6.05 5.86E-09 7.08E-10 b351 5.3 5.35 5.65E-08 4.44E-08
b258 7.79 5.79 3.28E-15 3.62E-09 b458 3.72 4.48 9.86E-05 3.79E-06
b257 7.81 5.88 2.89E-15 2.06E-09 b457 3.72 4.48 9.99E-05 3.69E-06
b256 5.27 5.68 6.84E-08 6.58E-09 b456 6.37 4.35 9.53E-11 6.85E-06
b255 5.26 5.66 7.26E-08 7.37E-09 b455 6.34 4.64 1.18E-10 1.73E-06
b254 7.51 5.69 3.01E-14 6.38E-09 b454 3.79 4.69 7.65E-05 1.35E-06
b253 7.51 5.71 2.87E-14 5.70E-09 b453 3.79 4.32 7.57E-05 7.74E-06
b252 5.25 5.85 7.71E-08 2.47E-09 b452 6.33 4.45 1.22E-10 4.23E-06
b251 5.27 5.74 6.93E-08 4.78E-09 b451 6.37 4.51 9.75E-11 3.20E-06
Table 27: Reliability Index of Vertical Members
192
ࢼ ࡼࢌ
Member
ID 0 degs 45 degs 0 degs 45 degs
b404 14.72 10.82 2.32E-49 1.40E-27

b403 10.66 10.20 7.76E-27 1.03E-24
b402 10.66 10.88 7.80E-27 7.17E-28
b401 14.73 19.52 2.13E-49 4.03E-85
b304 8.72 13.87 1.44E-18 4.54E-44
b303 8.14 7.36 1.99E-16 9.24E-14
b302 8.15 13.83 1.83E-16 7.84E-44
b301 8.71 7.23 1.54E-18 2.37E-13
b204 5.74 19.27 4.78E-09 4.54E-83
b203 7.45 6.26 4.52E-14 1.95E-10
b202 7.46 19.96 4.39E-14 5.78E-89
b201 5.74 5.72 4.76E-09 5.29E-09
b104 4.79 10.07 8.45E-07 3.93E-24
b103 5.23 4.50 8.62E-08 3.41E-06
b102 5.22 10.13 8.74E-08 2.09E-24
b101 4.79 4.66 8.41E-07 1.59E-06
Table 28: Reliability Index of Legs Members
From the results presented above it can be observed that reasonable values
are derived for the reliability indices of the members. The large values that
occur in members b111-b116 and refer to the horizontal members of the
seabed elevation are due to the stiff piling conditions that have been selected in
this analysis that result them to carry very small loads. In general, horizontal X
brace elements are designed to carry small loads, but should be included in the
design for practical reasons.
Variation in the direction of loading shows the worse case that should be
considered in the minimum reliability that will eventually size the members. For
the example of the legs, this fact is more obvious since in the 0 degrees of
approach there is symmetry in loading for two sets of legs while in 45 degrees,
two legs have symmetrically equal reliability, while the other two have different
values referring to the different loads acting on them.
As far as the legs reliability index of different parts along its length is concerned,
due to the uniform cross section that has been selected it is observed that
reliability index values follow the cumulative load distributed along each of the
193
elevation of the structures; this indicates that members at lower elevations
which suffer from greater stresses, will have lower values of reliabilities
(ߚ = 4.79 − 14.73 for 0 degrees and ߚ = 4.50 − 10.20 for 45 degrees).
Smaller values that have can noticed on the top vertical X-braces of the
structure (members b457-b458, ߚ = 3.72), are due to the fact that the topside of
the structure has not been analytically designed so those members locally
suffer from increased loads that carry the loads of the topside. In a more
realistic structure this phenomenon would have been avoided by appropriate
design of the topside-structure interaction with adequate load paths that would
transform those loads more uniformly to the rest of the structure.
Small deviations on the reliability indexes of symmetrical members in 45

degrees case is due to the fact that wind load is acting in a constant direction (0
degrees) compared to the different directions of the wave and current loads
slightly “disturbing” this symmetry. However, the small scale of this deviation
illustrates that the effect of the wind load is minimum compared to that of the
sea state loads. Further, it should be essential to point out that the large values
that are presented on the above tables refer only to two cases of loading where
some of the members suffer more than others while considering the different
directional load cases other members would be in this place. This means that in
order to derive the total reliability index of each member, every possible
direction of loading should be indentified and examined, and the final reliability
of each member would equal the minimum of the values that have been partially
calculated. For the case that has been examined, and due to the assumption
that the same loading conditions act on the structure every 90 degrees, the final
values of reliability indices can be derived based on the geometrical symmetry
of the structure. Table 29, summarizes those results incorporating 8 different,
but symmetrical, directions acting on the structure.
194
Member Member Member Member
࢓ ࢏࢔(ࢼ) ࢓ ࢏࢔(ࢼ) ࢓ ࢏࢔(ࢼ) ࢓ ࢏࢔(ࢼ)
ID ID ID ID
b116 35.01 b416 6.02 b104 4.50 b252 5.25
b115 35.01 b415 6.02 b103 4.50 b251 5.25
b114 54.83 b414 5.76 b102 4.50 b358 5.12
b113 54.83 b413 5.76 b101 4.50 b357 5.12
b112 54.83 b412 5.76 b158 5.54 b356 5.12
b111 54.83 b411 5.76 b157 5.54 b355 5.12
b216 16.38 b404 10.20 b156 5.54 b354 5.12
b215 16.38 b403 10.20 b155 5.54 b353 5.12
b214 9.33 b402 10.20 b154 5.54 b352 5.12
b213 9.33 b401 10.20 b153 5.54 b351 5.12
b212 9.33 b304 7.23 b152 5.54 b458 3.72
b211 9.33 b303 7.23 b151 5.54 b457 3.72
b316 9.05 b302 7.23 b258 5.25 b456 3.72
b315 9.05 b301 7.23 b257 5.25 b455 3.72
b314 6.94 b204 5.72 b256 5.25 b454 3.72
b313 6.94 b203 5.72 b255 5.25 b453 3.72
b312 6.94 b202 5.72 b254 5.25 b452 3.72
b311 6.94 b201 5.72 b253 5.25 b451 3.72
Table 29: Minimum Reliability index of members, incorporating 8 different

directions
Further to the above analysis for ultimate strength of members, reliability

assessment for those under compression for buckling was performed. During
the analysis, two different values of the parameter ‫ ܥ‬in the critical buckling force
calculation were considered, 1.2 (recommended value) and 4 (theoretical value)
as it can be found in [278], in order to evaluate the effect of this parameter on
the derived values of reliability indices. Results of this analysis for the 10 more
critical members are presented in Table 30 which corresponds to vertical
diagonal bracing members. From those values the conclusion that can be
drawn is that coefficient ‫ ܥ‬plays a significant role to the outcome of this analysis
affecting significantly the value of ߚ.
195
Member ࡯=૝ ࡯ = ૚. ૛
ID 0 degs 45 degs 0 degs 45 degs
b155 7.58 8.44 3.55 4.02
b152 7.58 10.82 3.55 5.25
b255 8.09 9.51 3.84 4.63
b252 8.09 10.61 3.85 5.17
b154 17.38 8.45 9.00 4.05
b355 9.04 10.72 4.45 5.38
b352 9.05 11.54 4.45 5.77
b254 28.32 9.53 15.10 4.68
b258 25.77 9.70 13.76 4.84
b257 25.74 10.65 13.76 5.25
Table 30: Reliability indices for buckling limit states (10 critical members)
5.5.2 Sensitivity analysis of design parameters
In this section, the effect of design parameters in the resultant values of

reliability index will be investigated. Surface roughness, expressed through
appropriate consideration of ‫ܥ‬ௗ and ‫ܥ‬௠ coefficients, selection of wave modelling
theory, effect of different statistical distributions and variation in stochastic loads
properties are studied as the main model uncertainties; further sources of model
uncertainty will not be investigated as they stand out of the scope of this
contribution. A further analysis regarding the effect of corrosion modelling to the
resulting values of ߚ will show the reliability deterioration of the structure
throughout its service life. For each case, the results with the major effect of the
parameter examined will be illustrated.
5.5.2.1 Wave modelling
5.5.2.1.1 Effect of surface roughness
The effect of surface roughness of structural members will be examined based

on provisions of standards for drag and inertial coefficients. Based on the fact
that the base case accounts for rough cylinders (‫ = ܦܥ‬1.05 and ‫ܥ‬ெ = 1.2)
according to API’s provisions, smooth surface cylinders should be modelled
with appropriate coefficients (‫ = ܦܥ‬0.65 and ‫ܥ‬ெ = 1.6). It is expected that
members of greater surface roughness will generate greater drag force and
therefore their relevant reliability index would be higher. Table 31 summarizes
196
the results of ߚ and the related difference to the base case for the 30 members
with the greatest effect for 0 degrees.
Relative Relative
Member error Member
ࢼ࢘࢕࢛ࢍࢎ ࢼ࢙࢓ ࢕࢕࢚ࢎ ࢼ࢘࢕࢛ࢍࢎ ࢼ࢙࢓ ࢕࢕࢚ࢎ error
ID ID
(%) (%)
b415 8.36 15.05 80.03 b257 7.81 10.65 36.32
b416 8.37 15.06 79.98 b201 5.74 7.81 36.11
b301 8.71 13.62 56.33 b204 5.74 7.81 36.09
b304 8.72 13.63 56.29 b151 5.70 7.75 36.00
b158 6.74 9.61 42.62 b156 5.70 7.75 35.93
b157 6.76 9.63 42.53 b302 8.15 10.96 34.48
b404 14.72 20.54 39.54 b303 8.14 10.95 34.48
b401 14.73 20.55 39.50 b212 9.70 13.03 34.36
b357 6.16 8.56 39.02 b403 10.66 14.27 33.90
b358 6.16 8.56 38.99 b402 10.66 14.27 33.90
b458 3.72 5.13 37.80 b414 5.76 7.63 32.52
b457 3.72 5.12 37.67 b312 6.96 9.16 31.66
b152 5.54 7.56 36.55 b256 5.27 6.91 31.06
b155 5.55 7.58 36.49 b251 5.27 6.90 31.00
b258 7.79 10.62 36.39 b352 5.33 6.97 30.80
Table 31: Reliability index of smooth and rough cylinders (30 critical
members)
The increase of reliability index is greater for members that have already
greater values of ߚ in both cases. Further, due to the increase to the distributed
loads along the height of the structure, the resultant load acting on the structure
with the rougher members will be greater on the members of the lower
elevation, leading to greater difference of reliability compared to a structure with
smoother components. This phenomenon, illustrates that as the structure ages,
and the smoothness of the members’ surfaces change due to corrosion, marine
growth etc, the effect of environmental loads acting on the structure becomes
more significant. This verifies the assumption presented in theory regarding
deterioration of reliability due to decrease in the relative distance between load
effect and resistance of the structure.
5.5.2.1.2 Different wave theories
Selection of the appropriate wave modelling theory is another decision that has
a crucial impact on the accuracy of the prediction of the performance of the
197
structure for given input conditions. Increased complexity of more analytical
methods may raise restriction to engineers, especially when appropriate
software for modelling is not available; however, this fact might under or
overestimate the actual performance of the structure. The effect of this
inaccuracy is therefore transferred in the reliability calculation misjudging the
actual performance of members. The base case has been modelled using 5th
order Stokes equations model. Table 32, presents the results of the 20
members with the most significant deviation when modelled with different wave
modelling theories, for the case of 0 degrees of wave angle. For this
comparison, Airy wave theory and Stream function theory (3rd and 11th order)
have been considered.
th
ࢼ Deviation from Stokes 5 order
th
Airy Stream Stream
Stokes 5 Stream Stream Theory function function
Member Airy
Order function function rd
(3 order)
rd
(11 order)
ID Theory rd rd
Theory (3 order) (11 order)
(%) (%) (%)
b301 8.71 14.16 8.73 8.73 62.5 0.3 0.2

b304 8.72 14.17 8.74 8.74 62.5 0.2 0.2
b203 7.45 11.23 7.55 7.54 50.8 1.3 1.3
b202 7.46 11.24 7.55 7.55 50.6 1.2 1.2
b302 8.15 12.07 8.20 8.19 48.1 0.6 0.5
b303 8.14 12.05 8.19 8.18 48.0 0.6 0.5
b452 6.33 8.89 6.38 6.38 40.5 0.8 0.8
b455 6.34 8.90 6.39 6.39 40.4 0.7 0.7
b456 6.37 8.81 6.43 6.43 38.3 0.9 0.9
b451 6.37 8.80 6.42 6.42 38.2 0.9 0.8
b102 5.22 7.05 5.29 5.29 35.1 1.3 1.3
b103 5.23 7.05 5.29 5.29 34.9 1.1 1.1
b453 3.79 5.00 3.82 3.82 31.8 0.9 0.9
b454 3.79 4.99 3.82 3.82 31.7 0.8 0.8
b458 3.72 4.79 3.75 3.74 28.9 0.7 0.6
b457 3.72 4.79 3.74 3.74 28.8 0.6 0.5
b403 10.66 13.71 10.77 10.78 28.6 1.0 1.1
b402 10.66 13.71 10.77 10.77 28.6 1.0 1.1
b352 5.33 6.81 5.38 5.38 27.7 1.0 1.0
b355 5.33 6.81 5.38 5.38 27.7 1.0 1.0
Table 32: Reliability index for different wave theories (20 members with
critical effect)
198
From the figures above, the conclusion that can be derived is that compared to
the 5th order Stokes wave theory, Airy wave theory tend to underestimate the
environmental loads, resulting to greater values of reliability indices. Greater
deviation is found to occur in the mid elevation members. Stream function
produces similar results than the base case with limited deviation (up to 4.3 %).
The order of the stream function does not seem to provide a significant
difference on the values of reliability index for the case studied. Finally from the
sea state conditions that have been selected and the structure that has been
examined, the results provided from the 5th order Stokes wave theory, which
compiles to API’s provisions for given environment, provide the most
conservative results. This comparison illustrates the significance of the selection
of the appropriate wave theory.
5.5.2.2 Effect of statistical distributions
Appropriate modelling of the stochastic variables is another issue that should be

treated efficiently in the pre-processing of the reliability assessment. In the
literature review of this Thesis, different statistical distributions have been
proposed for different loads and design parameters. In the base case, relevant
distributions have been selected based on the nature of the variables. In this
section, the values of ߚ that have been derived for the base case will be
compared to values obtained by a new calculation of reliability derived by
considering equivalent normal distributions for all of the variables modeled
stochastically. Table 33 presents the properties of those equivalent normal
distributions while Table 34 provides comparative results of the 20 members
with greater deviations on the results (0 degrees).
Variable Coefficients Units

‫ݔ‬1 Wave height (8.799,4.5932) m
‫ݔ‬2 Wind Force (400,40) kN
‫ݔ‬3 Current (0.8,0.15) m/sec
‫ݔ‬4 Yield (355,25) MPa
Table 33: Parameters of equivalent normal distributions
199
Member Deviation Member Deviation
ࢼ࢈ࢇ࢙ࢋ ࢼ࢔࢕࢘࢓ ࢇ࢒ ࢼ࢈ࢇ࢙ࢋ ࢼ࢔࢕࢘࢓ ࢇ࢒
ID (%) ID (%)
b101 4.79 5.19 8.4 b304 8.72 9.13 4.7
b104 4.79 5.19 8.4 b258 7.79 8.15 4.6
b201 5.74 6.20 8.0 b212 9.70 10.14 4.6
b204 5.74 6.20 8.0 b303 8.14 8.51 4.6
b415 8.36 8.96 7.2 b411 8.98 9.39 4.5
b416 8.37 8.96 7.1 b413 8.98 9.39 4.5
b404 14.72 15.49 5.2 b313 10.48 10.95 4.5
b401 14.73 15.49 5.2 b158 6.74 7.04 4.5
b402 10.66 11.19 5.0 b315 20.86 21.79 4.5
b403 10.66 11.19 5.0 b302 8.15 8.51 4.4
Table 34: Reliability indices for equivalent normal distributions
From the above results, it can be concluded that consideration of the equivalent
normal variables overestimates values of reliability indices. This results in less
conservative approximation of ߚ. In the list provided, greater deviation occurs in
the lower elevation legs members which are the most significant members in
the design of a jacket structure. The maximum of 8.4 % in the value of ߚ
highlights the significance of appropriate statistical modeling of the stochastic
variables.
5.5.2.3 Stochastic variables variation
5.5.2.3.1 Loading variables
Each of the variables that are considered stochastically in the reliability

assessment of a structure, has a different contribution in the resultant value of
ߚ. This section presents the comparison of the base case, with three different
simulations of cases where the three environmental loads, wave, wind and
current, have been decreased by 25 %. The effect on ߚs are presented in Table
35, for the case of the 30 more critical members (lower values of ߚ in the base
case) and wave angle of 0 degrees.
200
Member Member
ࢼ࢈ࢇ࢙ࢋ ࢼ૙.ૠ૞×࢝ ࢇ࢜ࢋ ࢼ૙.ૠ૞×࢝ ࢏࢔ࢊ ࢼ૙.ૠ૞×ࢉ࢛࢘࢘ ࢼ࢈ࢇ࢙ࢋ ࢼ૙.ૠ૞×࢝ ࢇ࢜ࢋ ࢼ૙.ૠ૞×࢝ ࢏࢔ࢊ ࢼ૙.ૠ૞×ࢉ࢛࢘࢘
ID ID
b457 3.72 5.52 3.72 3.76 b352 5.33 7.67 5.33 5.35
b458 3.72 5.53 3.72 3.75 b355 5.33 7.66 5.33 5.35
b453 3.79 5.62 3.79 3.82 b152 5.54 7.96 5.55 5.57
b454 3.79 5.61 3.78 3.82 b155 5.55 7.96 5.56 5.57
b101 4.79 5.81 4.79 4.79 b151 5.70 8.18 5.71 5.71
b104 4.79 5.82 4.79 4.80 b156 5.70 8.18 5.71 5.73
b102 5.22 7.52 5.24 5.24 b201 5.74 7.09 5.73 5.74
b103 5.23 7.53 5.23 5.24 b204 5.74 7.08 5.74 5.74
b252 5.25 7.57 5.26 5.28 b414 5.76 8.23 5.75 5.80
b255 5.26 7.56 5.25 5.28 b412 6.00 8.56 6.00 6.05
b251 5.27 7.58 5.27 5.29 b353 6.08 8.68 6.09 6.13
b256 5.27 7.58 5.27 5.30 b354 6.09 8.67 6.09 6.13
b351 5.30 7.64 5.31 5.34 b357 6.16 8.77 6.16 6.20
b356 5.31 7.64 5.32 5.34 b358 6.16 8.78 6.16 6.21
b352 5.33 7.67 5.33 5.35 b452 6.33 9.01 6.34 6.36
b355 5.33 7.66 5.33 5.35 b455 6.34 9.01 6.34 6.36
Table 35: Reliability indices for cases of 25% reduced loads
The above results show that the wave loading has the most significant effect on
the structure. A decrease in this load decreases crucially the loads acting on the
structure and therefore increases the reliability indices of the members. This
deviation increases to the members with higher reliability in the base case. The
effect of wind and current is less important on the structure, verifying that the
assumption made for symmetrical wave load sets every 90 degrees, is not
influenced significantly by the different direction of the wind loads.
A useful note in this comparison is that the code that has been developed
incorporating FORM and SORM can provide sensitivity factors of each of the
variables, which illustrate the relative contribution of each to the calculated
reliability index.
5.5.2.3.2 Material Yield
Different grades of steel, with different yield strength influence the reliability of
the members since it directly forms the utilization ratio (ratio of actual stress to
yield) that will later form the response surface of the structure. In this
comparison, the S355 steel that has been used in the base case simulation is
201
compared to a steel of lower grade, S275, and the results are presented in
Table 36.
Member Member Member

ࢼ૜૞૞ ࢼ૛ૠ૞ ࢼ૜૞૞ ࢼ૛ૠ૞ ࢼ૜૞૞ ࢼ૛ૠ૞
ID ID ID
b101 4.79 4.11 b213 12.33 10.43 b353 6.08 5.06
b102 5.22 4.35 b214 9.33 7.94 b354 6.09 5.05
b103 5.23 4.36 b215 18.72 15.56 b355 5.33 4.43
b104 4.79 4.11 b216 18.71 15.53 b356 5.31 4.44
b111 54.83 36.36 b251 5.27 4.43 b357 6.16 5.07
b112 54.83 36.36 b252 5.25 4.38 b358 6.16 5.07
b113 54.83 36.36 b253 7.51 6.36 b401 14.73 12.43
b114 54.83 36.36 b254 7.51 6.37 b402 10.66 9.12
b115 35.01 22.57 b255 5.26 4.38 b403 10.66 9.13
b116 35.01 22.57 b256 5.27 4.43 b404 14.72 12.44
b151 5.70 4.82 b257 7.81 6.62 b411 8.98 7.57
b152 5.54 4.68 b258 7.79 6.63 b412 6.00 4.94
b153 6.93 5.91 b301 8.71 7.50 b413 8.98 7.57
b154 6.94 5.92 b302 8.15 6.96 b414 5.76 4.75
b155 5.55 4.69 b303 8.14 6.96 b415 8.36 7.10
b156 5.70 4.82 b304 8.72 7.50 b416 8.37 7.09
b157 6.76 5.80 b311 10.46 8.83 b451 6.37 5.38
b158 6.74 5.81 b312 6.96 5.92 b452 6.33 5.35
b201 5.74 4.91 b313 10.48 8.84 b453 3.79 3.13
b202 7.46 6.30 b314 6.94 5.87 b454 3.79 3.13
b203 7.45 6.30 b315 20.86 17.49 b455 6.34 5.35
b204 5.74 4.90 b316 20.87 17.50 b456 6.37 5.38
b211 12.34 10.41 b351 5.30 4.45 b457 3.72 3.05
b212 9.70 8.25 b352 5.33 4.43 b458 3.72 3.05
Table 36: Reliability indices for S355 and S275 steels
The results that have been derived illustrate as it was expected, that the steel of
lower yield strength leads to lower values of target reliability. This deviation is
greater in members of higher reliability. For the critical members (b457-b458)
minimum derived values of ߚ become very low form S275 steels, getting out of
the acceptable limits set by standards, as it has been discussed earlier in this
Thesis.
202
5.5.2.4 Corrosion models
Corrosion is one of the most important phenomena related to capacity

deterioration of the structure. Different models for this time-dependent
phenomenon have already been presented and express this problem as a time-
independent one through a relative decrease in the thickness of members. In
this section, different models have been considered in the reliability assessment
and the corresponding reliability deterioration throughout the structure’s service
life is presented. In Figure 58, the cumulative relative degradation in members’
thickness is expressed based on the models examined. For the models where
thickness decrease is expressed through statistical distributions (mean value
and standard deviation) the thickness value that corresponds to probability of
occurrence of 95% is considered. From the results that have been collected,
Figure 60 (i-iv) presents graphs that illustrate the degradation of the reliability
index according to each method for 4 members of one leg, as they represent
values of reliability index of different range. Table 37, shows the degradation of
reliability between 0 and 20 years, since this is the most common prescribed life
of offshore structures, for the 20 more critical members according to each
method.
Thickness Deterioration - Corrosion models

6
linear 0.5
linear 1.0
5 linear 1.5
Southwell Linear
Southwell extended
4 Melchers-Southwell
dt (mm)
0
0 5 10 15 20 25 30 35 40
years
Figure 58: Thickness deterioration as a function of time
203
Member Base Linear Linear Linear Southwell Southwell Melchers-
ID case 0.5 1.0 1.5 Linear Extended Southwell
b457 3.72 3.59 3.47 3.33 3.62 3.50 3.46

b458 3.72 3.60 3.47 3.33 3.62 3.50 3.46
b453 3.79 3.66 3.54 3.40 3.69 3.57 3.52
b454 3.79 3.66 3.53 3.40 3.69 3.57 3.53
b101 4.79 4.75 4.70 4.72 4.73 4.70 4.70
b104 4.79 4.75 4.70 4.72 4.73 4.70 4.70
b102 5.22 5.11 4.99 4.85 5.13 5.02 4.98
b103 5.23 5.11 4.99 4.86 5.13 5.02 4.98
b252 5.25 5.08 4.93 4.76 5.13 4.98 4.92
b255 5.26 5.09 4.94 4.77 5.14 4.99 4.92
b251 5.27 5.11 4.96 4.79 5.15 5.01 4.95
b256 5.27 5.11 4.97 4.79 5.16 5.01 4.95
b351 5.30 5.15 5.01 4.83 5.20 5.05 4.99
b356 5.31 5.15 5.01 4.84 5.20 5.05 4.99
b352 5.33 5.15 5.00 4.81 5.21 5.05 4.98
b355 5.33 5.15 5.00 4.81 5.21 5.04 4.99
b152 5.54 5.35 5.18 4.99 5.40 5.23 5.18
b155 5.55 5.36 5.19 5.00 5.41 5.24 5.17
b151 5.70 5.51 5.32 5.11 5.55 5.37 5.31
b156 5.70 5.50 5.32 5.11 5.55 5.37 5.31
Table 37: Reliability indices for 20 years for different corrosion models
Reliability index deteriorationm Member 62 Reliability index deteriorationm Member 66

3.8 3.8
3.7 3.7
3.6
3.6
3.5
3.5
3.4
reliability index
reliability index
3.4
3.3
3.3
3.2
linear 0.5 linear 0.5
3.2
linear 1.0 3.1 linear 1.0
3.1
Southwell Linear 3 Southwell Linear
Southwell extended Southwell extended
3 2.9
Melchers-Southwell Melchers-Southwell
2.8
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
years years
Figure 59: Reliability index deterioration of critical members (b454, b458)
204
11 8.2
10.8 linear 1.0 linear 1.0
linear 1.5 8 linear 1.5
Southwell Linear Southwell Linear
10.6
Melchers-Southwell 7.8 Melchers-Southwell
10.4
reliability index
reliability index
10.2 7.6
10
7.4
9.8
7.2
9.6
9.4 7
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
years years

7.5 5.4
7.4 linear 1.0 5.3 linear 1.0
7.3 5.2
Southwell Linear Southwell Linear
7.2 5.1
Melchers-Southwell Melchers-Southwell
7.1 5
reliability index
reliability index
7 4.9
6.9 4.8
6.8 4.7
6.7 4.6
6.6 4.5
6.5 4.4
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
years years
Figure 60: Reliability index deterioration of members (b403-b103)
From the simulations that have been executed and the results that are
presented above, it can be pointed out that the effect of corrosion is significant
in the reliability of the structure since its load bearing capacity is reduced.
Selection of the appropriate corrosion model is essential since comparing
different models it can double the effect of corrosion throughout its service life.
This decision should be taken according to the type of the structure and the
environment of deployment. The effect in reliability deterioration is more
significant for thinner members and in members with greater values of reliability
index. The fact that for specific members the deterioration rate changes, even
implying small scale increase of reliability as time progresses, identifies load
redistribution through alternative load paths of the structure, with different
members suffering from greater loads than others as the thickness of members
decreases uniformly.
205
For critical members, as it can be observed in Figure 59, appropriate modelling
of the structures can classify them within an acceptable region of reliability or in
an unsafe region yielding intervention when in operational phase or re-
dimensioning when in the design phase. Further, in a more accurate
assessment, different parts of the structure at different elevations shall be
modelled according to different corrosion models.
The above analysis has considered general (wastage) corrosion on the

structural members that accounts for wall thickness reduction and is considered
by appropriate models. Additionally, pitting corrosion may occur locally on
members leading to stress concentration effects. Wall thinning due to corrosion
(of any type, including corrosion fatigue), can be largely mitigated against by
use of an appropriate corrosion protection system, as it has been referred to
earlier and is presented in Appendix B of the Thesis.
5.5.3 System Reliability Integration
Although the main interest of this Thesis is the estimation of reliability in a

component level since this is the approach followed by design standards, in this
section the effects of uncertainties on the structure as a system will be
examined following the background theory that has been presented earlier.
For the purpose of this study, pushover analysis has been selected among
other methods for the identification of the most dominant failure path of the
structure. This analysis identifies in a deterministic way the most critical
members of the structure forming a failure path that will derive the resultant
global reliability of the structure. Available studies, verify that for cases of
extreme loading conditions, the level of reliability of the identified failure path
based on a pushover analysis is in accordance to values that may be obtained
using extensive methods of search and simulations [299], [300].
Design simulation was combined with pushover analysis to address the stress
variation on the structure and identify the sequence of failure of structural
members as well as the global response of the structure. The difficulty of this
206
approach is that due to the simulations required, in structures of high
complexity, the computational effort required increases; therefore it is not
indicated for failure paths with large number of members. In [301] and [302],
guidelines for simple methods of failure paths identification can be found.
An important assumption following this approach is that global failure occurs at

one instance, for example in the case when the lateral wave load reaches its
maximum, and all of the members of a failure paths fail at the same time.
Failures are considered to occur over a short period of time during which the
load is applied proportionally. This assumption transforms the problem to a time
independent one, without demanding recalculation of reliability of the members
of the damaged structure.
Once a member’s failure is identified, its stiffness and therefore the global
stiffens of the structure is modified by modelling a residual strength either by
applying appropriate forces at the failed member or by changing its structural
properties. Figure 61, presents the post-failure behaviour that has been
considered in this analysis based on [86]. A new structural stress calculation is
initiated and this loop continues until a successive sequence of members’
failures is identified. The global response of the structure will illustrate the step
where the global failure is considered to occur through criteria of extensive non
linear deformation or displacements. In order to proportionally increase loads,
the design case is selected as a base case and the loads are factored until the
structure collapses.
207
C = Compression Capacity
Force
0.4C
Deformation
T = Tension Capacity
Figure 61: Post failure behaviour
For the reference structure that is examined, the resultant failure path,
consisting of four different failure branches, 90 degrees apart for each load
direction, that are been identified are presented in Figure 62Error! Reference
source not found.Error! Reference source not found.. In this graph, each of the
branches is modelled as a sequence of parallel events, where they should all
fail in order for the branch to fail, while branches are linked together in a series
formation, indicating the once the weakest link has failed the whole structure
fails. Graphs of displacement as a function of overturning moment and load
factor are presented in Figure 63 and Figure 64.
b457 b458 b453 b414 b103
b451 b452 b455 b411 b104
b453 b454 b457 b412 b101
b455 b456 b451 b413 b102
Figure 62: Critical Failure paths
208
Diagram of intact structure
550
500
450
400
Base OTM (MNm)
350
300
250
200
150
100
50
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22
displacement (m)
Figure 63: Displacement vs overturning moment (intact structure)
Diagram of intact structure

4
3.5
3
Load Factor
2.5
1.5
0.5
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22
displacement (m)
Figure 64: Displacement vs load factor (intact structure)
From the first failure path identified, failures of members 457-103 refer to the
effective path of the structure. Failure of members 103 and 102, which refer to
the legs of the lower elevation of the structure has the most significant effect on
the global response of the structure since upon their failure, the structure stops
209
to deform linearly. For the case of 45 degrees, the first member to fail is 103,
resulting to nonlinear displacement following the first failure of the structure,
indicating absence of redundancy in this direction.
Calculation of the reserve strength ratio as the ratio of the base overturning
moment of the ultimate resistance of the structure over that of the design load
case derives a value of 3.7. The corresponding value for the case of 45 degrees
is 3.1.
Further analysis has been executed for the case of damaged structure, by
completely removing structural members, simulating the case when the two
most likely to fail members are sequentially removed. The diagrams that
correspond to those cases are presented in Figure 65 to Figure 68 presenting
displacement as a function of overturning base moment and load factor.
From the results obtained, the capacity of the damaged structure can be
determined and the residual strength can be expressed by calculation of the
residual resistance factor (RIF) and the damage tolerance ratio (DTR). The RIF
for the first case where the member b457 is removed is calculated as 0.96 while
for the second case where additionally member b458 is removed, RIF is found
to be 0.92. Those results show that for both damage scenarios, removal of
members reduce the maximum environmental load the structure might resist.
The DTR calculated for the first case is 0.92 while for the second case 0.88.
Those numbers indicate the reduction in the reserve strength of the structure.
The values of the DTR obtained, indicate that the platform has a high tolerance
to damages.
210
Diagram of damaged structure (1 member removed)
550
500
450
400
Base OTM (MNm)
350
300
250
200
150
100
50
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22
displacement (m)
Figure 65: Displacement vs overturning moment (one member removed)
Diagram of damaged structure (2 members removed)

550
500
450
400
Base OTM (MNm)
350
300
250
200
150
100
50
0 0.05 0.1 0.15 0.2 0.25
displacement (m)
Figure 66: Displacement vs overturning moment (two members removed)
211
Diagram of damaged structure (1 member removed)
4
3.5
3
Load Factor
2.5
1.5
0.5
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22
displacement (m)
Figure 67: Displacement vs load factor (one member removed)
Diagram of damaged structure (2 members removed)

4
3.5
3
Load Factor
2.5
1.5
0.5
0 0.05 0.1 0.15 0.2 0.25
displacement (m)
Figure 68: Displacement vs load factor (two members removed)
For a rough estimation of system reliability a simple bound approach is applied

based on the assumptions stated earlier. Although this approach provides a
wide range of the value of the system reliability it can derive an initial estimate
of the system integrated reliability. Due to the fact that symmetric loads are
212
considered to act on the structure, the total reliability of the structure will equal
that of anyone of the branches. For the calculation of the reliability of each
branch, the laws of parallel systems are applied.
The values that have been derived both for the reserve strength ratio, residual
resistance factor and damage tolerance ratio are reasonable and close to ones
found in similar studies [92]. The value of the lower and upper bounds of the
system reliability that have been calculated are presented in Table 38.
Lower Bound Upper Bound
10.97 2.66E-28 5.76 4.21E-09
Table 38: Bounds of System Reliability
The analysis that has been executed quantifies the level of system reliability of
a typical jacket structure based on a simplified method. More accurate
approximation of the index can be obtained through a more accurate method
such as the Hobhenbicler approximation. The reliability index that has been
calculated from the system of failure paths equals 5.76. This result, that equals
the minimum reliability index of the members included in a failure path, is based
on a new un-corroded platform. Capacity deterioration due to corrosion, weld
imperfections, marine growth etc, will change the response of the structure and
in some cases may create different load and failure paths; therefore an
individual analysis should be executed varying members’ thicknesses and
surface properties in order to reflect the structure’s deteriorated conditions.
5.6 A Note on the dynamic loading of structures
Although the main aim of this PhD has been the investigation of the response of
the response of complex offshore structures due to static or quasi-static loads,
dynamic loads might have significant effect on structures and therefore a
separate complementary analysis should then be considered. This effect is due
213
to the inability of the structures to respond instantly to the loading applied. The
increase of the effect in the response of a structure under dynamic load can be
given by a Dynamic Amplification Factor (DAF) which will magnify stresses or
displacements derived from static analysis to represent the corresponding
values for dynamic loads:
‫ݑ‬௠ ௔௫
‫= ܨܣܦ‬ (5-33)
‫ݑ‬௦௧௔௧௜௖
The Dynamic Amplification Factor is practically a function of the geometry of the

structure and the material properties, and corresponds to its ability to dissipate
energy [303]. In [304], values for DAF are provided within a range of 1.10 − 1.30
based on a classification according to the scale of the structure, while in [305]
values between 1.5 − 2.0 are given for wind turbine structures that are
subjected to severe dynamic loads; the later values are considered as the most
conservative ones [306].
Another effect of dynamic loads is the corrosion fatigue phenomenon. This

refers to the mechanical deterioration of a material under combination of
corrosion and cyclic loading. This phenomenon can be initiated by pitting,
facilitating crack initiation and has a more significant effect to steels of higher
grades, constituting a constraint on their extensive application. Fatigue
corrosion can be avoided by addition of appropriate alloy or surface treatment of
the members.
In those cases where cyclic loads are significant, direct application of the
amplification factor to static loads obtained by ultimate strength do not produce
accurate results; therefore a fatigue limit state should be employed as it can be
found in literature [307], [308] and [309]. Further presentation of fatigue limit
states is beyond the scope of this Thesis.
214
5.7 Summary
This Chapter has presented an application of the Stochastic Response Surface

Method for the reliability assessment of ultimate strength for a typical offshore
jacket structure, deployed in a hypothetical site in the North Sea. Basis of the
analysis have been analytical limit states that have been derived, based on
failure criteria of structural mechanics. An application of the joint probability
distribution, which has been presented in Chapter 5, is included for the
reference site. For a reference structure, the reliability of structural members
has been executed, based on the theory and data, and using the computational
tools that have been presented in the previous chapters considering four
stochastic variables (wave, wind, current, material yield). Using a consistent
methodology, a sensitivity analysis of the effect of different design parameters
of the design, such as the angle of loads, buckling coefficients, statistical
distribution and relative effect of variables has been presented. After an
extensive investigation of different corrosion models and their effect on the
estimated reliability indices, integration from a local to system reliability has
been realized illustrating the structure’s potential to redundancy.
215
6 COMPARISON OF THE ULTIMATE STRENGTH
RELIABILITY PERFORMANCE OF RELEVANT
DESIGN PROCEDURES
6.1 Introduction
In this Chapter, the reliability assessment of offshore jacket structures will be

investigated based on limit states formulated by design standards’
requirements. The basic clauses of each standard have been included in this
Chapter. For the four standards that reference has been made in the first
chapters of this Thesis, API LRFD [2], ISO 19902 [3], EN 1993 [5], AISC/ANSI
[4], the applicable design provisions will be initially presented, resulting to the
formulation of corresponding limit state functions. For the definition of loads and
load combinations, load factors on both loads and material properties have
been considered to equal unity, while reduction (resistance) factors for the
strength of members under different loading modes have been selected
deterministically following standards’ recommendations.
Once this background is set, reliability indices for the members of the reference
structure that was described in the previous Chapter will be calculated based on
the new limit states. This study will illustrate the level of conservatism of
standards, comparing obtained results to the ones that have been derived by
the limit state that was based on the von Mises failure criterion in the previous
Chapter. This study will focus on the component reliability assessment since
this is the approach that design standards adopt; however following the
procedure that was described and applied in the last section of the previous
chapter, integration to a systemic level can be achieved.
217
6.2 API RP-2A: Recommended practice for planning, designing
and constructing fixed offshore platforms LRFD
This section, will present the design procedure of tubular members subjected to
tension, compression, bending, shear, hydrodynamic pressure, or combined
action of the above loads, as included in Chapter D of the standard. Ultimate
strength and stability criteria are derived. The recommendations provided, are
considered to be applicable for stiffened or un-stiffened members with thickness
of ‫ ≥ݐ‬6 ݉ ݉ and materials with yield of less than 414 MPa [2].
6.2.1 Design Provisions
6.2.1.1 Members in Tension
For cylindrical members that are subjected to tension, the condition that should
be satisfied is:
݂௧ ≤ ߮௧‫ܨ‬௬ (6-1)
Where:
‫ܨ‬௬ : Nominal yield strength (MPa)
݂௧ : Axial tensile stress (MPa)
߮௧ : Resistance factor for axial tensile strength (= 0.95)
6.2.1.2 Members in Compression
For cylindrical members that are subjected to compression, the condition that
should be satisfied is:
݂௖ ≤ ߮௖‫ܨ‬௖௡ (6-2)
Where:
‫ܨ‬௖௡ : Nominal axial compressive strength (MPa)
݂௖ : Axial compressive stress (MPa)
߮௖ : Resistance factor for axial compressive strength (= 0.85)
218
An additional check should take place for members subjected to column or local
buckling. For column buckling, the nominal axial compressive strength ‫ܨ‬௖௡
should be determined as:
‫ܨ‬௖௡ = [1.0 − 0.25ߣଶ]‫ܨ‬௬ , for ߣ < √2
ଵ
‫ܨ‬௖௡ = ఒమ ‫ܨ‬௬ , for ߣ ≥ √2
(6-3)
‫ܨ ܮܭ‬௬ ଴.ହ
ߣ= ൤ ൨
ߨ‫ܧ ݎ‬
Where:
ߣ : Column slenderness parameter (see Table 39)
‫ ܧ‬: Young’s modulus of Elasticity (MPa)
‫ ܭ‬: Effective length factor
‫ ܮ‬: Un-braced Length (m)
‫ ݎ‬: Radius of gyration (m)
‫ܨ‬௬ : Nominal yield strength (smaller of ‫ܨ‬௫௘ and ‫ܨ‬௫௖)
The nominal elastic local buckling strength ‫ܨ‬௫௘ should be determined from:
‫ܨ‬௫௘ = 2‫ܥ‬௫‫ݐ( ܧ‬⁄‫) ܦ‬ (6-4)
Where:
‫ܥ‬௫ : Critical elastic buckling coefficient (= 0.3 − 0.6)
‫ ܦ‬: Outside diameter (m)
‫ ݐ‬: Wall thickness (m)
‫ ݔ‬: Subscript for the member longitudinal axis
The nominal inelastic local buckling stress ‫ܨ‬௫௖ should be determined from:
‫ܨ‬௫௖ = ‫ܨ‬௬ , for (‫ ܦ‬⁄‫ ≤ )ݐ‬60

(6-5)
‫ܨ‬௫௖ = ൣ1.64 − 0.23 ∙ (‫ ܦ‬⁄‫ݐ‬ )ଵ⁄ସ
൧‫ܨ‬௬ , for (‫ ܦ‬⁄‫ > )ݐ‬60
219
6.2.1.3 Members in Bending
For cylindrical members subjected to bending stress, the condition that should
be satisfied is:
݂௕ ≤ ߮௕‫ܨ‬௕௡ (6-6)
Where:
݂௕ : (= ‫ ܯ‬/ܵ) bending stress (MPa)
ܵ : Elastic section Modulus (mm3)
‫ ܯ‬: Applied bending moment (Nm)
߮௕ : Resistance factor for bending strength (= 0.95)
‫ܨ‬௕௡ : Nominal bending strength (MPa)

௓ ஽ ଵହ଴଴
‫ܨ‬௕ = ௌ ‫ܨ‬௬ , for ≤
௧ ி೤
ி೤ ∙஽ ௓ ଵ଴ଷସ଴ ஽ ଶ଴଺଼଴
‫ܨ‬௕ = ቂ1.13 − 2.58 ∙ ቃௌ ‫ܨ‬௬ , for < ≤ (6-7)
ா௧ ி೤ ௧ ி೤
ி೤ ∙஽ ௓ ଶ଴଺଼଴ ஽
‫ܨ‬௕ = ቂ0.94 − 0.76 ∙ ቃௌ ‫ܨ‬௬ , for < ≤ 300
ா௧ ி೤ ௧
Where:
ܼ : Plastic Section Modulus (mm3)
6.2.1.4 Shear loads
For cylindrical members subjected to shear loads, the condition to be satisfied

is:
2ܸ
݂௩ = ≤ ߮௩‫ܨ‬௩௡ (6-8)
‫ܣ‬
Where:
‫ܨ‬௩௡ : ൫= ‫ܨ‬௬ /√3൯Nominal Shear strength (MPa)
݂௩ : Maximum shear stress (MPa)
ܸ : Beam shear (N)
220
‫ ܣ‬: Cross sectional area (m2)
߮௩ : Resistance factor for beam shear strength (= 0.95)
For cylindrical members subjected to torsional loads, the condition to be

satisfied is:
‫ ܯ‬௩௧‫ܦ‬
݂௩௧ = ≤ ߮௩‫ܨ‬௩௧௡ (6-9)
2‫ܫ‬௣
Where:
‫ܨ‬௩௧௡ : ൫= ‫ܨ‬௬ /√3൯Nominal torsional strength (MPa)
݂௩௧ : Torsional shear stress (MPa)
‫ ܯ‬௩௧ : Torsional moment (Nm)
‫ܫ‬௣ : Polar moment of inertia (m4)
6.2.1.5 Hydrostatic Pressure
For the consideration of the hydrostatic pressure, the design hydrostatic head
should be derived as:
‫ߛ = ݌‬஽ ‫ܪ ݓ‬௭
‫ܪ‬௪ ܿ‫ݏ݋‬ℎ൫݇(݀ − ‫)ݖ‬൯ (6-10)

‫ܪ‬௭ = ‫ݖ‬+ ቈ ቉
2 ܿ‫ݏ݋‬ℎ(݇݀)
Where:
‫ ݌‬: Hydrostatic Pressure (MPa)
ߛ஽ : Hydrostatic pressure load factor
‫ ݓ‬: Sea water density (MN/m3)
‫ ݖ‬: Depth below still water surface (m)
‫ܪ‬௪ : Wave Height (m)
݇ : (= 2ߨ/‫)ܮ‬, where ‫ ܮ‬: Wave length (m)
݀ : Still water depth (m)
221
For cylindrical members subjected to external (hydrostatic) pressure the
condition to be satisfied is:
‫ܦ݌‬
݂௛ = ≤ ߮௕‫ܨ‬௛௖ (6-11)
2‫ݐ‬
Where:
݂௛ : Hoop stress due to factored hydrostatic pressure (MPa)
߮௕ : Resistance factor for hoop buckling strength (= 0.80)
‫ܨ‬௛௖ : Nominal critical hoop buckling strength
The elastic buckling nominal critical strength ‫ܨ‬௛௖ should be determined as:
‫ܨ‬௛௖ = ‫ܨ‬௛௘, for ‫ܨ‬௛௘ ≤ 0.55‫ܨ‬௬ (6-12)
The inelastic buckling nominal critical strength ‫ܨ‬௛௖ should be determined as:
଴.ସ
ி
‫ܨ‬௛௖ = 0.7‫ܨ‬௬ ൤ி೓೐൨ ≤ ‫ܨ‬௬ , for ‫ܨ‬௛௘ > 0.55‫ܨ‬௬ (6-13)
೤
Where the elastic hoop buckling stress ‫ܨ‬௛௘ is determined as:
‫ܨ‬௛௘ = 2‫ܥ‬௛ ‫ݐ( ܧ‬⁄‫) ܦ‬

଴.ସସ௧
‫ܥ‬௛ = , for ‫ ≥ ܯ‬1.6‫ܦ‬/‫ݐ‬
஽
଴.ସସ௧ ଴.ଶଵ(஽/௧)య (6-14)

‫ܥ‬௛ = + , for 0.825‫ܦ‬/‫ ≤ ܯ ≤ݐ‬1.6‫ܦ‬/‫ݐ‬
஽ ெర
‫ܥ‬௛ = 0.737/(‫ ܯ‬− 0.579) for 1.5‫ܦ‬/‫ ≤ ܯ ≤ݐ‬0.825‫ܦ‬/‫ݐ‬
‫ܥ‬௛ = 0.8 for ‫ < ܯ‬1.5
Where:
௅ ଶ஽
‫= ܯ‬஽ට : Geometrical parameter
௧
‫ ܮ‬: Length of cylinder between stiffeners (m)
6.2.1.6 Members under combined loads
For cylindrical members under axial tension and bending loads, the condition
that should be satisfied is:
222
ଶ ଴.ହ
ߨ (݂௧) ቂ൫݂௕௬ ൯ + (݂௕௭)ଶቃ
ܷ௠ = 1 − ܿ‫ݏ݋‬ቈ ቉+ ≤ 1.0 (6-15)
2 ߮௧‫ܨ‬௬ ߮௕‫ܨ‬௕௡
Where:
݂௕௬ , ݂௕௭ : Bending stresses about y and z- axis (MPa)
For cylindrical members under combined axial compressive and bending loads,
the conditions that should be satisfied is:
ଶ ଶ ଴.ହ
݂௖ 1 ⎧ ‫ܥ‬௠ ௬ ∙ ݂௕௬ ‫ܥ‬௠ ௭ ∙ ݂௕௭ ⎫

ܷ௠ = + ൦ ൪ +൦ ൪ ≤ 1.0 (6-16)
߮௖‫ܨ‬௖௡ ߮௕‫ܨ‬௕௡ ⎨ ݂௘ ݂௘
1−߮ ‫ܨ‬ 1−߮ ‫⎬ ܨ‬
⎩ ௖ ௘௬ ௖ ௘௭ ⎭
ଶ ଴.ହ
ߨ (݂௖) ቂ൫݂௕௬ ൯ + (݂௕௭)ଶቃ
ܷ௠ = 1 − ܿ‫ݏ݋‬ቈ ቉+ ≤ 1.0 (6-17)
2 ߮௖‫ܨ‬௫௖ ߮௕‫ܨ‬௕௡
‫ܨ‬௖ < ߮௖‫ܨ‬௫௖ (6-18)
Where:
‫ܥ‬௠ ௬ , ‫ܥ‬௠ ௭ : Reduction factors (Table 39)
‫ܨ‬௘௬ = ‫ܨ‬௬ /ߣଶ௬ , ‫ܨ‬௘௭ = ‫ܨ‬௭/ߣଶ௭ : Euler buckling strengths (MPa)
ߣ௬ , ߣ௭ : Column slenderness parameters (eq. 7-3c)
Incorporating the hydrostatic pressure, the condition that should be satisfied for
members subjected to longitudinal tensile stresses due to axial tension and
bending, and hoop compressive strength is:
ଶ
݂௧ + ݂௕ − (0.5݂௛ ) ݂௛ ଶ௡ ݂௧ + ݂௕ − (0.5݂௛ ) ݂௛
ܷ௠ = ቆ ቇ +൬ ൰ + 2‫ ݑ‬ቤ ቤ ≤ 1.0 (6-19)
߮௧‫ܨ‬௬ ߮௛ ‫ܨ‬௛௖ ߮௧‫ܨ‬௬ ߮௛ ‫ܨ‬௛௖
Where:
‫ ݑ‬: Poisson’s Ration (= 0.3)
For members subjected to longitudinal compressive stresses due to axial

compression, bending, and hoop compressive strength the fundamental
equations for combined loads should be applied. Further, when the axial
223
utilization factor exceeds 0.5 (݂௫ > 0.5߮௛ ‫ܨ‬௛௘), the following condition should be
satisfied:
݂௫ − 0.5߮௛ ‫ܨ‬௛௘ ݂௛ ଶ
ܷ௠ = +൬ ൰ ≤ 1.0 (6-20)
߮௖‫ܨ‬௫௘ − 0.5߮௛ ‫ܨ‬௛௘ ߮௛ ‫ܨ‬௛௖
Where: ݂௫ = ݂௖ + ݂௕ + (0.5݂௛ )
Situation K Cm1
Superstructure legs
Braced 1.0 a)
Portal (un-braced) K2 a)
Structure legs and piling
Grouted composite section 1.0 c)
Un-grouted legs 1.0 c)
Un-grouted piling between shim points 1.0 b)
Jacket Braces
Face to face length of main diagonals 0.8 b) or c)
Face of leg to centerline of joint length of K-braces 0.8 c)
Longer segment Length of X-Braces 0.9 c)
Secondary Horizontals 0.7 c)
Deck Truss Chord Members 1.0 a) b) or c)
Deck truss web members
In plane action 0.8 b)
Out of plane action 1.0 a) or b)
a. Cm values for the three cases defined in this table are as follows:
1) 0.85;
‫ܯ‬ଵ
2) 0.40 < ‫ܥ‬௠ = 0,6 − 0,4 ∙ ൗ‫ < ܯ‬0.85, where ‫ ܯ‬ଵ⁄‫ ܯ‬ଶ is the ratio of smaller to larger moments at the ends
ଶ
of the unbraced portion of the member in the plane of bending under consideration. ‫ ܯ‬ଵ⁄‫ ܯ‬ଶ is positive
when the member is bent in reverse curvature, negative when bent in single curvature.
3) ‫ܥ‬௠ = 1,0 − 0,4 ∙ (ߪ௖⁄݂௘), or 0,85, whichever is less
b. Use effective length alignment chart. This may be modified to account for conditions different from those assumed
in the development of the chart.
c. At least one pair of members framing into a joint shall be in tension, if the joint is not braced out-of-plane.
Table 39: Effective length and bending reduction factors
6.2.2 Numerical Results
Table 40, summarizes the limit states that have been considered in the
reliability assessment of the reference structure. The corresponding paragraph
and number of each limit state as it is referred to in the standard is included, as
224
well as the identification number that corresponds to this limit states in this
Chapter. Following, in Table 41 to Table 43, the results of reliability analysis of
each limit state are presented for the different classes of members.
Limit State Description Equation
ufShear Usage factor due to shear action 6-8

ufTorsion Usage factor due to torsional action 6-9
ufD211 Usage factor according to equation (D.2.1-1), i.e. axial tension 6-1
ufD252 Usage factor according to equation (D.2.5-2), i.e. hydrostatic pressure 6-11
Usage factor according to equation (D.3.1-1), i.e. axial tension and
ufD311 6-15
bending
Usage factor according to equation (D.3.2-1), i.e. axial compression and
ufD321 6-16
bending
Usage factor according to equation (D.3.2-2), i.e. axial compression and
ufD322 6-17
bending
ufD323 Usage factor according to equation (D.3.2-3), i.e. axial compression 6-18
Usage factor according to equation (D3.3-1), i.e. axial tension, bending
ufD331 6-19
and hydrostatic pressure
Usage factor according to equation (D.3.4-1), i.e. axial compression,
ufD341 6-20
bending and hydrostatic pressure
Table 40: Limit States according to API-LRFD
ufShear ufTorsion ufD211 ufD252 ufD311 ufD321 ufD322 ufD323 ufD331 ufD341
b101 0.00 21.43 5.97 0.00 15.30 7.80 13.26 10.78 0.00 0.00
b102 33.85 18.32 0.00 0.00 0.00 4.76 6.33 5.73 0.00 0.00
b103 33.82 18.33 0.00 0.00 0.00 4.76 6.32 5.74 0.00 0.00
b104 0.00 21.41 5.97 0.00 15.32 7.80 13.26 10.80 0.00 0.00
b201 0.00 26.62 7.06 0.00 0.00 7.36 12.81 10.05 0.00 0.00
b202 53.40 27.99 0.00 0.00 0.00 6.62 10.52 7.06 0.00 0.00
b203 53.33 28.00 0.00 0.00 0.00 6.62 10.51 7.07 0.00 0.00
b204 0.00 26.66 7.04 0.00 0.00 7.36 12.82 10.05 0.00 0.00
b301 37.35 30.82 10.49 0.00 9.56 0.00 0.00 29.69 0.00 0.00
b302 24.56 26.08 0.00 0.00 0.00 7.75 11.32 10.02 0.00 0.00
b303 24.56 26.06 0.00 0.00 0.00 7.76 11.30 10.03 0.00 0.00
b304 37.37 30.85 10.51 0.00 9.56 0.00 0.00 29.64 0.00 0.00
b401 16.84 11.74 0.00 0.00 0.00 31.61 17.49 0.00 0.00 0.00
b402 18.31 12.91 0.00 0.00 0.00 11.65 13.73 23.61 0.00 0.00
b403 18.32 12.91 0.00 0.00 0.00 11.65 13.73 23.58 0.00 0.00
b404 16.84 11.73 0.00 0.00 0.00 31.59 17.47 0.00 0.00 0.00
Table 41: Reliability indices for Leg Members
225
b111 0.00 0.00 0.00 0.00 0.00 59.43 59.43 0.00 0.00 0.00
b112 0.00 0.00 0.00 0.00 0.00 59.43 59.43 0.00 0.00 0.00
b113 0.00 0.00 0.00 0.00 0.00 59.43 59.43 0.00 0.00 0.00
b114 0.00 0.00 0.00 0.00 0.00 59.43 59.43 0.00 0.00 0.00
b115 0.00 0.00 0.00 82.54 0.00 43.01 38.75 0.00 0.00 0.00
b116 0.00 0.00 0.00 82.54 0.00 43.01 38.75 0.00 0.00 0.00
b211 0.00 53.23 0.00 0.00 13.84 0.00 0.00 0.00 0.00 0.00
b212 30.24 30.21 13.24 0.00 14.53 0.00 0.00 0.00 0.00 0.00
b213 0.00 53.21 0.00 0.00 13.84 0.00 0.00 0.00 0.00 0.00
b214 31.94 31.94 57.97 0.00 6.50 10.28 0.00 12.49 0.00 0.00
b215 0.00 53.05 0.00 92.59 0.00 0.00 0.00 0.00 0.00 0.00
b216 80.00 44.07 0.00 92.64 0.00 0.00 0.00 0.00 33.45 0.00
b311 56.65 71.67 57.76 0.00 11.70 0.00 0.00 0.00 0.00 0.00
b312 20.90 20.89 15.89 0.00 8.75 0.00 0.00 0.00 0.00 0.00
b313 56.62 71.65 57.78 0.00 11.69 0.00 0.00 0.00 0.00 0.00
b314 22.57 22.58 32.73 0.00 6.15 7.24 0.00 14.18 0.00 0.00
b315 46.56 71.55 0.00 85.05 0.00 0.00 0.00 0.00 35.53 0.00
b316 46.60 71.61 0.00 85.06 0.00 0.00 0.00 0.00 35.58 0.00
b411 53.26 38.70 0.00 0.00 10.19 0.00 0.00 0.00 0.00 0.00
b412 18.54 18.53 20.55 0.00 7.00 0.00 0.00 0.00 0.00 0.00
b413 53.17 38.70 0.00 0.00 10.18 0.00 0.00 0.00 0.00 0.00
b414 17.75 17.74 36.06 0.00 0.00 4.29 4.69 17.09 0.00 0.00
b415 22.83 19.25 0.00 59.57 0.00 7.11 6.64 80.46 0.00 0.00
b416 22.81 19.27 0.00 59.60 0.00 7.11 6.63 80.39 0.00 0.00
Table 42: Reliability indices for Horizontal Brace Members
b151 0.00 26.31 0.00 0.00 0.00 0.00 0.00 0.00 8.26 0.00
b152 0.00 0.00 0.00 82.54 0.00 0.00 8.63 6.11 0.00 18.92
b153 29.71 23.02 0.00 82.54 0.00 0.00 8.61 14.24 0.00 24.20
b154 29.69 23.01 0.00 82.53 0.00 0.00 8.62 14.24 0.00 24.20
b155 0.00 0.00 0.00 82.53 0.00 0.00 8.64 6.10 0.00 18.91
b156 0.00 26.33 0.00 0.00 0.00 0.00 0.00 0.00 8.26 0.00
b157 27.85 22.12 0.00 82.54 0.00 11.14 0.00 21.59 11.79 0.00
b158 27.83 22.12 0.00 82.54 0.00 11.14 0.00 21.52 11.80 0.00
b251 40.07 22.59 0.00 0.00 0.00 0.00 0.00 0.00 7.01 0.00
b252 0.00 29.70 0.00 92.69 0.00 0.00 7.47 6.00 0.00 18.38
b253 22.92 20.95 0.00 92.71 0.00 0.00 8.99 20.81 0.00 32.79
b254 22.93 20.96 0.00 92.70 0.00 0.00 8.99 20.80 0.00 32.77
b255 0.00 29.70 0.00 92.69 0.00 0.00 7.47 6.00 0.00 18.41
b256 40.07 22.62 0.00 0.00 0.00 0.00 0.00 0.00 7.01 0.00
b257 23.45 20.25 0.00 92.72 0.00 0.00 0.00 0.00 12.35 0.00
226
b258 23.45 20.29 0.00 92.72 0.00 0.00 0.00 0.00 12.35 0.00
b351 65.11 23.30 0.00 0.00 0.00 0.00 0.00 0.00 7.10 0.00
b352 32.76 19.94 0.00 0.00 0.00 0.00 7.32 6.29 0.00 18.89
b353 19.72 14.84 0.00 11.15 0.00 0.00 7.33 16.61 0.00 20.05
b354 19.73 14.83 0.00 11.15 0.00 0.00 7.34 16.61 0.00 20.05
b355 32.73 19.95 0.00 0.00 0.00 0.00 7.32 6.30 0.00 18.91
b356 65.14 23.31 0.00 0.00 0.00 0.00 0.00 0.00 7.11 0.00
b357 46.43 0.00 0.00 17.12 0.00 0.00 9.25 0.00 0.00 21.42
b358 46.44 0.00 0.00 17.15 0.00 0.00 9.25 0.00 0.00 21.43
b451 0.00 0.00 6.68 0.00 7.85 0.00 0.00 0.00 0.00 0.00
b452 59.92 29.74 0.00 0.00 0.00 4.56 18.59 7.40 0.00 0.00
b453 13.68 11.57 0.00 0.00 0.00 3.03 4.40 0.00 0.00 0.00
b454 13.68 11.56 0.00 0.00 0.00 3.02 4.41 0.00 0.00 0.00
b455 59.86 29.73 0.00 0.00 0.00 4.56 18.58 7.41 0.00 0.00
b456 0.00 0.00 6.67 0.00 7.84 0.00 0.00 0.00 0.00 0.00
b457 9.80 9.81 0.00 15.57 0.00 4.45 4.10 24.76 0.00 0.00
b458 9.81 9.82 0.00 15.57 0.00 4.44 4.10 24.73 0.00 0.00
Table 43: Reliability indices for Vertical Brace Members
The final value of the reliability index for every member will be derived as the
minimum value calculated from each of the limit states examined. Results
referring to the legs members, starting from the first elevation, distinguish one
set of members in axial compression and bending (b102-b103) with ߚ = 4.76
and one set in axial tension (b101-b104) with ߚ = 5.97. Due to the uniform
cross section of the legs, it is expected that the values of ߚ of this elevation will
be the lowest one, sizing the rest of the corresponding leg parts. Moving to the
second elevation the same pattern is followed with members (b202-b203) in
axial compression and bending with ߚ = 6.62 and the second set (b201-b204)
in axial tension with ߚ = 7.06. The same pattern can be observed in the third
elevation with the set of members (b302-b303) subjected to axial compression
and bending with ߚ = 7.75 and the set of (b301-b304) subjected to axial tension
with ߚ = 10.51. A different behaviour can be observed in the fourth elevation leg
members which are characterized by their performance to torsion with ߚ =
11.73 for (b401-b404) and ߚ = 12.91 for (b402-b403); this varying pattern is due
to the fact that topside loads have been incorporated in the design using ultra
227
stiff elements. An analytical representation of those loads would reduce this
effect on the reliability assessment.
Horizontal bracing members of the first elevation, have great values of reliability
due to the insignificant loads they carry based on the piling conditions that have
been applied. Members of the first elevation that stand parallel to the flow
(b211-b213) have ߚ = 13.24 due to axial tension and bending, while for the
members vertical to the flow, the one that meets the flow first (b214) has a
lower value of ߚ = 6.50 due to axial tension and bending and the second one
(b212) has ߚ = 13.24 due to axial tension. The two X-bracing members (b215-
b216) have great values of reliability since they carry minimum loads as they
have a non functional load to the operation of the structures. In the third
elevation, the reliability indices of all of the members forming the square braces
are determined by the axial tension and bending limit states where the two
symmetrical members, parallel to the flow (b311-b313) have ߚ = 11.69, the first
vertical to the flow (b314) ߚ = 6.15 and the second (b312) ߚ = 8.75. The
members parallel to the flow of the fourth elevation (b411-b413) have ߚ = 10.18
due to axial tension and bending, the first member vertical to the flow (b414)
ߚ = 4.29 due to axial compression and bending while the second member
(b412) ߚ = 7.00 due to axial tension and bending. At this elevation, the X brace
members show values of ߚ = 6.63 due to the axial compression and bending
criterion, due to the loads transmitted by the topside loads.
For the vertical X-braces of the first elevation, the two symmetrical sets of
members parallel to the flow, have one of their member (b152-b155) defined by
the axial compression limit state with ߚ = 6.10 and the other of their members
(b151-b156) with ߚ = 8.26 subjected to axial tension, bending and
hydrodynamic pressure. From the sets vertical to the flow, the first (b157-b158)
has ߚ = 11.14 due to axial compression and bending, while the second one
(b153-b154) has ߚ = 8.61 due to axial compression and bending. From the
members of the second elevation that stand parallel to the flow, members
(b252-b255) are subjected to axial compression with ߚ = 6.00 while members
(b251-b256) are subjected in axial tension, bending and hydrostatic pressure
228
with ߚ = 7.01. From the sets vertical to the flow, the first (b257-b258) has
ߚ = 12.35 due to axial tension, bending and hydrostatic pressure, while the
second one (b253-b254) has ߚ = 8.99 due to axial compression and bending.
From the members of the third elevation that stand parallel to the flow,
members (b352-b355) are subjected to axial compression with ߚ = 6.29 while
members (b351-b356) are subjected in axial tension, bending and hydrostatic
pressure with ߚ = 7.10. From the sets vertical to the flow, the first (b357-b358)
has ߚ = 9.25 due to axial compression and bending, while the second one
(b353-b354) has ߚ = 7.33 due to axial compression and bending. Finally for the
members of the fourth elevation, those that stand parallel to the flow, (b452-
b455) are subjected to axial compression and bending with ߚ = 4.56 while
members (b451-b456) are subjected in axial tension with ߚ = 6.68. From the
sets vertical to the flow, the first (b457-b458) has ߚ = 4.44 due to axial
compression and bending, while the second one (b453-b454) has ߚ = 3.02 due
to axial compression and bending.
6.3 ISO 19902:2002: Petroleum and natural gas industries-

general requirements for offshore structures
In this section, the relevant provisions of ISO 19902 will be presented, as they
are included in Chapter 13 of this standard. After an initial reference to the
design requirements of members subjected to individual tension, compression,
bending, shear or hydrostatic pressure, and provisions for combined forces
results, of the numerical application will be presented [3].
6.3.1.1 Members in Axial Tension
For members subjected to axial tension, the design condition that should be
satisfied is:
229
݂௧
ߪ௧ ≤ (6-21)
ߛோ,௧
Where:
ߪ௧ : Axial tensile stress (MPa)
݂௧ : Representative Axial tensile strength ݂௧ = ݂௬ (MPa)
݂௬ : Representative yield strength (MPa)
ߛோ,௧ : Partial resistance factor for axial tensile strength (= 1.05)
The derived utilization of a member subjected to tension is therefore:

ߪ௧
ܷ௠ =
݂௧ (6-22)
ߛோ,௧
6.3.1.2 Members in Axial Compression
Members subjected to axial compression should be checked both for ultimate

as well as buckling strength. The corresponding utilization factor for ultimate
strength would be:
ߪ௖
ܷ௠ =
݂௖ (6-23)
ߛோ,௖
Where:
ߪ௖ : Axial compressive stress (MPa)
݂௖ : Representative Axial compressive strength as will be derived below (MPa)
ߛோ,௖ : Partial resistance factor for axial compressive strength (= 1.18)
The representative axial compressive strength for tubular members ݂௖ without

hydrostatic pressure would be derived by the smallest values of the following
equations:
݂௖ = (1.0 − 0.278ߣଶ)݂௬௖, for ߣ ≤ 1.34

(6-24)
଴.ଽ
݂௖ = ݂ , for ߣ ≤ 1.34
ఒమ ௬௖
230
݂௬௖ ‫݂ ܮܭ‬௬௖
ߣ= ඨ = ඨ
݂௘ ߨ‫ܧ ݎ‬
Where:
݂௬௖ : Representative local buckling strength, as will be derived below (MPa)
ߣ : Column Slenderness parameter
݂௘ : Euler buckling strength (MPa)
‫ ܧ‬: Young’s Modulus of Elasticity (MPa)
‫ ܭ‬: Effective length factor
‫ ܮ‬: Un-braced length (m)
‫ ݎ‬: Radius of gyration ൫= ඥ‫ܫ‬/‫ܣ‬൯(m)
‫ܫ‬: Moment of inertia (kg·m²)
‫ ܣ‬: Cross-sectional area (m2)
The representative local buckling strength ݂௬௖ shall be determined as:
௙೤
݂௬௖ = ݂௬ , for ௙ ≤ 0.170
ೣ೐
(6-25)
௙೤ ௙೤
݂௬௖ = ቀ1.047 − 0.274 ௙ ቁ݂௬ , for ௙ > 0.170
ೣ೐ ೣ೐
݂௫௘ = 2‫ܥ‬௫‫ݐܧ‬/‫ܦ‬ (6-26)
Where:
݂௫௘ : Representative elastic local buckling strength (MPa)
‫ܥ‬௫ : Elastic critical buckling coefficient (= 0.3 − 0.6)

‫ ܦ‬: Outside diameter of member (m)
‫ݐ‬: Wall thickness of member (mm)
6.3.1.3 Bending
The corresponding utilization factor for members subjected to bending should

be:
231
ߪ௕ ‫ ܯ‬/ܼ௘
ܷ௠ = = ≤ 1.0
݂௕ ݂௕ (6-27)
ߛோ,௕ ߛோ,௕
Where:
ߪ௕ : Bending stress (MPa)
݂௕ : Representative bending strength (MPa)
ߛோ,௕ : Partial resistance factor for bending strength (= 1.05)
‫ ܯ‬: Bending moment (Nm)
ܼ௘ : Elastic section modulus (mm3)
The representative bending strength should be determined as:

௓೛ ௙೤ ஽
݂௕ = ቀ௓ ቁ݂௬ , for ≤ 0.0517
೐ ா௧
௙೤ ஽ ௓೛ ௙೤ ஽
݂௕ = ቂ1.13 − 2.58 ቀ ா௧ ቁቃቀ௓ ቁ݂௬ , for 0.0517 < ≤ 0.1034 (6-28)
೐ ா௧
௙೤ ஽ ௓೛ ௙೤ ஽ ௙೤
݂௕ = ቂ0.94 − 0.76 ቀ ா௧ ቁቃቀ௓ ቁ݂௬ , for 0.1034 < ≤ 120 ா
೐ ா௧
6.3.1.4 Shear
For members, subjected to shear force, the utilization factor should be derived
as:
߬௕ 2ܸ/‫ܣ‬
ܷ௠ = = ≤ 1.0
݂ఔ ݂ఔ (6-29)
ߛோ,ఔ ߛோ,ఔ
Where:
߬௕ : Maximum beam shear stress (MPa)
݂ఔ : Representative shear strength ൫= ݂௬ /√3൯(MPa)
ߛோ,ఔ : Partial resistance factor for shear strength (= 1.05)
ܸ : Beam shear (N)
232
For members subjected to torsional shear force, the utilization factor should be
derived as:
߬௧ ‫ ܯ‬௩,௧‫ܦ‬/2‫ܫ‬௣
ܷ௠ = = ≤ 1.0
݂ఔ ݂ఔ (6-30)
ߛோ,ఔ ߛோ,ఔ
Where:
߬௧ : Torsional shear stress (MPa)
‫ ܯ‬௩,௧ : Torsional moment due to factored actions (Nm)

గ
‫ܫ‬௣ = ଶଵ [‫ ܦ‬ସ − (‫ ܦ‬− 2‫)ݐ‬ସ] : Polar moment of inertia (m3)
6.3.1.5 Hydrostatic Pressure
Provisions of ISO 19902 for calculation of hydrostatic pressure and formulation

of utilization criteria for hoop buckling stress are exactly the same as the ones
included in API RP-2A LRFD and have been presented in the previous section.
Therefore, they will not be analytically presented here. The utilization factor of a
member under external pressure is derived as:
ߪ௛ ‫ܦ݌‬/2‫ݐ‬
ܷ௠ = = ≤ 1.0
݂௛ ݂௛ (6-31)
ߛோ,௛ ߛோ,௛
6.3.1.6 Members subjected to combined forces without hydrostatic

pressure
For tubular members subjected to combined axial tension and bending forces,
without hydrostatic pressure, the condition that should be satisfied is:
ଶ ଶ
ߛோ,ௗ ∙ ට ߪ௕,௬ + ߪ௕,௭
ߛோ,௧ ∙ ߪ௧ (6-32)
ܷ௠ = + ≤ 1.0
݂௧ ݂௕
Where:
ߪ௕,௬ , ߪ௕,௭ : Bending stress about the member y and z-axis (MPa)
233
For tubular members subjected to combined axial compression and bending
forces without hydrostatic pressure, the condition that should be satisfied is:
ଶ ଶ ଴,ହ
ߛோ,௖ ∙ ߪ௖ ߛோ,௕ ‫ܥ‬௠ ,௬∙ ߪ௕,௬ ‫ܥ‬௠ ,௭∙ ߪ௕,௭
ܷ௠ = + ∙ ൥ቆ ቇ +ቆ ቇ൩ ≤ 1.0 (6-33)
݂௖ ݂௕ 1 − ߪ௖⁄݂௘,௬ 1 − ߪ௖⁄݂௘,௭
ଶ ଶ
ߛோ,௕ ∙ ටߪ௕,௬ + ߪ௕,௭
ߛோ,௖ ∙ ߪ௖ (6-34)
ܷ௠ = + ≤ 1.0
݂௬௖ ݂௕
Where:
‫ܥ‬௠ ,௬ , ‫ܥ‬௠ ,௭ : Moment reduction factors corresponding to the member y- and z-
axes (see Table 44)
గ మா గ మா
݂௘,௬ = మ , ݂௘,௭ = (௄ మ : Euler buckling strengths corresponding to the
൫௄೤ ∙௅೤ ⁄௥൯ ೥∙௅೥⁄௥)
member y- and z-axes (MPa)
‫ܭ‬௬ , ‫ܭ‬௭ : Effective length factors for the y- and z-directions (see Table 44)
‫ܮ‬௬ , ‫ܮ‬௭ : Un-braced lengths in the y- and z-directions (m)
6.3.1.7 Members subjected to combined forces with hydrostatic pressure
In the presence of hydrostatic pressure, for tubular members subjected to

combined axial tension, bending and hydrostatic pressure shall be designed to
satisfy:
ଶ ଶ
ߛோ,௕ ∙ ටߪ௕,௬ + ߪ௕,௭
ߛோ,௧ ∙ ߪ௧,௖ (6-35)
ܷ௠ = + ≤ 1.0
݂௧,௛ ݂௕,௛
Where:
݂௧,௛ = ݂௬ ∙ ቀඥ1 + 0,09 ∙ ‫ ܤ‬ଶ − ‫ ܤ‬ଶఎ − 0,3 ∙ ‫ܤ‬ቁ : Representative axial tensile strength
in the presence of external hydrostatic pressure
݂௕,௛ = ݂௕ ∙ ቀඥ1 + 0,09 ∙ ‫ ܤ‬ଶ − ‫ ܤ‬ଶఎ − 0,3 ∙ ‫ܤ‬ቁ : Representative bending strength in
the presence of external hydrostatic pressure
234
ఊೃ,೓ ∙ఙ೓ ௙
‫=ܤ‬ ≤ 1.0, ߟ = 5 − 4 ∙ ௙೓
௙೓ ೤
For tubular members subjected to combined axial compression, bending and

hydrostatic pressure shall be designed to satisfy:
ଶ ଶ
ߛோ,௕ ∙ ටߪ௕,௬ + ߪ௕,௭
ߛோ,௖ ∙ ߪ௖,௖ (6-36)
ܷ௠ = + ≤ 1.0
݂௬௖ ݂௕,௛
ଶ ଶ ଴,ହ
ߛோ,௖ ∙ ߪ௖ ߛோ,௕ ‫ܥ‬௠ ,௬∙ ߪ௕,௬ ‫ܥ‬௠ ,௭∙ ߪ௕,௭
ܷ௠ = + ∙ ൥ቆ ቇ +ቆ ቇ൩ ≤ 1.0 (6-37)
݂௖,௛ ݂௕,௛ 1 − ߪ௖⁄݂௘,௬ 1 − ߪ௖⁄݂௘,௭
ߪ௫ − 0,5 ∙ ݂௛௘⁄ߛோ,௛ ߛோ,௛ ∙ ߪ௛ ଶ

ܷ௠ = +൬ ൰ ≤ 1.0 (6-38)
݂௫௘⁄ߛோ,௖ − 0,5 ∙ ݂௛௘⁄ߛோ,௛ ݂௛௘
Where:
݂௖,௛ : Representative axial compressive strength in the presence of external

hydrostatic pressure (MPa)
ଵ ଶఙ೜ ఙ೜
݂௖,௛ = ݂௬௖ ቈ(1,0 − 0,278ߣଶ) − + ට (1,0 − 0,278ߣଶ)ଶ + 1,12ߣଶ ቉,
ଶ ௙೤೎ ௙೤೎
ିଵ
ଶఙ೜
for ߣ ≤ 1,34ඨ ൬1 − ൰ (6-39)
௙೤೎
ିଵ
଴,ଽ ଶఙ೜
݂௖,௛ = ඨ
మ ∙ ݂௬௖,for ߣ ≤ 1,34 ൬1 − ൰
ఒ ௙೤೎
ߪ௤ = 0.5ߪ௛ : Compressive axial strength due to capped end hydrostatic action

(MPa)
ߪ௛ : Hoop stress due to forces from factored hydrostatic pressure (MPa)
235
Structural component K Cma
Topsides legs
Braced 1,0 1)
Portal (unbraced) Kb 1)
Structure legs and piling
Grouted composite section 1,0 3)
Ungrouted legs 1,0 3)
Ungrouted piling between shim points 1,0 2)
Structure brace members
Primary diagonals and horizontals 0,7 2) or 3)
c
K-braces 0,7 2) or 3)
X-braces
Longer segment length c 0,8 2) or 3)
Full length d 0,7 2) or 3)
Secondary horizontals 0,7 2) or 3)
a. Cm values for the three cases defined in this table are as follows:
1) 0,85;
2) for members with no transverse loading, other than self weight,
‫ܯ‬ଵ
‫ܥ‬௠ = 0,6 − 0,4 ∙ ൗ‫ܯ‬
ଶ
where ‫ ܯ‬ଵ⁄‫ ܯ‬ଶ is the ratio of smaller to larger moments at the ends of the unbraced portion of the member
in the plane of bending under consideration;
‫ ܯ‬ଵ⁄‫ ܯ‬ଶ is positive when the member is bent in reverse curvature, negative when bent in single curvature.
Cm shall not be larger than 0,85;
3) for members with transverse loading, other than self weight,
‫ܥ‬௠ = 1,0 − 0,4 ∙ (ߪ௖⁄݂௘), or 0,85, whichever is less,

and ݂௘ = ݂௘௬ or ݂௘௭ as appropriate.
b. See effective length alignment chart in A.13.5. This may be modified to account for conditions different from those
assumed in the development of the chart.
c. For either in-plane or out-of-plane effective lengths, at least one pair of members framing into a K- or X-joint shall
be in tension, if the joint is not braced out-of-plane.
d. When all members are in compression and the joint is not braced out-of-plane.
Table 44: Effective length and moment reduction factors for member
strength checking
and number of each limit state, as it is referred to in the standard, is included,
as well as the identification number that corresponds to this limit state in this
236
(13.2-2) Usage factor according to (13.2-2) – Axial Tension 6-22

(13.2-4) Usage factor according to (13.2-4) – Axial compression 6-23
Euler Usage factor with respect to Euler load capacity
(13.2-12) Usage factor according to (13.2-12) – Bending 6-27
(13.2-17) Usage factor according to (13.2-17) – Shear 6-29
(13.2-19) Usage factor according to (13.2-19) – Torsion 6-30
(13.3-2) Usage factor according to (13.3-2) – Tension and bending 6-32
(13.3-7) Usage factor according to (13.3-7) – Compression and bending 6-33
(13.3-8) Usage factor according to (13.3-8) – Compression and bending 6-34
(13.2-31) Usage factor according to (13.2-31) – Hoop Buckling 6-31
(13.4-12) Usage factor according to (13.4-12) – Tension, Bending, hydrostatic pressure 6-35
(13.4-19) Usage factor according to (13.4-19) – Compression, Bending, hydrostatic pressure 6-36
Table 45: Limit States according to ISO 19902
(13.2-2) (13.2-4) Euler (13.2-12) (13.2-17) (13.2-19) (13.3-2) (13.3-7) (13.3-8) (13.2-31) (13.4-12) (13.4-19) (13.4-20) (13.4-21)
b101 6.16 9.64 25.02 11.88 32.06 22.76 5.41 7.26 8.92 0.00 0.00 0.00 0.00 0.00
b102 0.00 5.52 15.22 12.22 33.82 21.42 0.00 4.71 4.95 0.00 0.00 0.00 0.00 0.00
b103 0.00 5.52 15.21 12.20 33.85 21.42 0.00 4.71 4.96 0.00 0.00 0.00 0.00 0.00
b104 6.16 9.64 24.99 11.89 32.04 22.78 5.41 7.26 8.91 0.00 0.00 0.00 0.00 0.00
b201 6.86 10.22 24.99 18.22 25.38 33.66 6.05 16.55 13.07 0.00 0.00 0.00 0.00 0.00
b202 0.00 6.82 18.75 39.88 53.37 33.69 0.00 6.56 6.85 0.00 0.00 0.00 0.00 0.00
b203 0.00 6.82 18.72 39.94 53.36 33.68 0.00 6.56 6.85 0.00 0.00 0.00 0.00 0.00
b204 6.88 10.24 25.00 18.19 25.36 33.68 6.05 16.51 13.07 0.00 0.00 0.00 0.00 0.00
b301 10.49 41.79 90.92 10.16 19.35 0.00 7.17 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b302 0.00 9.42 22.84 27.19 50.18 0.00 0.00 7.75 9.01 0.00 0.00 0.00 0.00 0.00
b303 0.00 9.42 22.85 27.20 50.19 0.00 0.00 7.75 9.02 0.00 0.00 0.00 0.00 0.00
b304 10.48 41.76 90.96 10.18 19.37 0.00 7.17 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b401 0.00 0.00 0.00 17.02 25.49 19.43 0.00 35.07 29.35 0.00 0.00 0.00 0.00 0.00
b402 0.00 19.06 40.03 18.01 25.52 20.11 0.00 11.05 12.79 0.00 0.00 0.00 0.00 0.00
b403 0.00 19.09 40.04 18.04 25.52 20.08 0.00 11.05 12.79 0.00 0.00 0.00 0.00 0.00
b404 0.00 0.00 0.00 17.03 25.51 19.42 0.00 35.08 29.34 0.00 0.00 0.00 0.00 0.00
237
(13.2-2) (13.2-4) Euler (13.2-12) (13.2-17) (13.2-19) (13.3-2) (13.3-7) (13.3-8) (13.2-31) (13.4-12) (13.4-19) (13.4-20) (13.4-21)
b111 0.00 0.00 0.00 63.25 0.00 0.00 0.00 63.25 63.25 0.00 0.00 0.00 0.00 0.00
b112 0.00 0.00 0.00 63.25 0.00 0.00 0.00 63.25 63.25 0.00 0.00 0.00 0.00 0.00
b113 0.00 0.00 0.00 63.25 0.00 0.00 0.00 63.25 63.25 0.00 0.00 0.00 0.00 0.00
b114 0.00 0.00 0.00 63.25 0.00 0.00 0.00 63.25 63.25 0.00 0.00 0.00 0.00 0.00
b115 0.00 0.00 0.00 40.05 0.00 0.00 0.00 0.00 0.00 82.54 0.00 37.50 0.00 0.00
b116 0.00 0.00 0.00 40.05 0.00 0.00 0.00 0.00 0.00 82.55 0.00 37.50 0.00 0.00
b211 0.00 0.00 0.00 13.85 0.00 80.45 13.77 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b212 13.45 0.00 0.00 15.61 30.22 0.00 9.90 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b213 0.00 0.00 0.00 13.86 0.00 80.44 13.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b214 57.94 7.96 9.26 15.74 31.93 0.00 6.43 9.44 0.00 0.00 0.00 0.00 0.00 0.00
b215 0.00 0.00 0.00 20.97 79.86 53.37 0.00 0.00 0.00 92.58 22.67 0.00 0.00 0.00
b216 0.00 0.00 0.00 20.99 79.92 53.38 0.00 0.00 0.00 92.68 22.72 0.00 0.00 0.00
b311 57.73 0.00 0.00 11.70 56.63 0.00 11.48 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b312 15.89 0.00 0.00 8.85 20.89 0.00 7.57 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b313 57.71 0.00 0.00 11.71 56.68 0.00 11.48 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b314 32.77 8.32 9.54 9.26 23.55 0.00 6.02 7.01 0.00 0.00 0.00 0.00 0.00 0.00
b315 0.00 0.00 0.00 19.59 0.00 46.39 0.00 0.00 0.00 85.05 27.92 0.00 0.00 0.00
b316 0.00 0.00 0.00 19.59 0.00 46.42 0.00 0.00 0.00 85.06 27.94 0.00 0.00 0.00
b411 0.00 0.00 0.00 10.31 53.22 0.00 10.37 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b412 18.98 0.00 0.00 7.45 18.54 0.00 6.75 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b413 0.00 0.00 0.00 10.31 53.27 0.00 10.38 0.00 0.00 0.00 0.00 0.00 0.00 0.00
b414 36.09 11.42 13.16 7.14 17.76 0.00 0.00 4.28 4.55 0.00 0.00 0.00 0.00 0.00
b415 0.00 0.00 0.00 9.70 21.95 44.36 0.00 0.00 0.00 59.63 9.65 0.00 0.00 0.00
b416 0.00 0.00 0.00 9.70 21.96 44.37 0.00 0.00 0.00 59.60 9.65 0.00 0.00 0.00
(13.2-2) (13.2-4) Euler (13.2-12) (13.2-17) (13.2-19) (13.3-2) (13.3-7) (13.3-8) (13.2-31) (13.4-12) (13.4-19) (13.4-20) (13.4-21)
b151 6.44 0.00 0.00 13.73 0.00 25.38 0.00 0.00 0.00 0.00 5.67 0.00 0.00 0.00
b152 0.00 0.00 3.23 0.00 84.93 80.48 0.00 0.00 0.00 83.16 0.00 6.03 2.20 20.07
b153 0.00 0.00 8.34 22.20 0.00 37.33 0.00 0.00 0.00 0.00 0.00 11.60 5.10 0.00
b154 0.00 0.00 8.34 22.21 0.00 37.42 0.00 0.00 0.00 0.00 0.00 11.59 5.09 0.00
b155 0.00 0.00 3.23 0.00 84.83 80.48 0.00 0.00 0.00 83.15 0.00 6.04 2.20 20.06
b156 6.45 0.00 0.00 13.72 0.00 25.39 0.00 0.00 0.00 0.00 5.66 0.00 0.00 0.00
b157 13.34 0.00 0.00 8.90 27.85 37.39 0.00 0.00 0.00 82.53 6.08 0.00 0.00 0.00
b158 13.34 0.00 0.00 8.90 27.87 37.39 0.00 0.00 0.00 82.54 6.08 0.00 0.00 0.00
b251 6.71 0.00 0.00 11.32 40.11 29.49 0.00 0.00 0.00 10.69 5.43 0.00 0.00 0.00
b252 0.00 0.00 3.47 60.04 84.98 40.61 0.00 0.00 0.00 0.00 0.00 5.99 2.24 23.17
b253 0.00 0.00 13.88 14.80 0.00 42.39 0.00 0.00 0.00 0.00 0.00 12.34 6.76 0.00
b254 0.00 0.00 13.88 14.80 0.00 42.35 0.00 0.00 0.00 0.00 0.00 12.35 6.76 0.00
b255 0.00 0.00 3.47 60.07 84.98 40.64 0.00 0.00 0.00 0.00 0.00 5.99 2.24 23.19
b256 6.71 0.00 0.00 11.32 40.12 29.50 0.00 0.00 0.00 10.68 5.44 0.00 0.00 0.00
238
(13.2-2) (13.2-4) Euler (13.2-12) (13.2-17) (13.2-19) (13.3-2) (13.3-7) (13.3-8) (13.2-31) (13.4-12) (13.4-19) (13.4-20) (13.4-21)
b257 26.77 0.00 0.00 9.08 23.47 42.44 0.00 0.00 0.00 92.66 6.31 0.00 0.00 0.00
b258 26.77 0.00 0.00 9.07 23.47 42.41 0.00 0.00 0.00 92.68 6.32 0.00 0.00 0.00
b351 6.84 0.00 0.00 9.88 38.75 30.01 0.00 0.00 0.00 17.32 5.41 0.00 0.00 0.00
b352 0.00 0.00 4.01 15.59 30.27 30.10 0.00 0.00 0.00 0.00 0.00 5.93 2.65 26.79
b353 0.00 0.00 19.92 9.72 84.91 32.76 0.00 0.00 0.00 0.00 0.00 9.02 6.36 46.37
b354 0.00 0.00 19.93 9.71 84.97 32.81 0.00 0.00 0.00 0.00 0.00 9.02 6.35 46.44
b355 0.00 0.00 4.01 15.62 30.28 30.07 0.00 0.00 0.00 0.00 0.00 5.91 2.65 26.74
b356 6.84 0.00 0.00 9.89 38.74 30.03 0.00 0.00 0.00 17.35 5.40 0.00 0.00 0.00
b357 28.38 0.00 99.26 5.85 11.76 18.34 0.00 0.00 0.00 0.00 4.71 0.00 0.00 0.00
b358 28.40 0.00 99.28 5.85 11.78 18.35 0.00 0.00 0.00 0.00 4.71 0.00 0.00 0.00
b451 7.91 0.00 0.00 12.75 31.73 25.50 0.00 0.00 0.00 80.91 6.51 0.00 0.00 0.00
b452 0.00 5.28 5.05 25.72 53.41 30.05 0.00 4.73 9.25 0.00 0.00 7.91 3.87 46.46
b453 0.00 7.64 26.26 3.86 10.12 22.95 0.00 2.82 3.28 0.00 0.00 0.00 0.00 0.00
b454 0.00 7.64 26.26 3.86 10.12 22.95 0.00 2.83 3.28 0.00 0.00 0.00 0.00 0.00
b455 0.00 5.28 5.05 25.73 53.39 30.07 0.00 4.72 9.25 0.00 0.00 7.91 3.87 46.42
b456 7.90 0.00 0.00 12.75 31.75 25.47 0.00 0.00 0.00 80.83 6.52 0.00 0.00 0.00
b457 0.00 9.44 0.00 4.38 11.42 80.38 0.00 4.90 8.32 0.00 0.00 4.53 4.66 13.16
b458 0.00 9.43 0.00 4.39 11.42 80.43 0.00 4.91 8.32 0.00 0.00 4.53 4.65 13.15
referring to the leg members, starting from the first elevation, distinguish one set
of members in axial compression and bending (b102-b103) with ߚ = 4.71 and
one set in axial tension and bending (b101-b104) with ߚ = 5.41. Due to the
uniform cross section of the legs, it is expected that the values of ߚ of this
elevation will be the lowest one, sizing the rest of the leg members. Moving to
the second elevation a different pattern is followed with members (b202-b203)
in axial compression and bending with ߚ = 6.56 having a greater value of
reliability index than the second member (b201-b204) subjected in axial tension
and bending with ߚ = 6.05. The same pattern as in the second elevation can be
observed in the third elevation with the set of members (b302-b303) subjected
to axial compression and bending with ߚ = 7.75 and the set of (b301-b304)
subjected to axial tension and bending with ߚ = 7.17. A slightly different
behaviour can be observed in elevation 4 leg members with the set of members
239
(b402-b403) subjected to axial compression and bending with ߚ = 11.05 and
the set of (b401-b404) subjected to bending with ߚ = 17.02.
due to the insignificant loads they carry due to the piling conditions that have
been applied. Members of the first elevation that stand parallel to the flow
(b211-b213) have ߚ = 13.77 due to axial tension and bending, while for the
members vertical to the flow, the one that meets the flow first (b214) has a
lower value of ߚ = 6.43 and the second one (b212) has ߚ = 9.90 both due to
axial tension and bending. The two X-bracing members (b215-b216) have great
values of reliability since they carry minimum loads as they have a non
functional load to the operation of the structures. In the third elevation, the
reliability indices of all of the members forming the square braces are
determined by the axial tension and bending limit states where the two
symmetrical members, parallel to the flow (b311-b313) have ߚ = 11.48, the first
vertical to the flow (b314) ߚ = 6.02 and the second (b312) ߚ = 7.57. The
members parallel to the flow of the fourth elevation (b411-b413) have ߚ = 10.31
due to bending, the first member vertical to the flow (b414) ߚ = 4.28 due to axial
compression and bending while the second member (b412) ߚ = 6.75 due to
axial tension and bending. At this elevation the X brace members show values
of ߚ = 9.65 due to the axial compression and bending criterion, due to the loads
transmitted from the topside loads.
members parallel to the flow, have one of their members (b152-b155) defined
from the axial compression, bending and hydrodynamic limit state with ߚ = 2.20
and the other of their members (b151-b156) with ߚ = 5.67 subjected to axial
tension, bending and hydrodynamic pressure. From the sets vertical to the flow,
the first (b157-b158) has ߚ = 6.08 due to axial tension, bending and hydrostatic
pressure, while the second one (b153-b154) has ߚ = 5.10 due to axial
compression, bending and hydrostatic pressure. From the members of the
second elevation that stand parallel to the flow, members (b252-b255) are
subjected to axial compression, bending and hydrodynamic pressure with
240
ߚ = 2.24 while members (b251-b256) are subjected in axial tension, bending
and hydrodynamic pressure with ߚ = 5.43. From the sets vertical to the flow, the
first (b257-b258) has ߚ = 6.31 due to axial tension, bending and hydrostatic
compression, bending and hydrodynamic pressure. From the members of the
third elevation that stand parallel to the flow, members (b352-b355) are
subjected to axial compression, bending and hydrodynamic pressure with
ߚ = 2.65 while members (b351-b356) are subjected in axial tension, bending
and hydrostatic pressure with ߚ = 5.41. From the sets vertical to the flow, the
first (b357-b358) has ߚ = 4.71 due to axial tension, bending and hydrodynamic
compression, bending and hydrostatic pressure. Finally for the members of the
fourth elevation, those that stand parallel to the flow, (b452-b455) are subjected
to axial compression, bending and hydrodynamic pressure with ߚ = 3.87 while
members (b451-b456) are subjected in axial tension, bending and
hydrodynamic pressure with ߚ = 6.51. From the sets vertical to the flow, the first
(b457-b458) has ߚ = 4.38 due to bending, while the second one (b453-b454)
has ߚ = 2.82 due to axial compression and bending.
6.4 BS EN 1993-1-1:2005 Eurocode 3: Design of Steel

Structures
In this section, the design requirements of Eurocode 3 will be discussed.

Provisions for cylindrical cross sections will be presented resulting to derivation
of limit states that will be used later in the analysis. The basic requirement
demands that ‘the design value of one or several action effects in each cross
section shall not exceed the corresponding design resistance for that action or
the corresponding combination’ [5]. The design values of resistance should
depend on the classification of the cross-section, while four classes can be
distinguished:
241
 Class 1 cross-sections are those which can develop their plastic moment
capacity and provide significant amount of rotation resistance.
 Class 2 cross-sections are those which can develop their plastic moment
resistance, but have limited rotation capacity because of local buckling.
 Class 3 cross-sections are those in which the stress in the extreme
compression fibre of the steel member assuming an elastic distribution of
stresses can reach the yield strength, but local buckling is liable to
prevent development of the plastic moment resistance.
 Class 4 cross-sections are those in which local buckling will occur before
the attainment of yield stress in one or more parts of the cross-section.
Table 49, presents the criteria for classification of members based on design
yield and geometrical properties.
Class Section in bending and/or compression

1 ݀/‫ ≤ݐ‬50ߝଶ
2 ݀/‫ ≤ݐ‬70ߝଶ
3 ݀/‫ ≤ݐ‬90ߝଶ
fy 235 275 355 420 460
ε 1,00 0,92 0,81 0,75 0,71
ߝ = ට 235/݂௬
ε2 1,00 0,85 0,66 0,56 0,51
Table 49: Classification of members
For elastic verification, the following conservative yield criterion should be

satisfied along the length of the member under consideration:
ଶ ଶ ଶ
ߪ௫,ாௗ ߪ௭,ாௗ ߪ௫,ாௗ ߪ௭,ாௗ ߬ாௗ
ቆ ቇ +ቆ ቇ −ቆ ቇ∙ ቆ ቇ+ 3 ∙ ቆ ቇ ≤1 (6-40)
݂௬ ⁄ߛெ ଴ ݂௬ ⁄ߛெ ଴ ݂௬ ⁄ߛெ ଴ ݂௬ ⁄ߛெ ଴ ݂௬ ⁄ߛெ ଴
Where:
ߪ௫,ாௗ : Design value of the longitudinal stress at the point of consideration (MPa)
ߪ௭,ாௗ : Design value of the transverse stress at the point of consideration (MPa)
߬ாௗ : Design value of the shear stress at the point of consideration (MPa)
242
ߛெ ଴ : Resistance partial factor of cross section applicable to all classes (1.00)
Another conservative approximation applicable to all cross section classes

introduces a linear summation of the utilization ratios for each stress resultant.
ܰாௗ ‫ ܯ‬௬,ாௗ ‫ ܯ‬௭,ாௗ

+ + ≤1 (6-41)
ܰோௗ ‫ ܯ‬௬,ோௗ ‫ ܯ‬௭,ோௗ
Where:
ܰோௗ , ‫ ܯ‬௬,ோௗ , ‫ ܯ‬௭,ோௗ : Design values of the resistance depending on the cross
sectional classification and including any reduction that may be caused by
shear effects
ܰாௗ , ‫ ܯ‬௬,ாௗ , ‫ ܯ‬௭,ாௗ : Design values of the action at point of consideration
6.4.1.1 Members in Tension
For members in tension, the design value of the tension force ܰாௗ at each cross
section should satisfy:
ܰாௗ
≤ 1.0 (6-42)
ܰ௧,ோௗ
Where,
஺∙௙೤ ଴,ଽ஺೙೐೟∙௙ೠ
ܰ௧,ோௗ = ݉ ݅݊ቀܰ௣௟,ோௗ = ఊ , ܰ௨,ோௗ = ఊಾ మ
ቁ
ಾబ
ߛெ ଶ : Resistance of cross section in tension to fracture
6.4.1.2 Members in Compression
For members in compression, the design value of the compression force ܰாௗ at
each cross-section should satisfy:
ܰாௗ
≤ 1.0 (6-43)
ܰ௖,ோௗ
Where:
஺∙௙೤
ܰ௖,ோௗ = ఊ , for class 1, 2 or 3 cross-sections
ಾబ
243
஺೐೑೑∙௙೤
ܰ௖,ோௗ = ఊಾ బ
, for class 4 cross-sections
6.4.1.3 Bending
For members under bending moment, the design value of the bending moment
‫ ܯ‬ாௗ at each cross-section should satisfy:
‫ ܯ‬ாௗ
≤ 1.0 (6-44)
‫ ܯ‬௖,ோௗ
ௐ ೛೗∙௙೤
‫ ܯ‬௖,ோௗ = ‫ ܯ‬௣௟,ோௗ = , for class 1 or 2 cross sections
ఊಾ బ
ௐ ೐೗,೘ ೔೙ ∙௙೤
‫ ܯ‬௖,ோௗ = ‫ ܯ‬௘௟,ோௗ = ఊಾ బ
, for class 3 cross sections
ௐ ೐೑೑,೘ ೔೙ ∙௙೤
‫ ܯ‬௖,ோௗ = ఊಾ బ
Where: ܹ ௘௟,௠ ௜௡ and ܹ ௘௙௙,௠ ௜௡ is the maximum elastic stress (MPa).
6.4.1.4 Shear
For members under shear loads, the design value of the shear force ܸாௗ at each
cross section should satisfy:
ܸாௗ
≤ 1.0 (6-45)
ܸ௖,ோௗ
Where: ܸ௖,ோௗ is the design shear resistance (N)
In the absence of torsion the design plastic shear resistance is given by:
‫ܣ‬௏ ൫݂௬ ⁄√3൯

ܸ௣௟,ோௗ = (6-46)
ߛெ ଴
Where: ‫ܣ‬௏ 2‫ܣ‬⁄ߨ is the shear area for cylindrical cross sections.
A conservative verification based on elastic design, suggests:
244
ܸாௗ ∙ ܵ
‫ݐ ∙ܫ‬ ≤ 1.0 (6-47)
݂௬ ⁄൫√3 ∙ ߛெ ଴൯
ܸாௗ : Design value of the shear force (N)

ܵ : First moment of area about the centroidal axis of that portion of the cross-
section between the point at which the shear is required and the boundary of
the cross-section (m3)
‫ܫ‬: Second moment of area of the whole cross section (mm4)
‫ݐ‬: The thickness at the examined point (mm)
6.4.1.5 Torsion
For members subject to torsion for which distortional deformations may be

disregarded the design value of the torsional moment ܶாௗ at each cross-section
should satisfy:
ܶாௗ
≤ 1.0 (6-48)
ܶோௗ
Where:
ܶோௗ : Design torsional resistance of the cross section.
For the elastic verification the yield criterion, eq. (6-40) can be applied.
6.4.1.6 Buckling Resistance of members
Members in compression should also be verified against buckling as follows:
ܰாௗ
≤ 1.0 (6-49)
ܰ௕,ோௗ
Where:
ܰாௗ : Design value of the compression force (N)
ܰ௕,ோௗ : Design buckling resistance of the compression member (N)
The design buckling resistance of a compression member ܰ௕,ோௗ should be taken

as:
245
ఞ∙஺∙௙೤
ܰ௕,ோௗ = , for class 1,2 and 3 cross sections
ఊಾ భ
ఞ∙஺೐೑೑∙௙೤
ܰ௕,ோௗ = , for class 4 cross sections
ఊಾ భ
Where:
ߛெ ଵ : Resistance of members to instability (1.00)
߯ : Reduction factor for the relevant buckling mode and should be determined
as follows based on the relevant buckling curve according to:
1
߯= ≤ 1.0
ߔ + ඥ ߔ ଶ − ߣଶ̅
(6-50)
ߔ = 0,5 ∙ උ1 + ߙ ∙ ൫ߣ̅ − 0,2൯+ ߣ̅ଶඏ
The non-dimensional slenderness ߣ̅ is given by:
஺∙௙೤ ௅ ଵ
ߣ̅ = ට ே = ೐ೝ ∙ , for class 1,2 and 3 cross sections
೎ೝ௜ ఒ భ
ಲ
஺೐೑೑∙௙೤ ට ೐೑೑
௅೐ೝ
ߣ̅ = ට = ∙ ಲ
ே ೎ೝ ௜ ఒభ
௲ ଶଷହ
ߣଵ = ߨට ௙ = 93, ߝ = ට ൫݂௬ ݅݊ ܰ ⁄݉ ݉ ଶ൯
೤ ௙೤
Where:
ߙ : Imperfection factor (see Table 50)
ܰ௖௥ : Elastic critical force for the relevant buckling mode based on the gross
cross sectional properties.
‫ܮ‬௘௥ : Buckling length in the buckling plane (m)
݅: Radius of gyration about the relevant axis, determined using the properties of
the gross cross-section (m)
Buckling curve a0 a b c d
Imperfection factor α 0,13 0,21 0,34 0,49 0,76
Table 50: Imperfection factors for buckling curves
246
Figure 69: Buckling curves
ே ಶ೏
Buckling effects may be ignored for slenderness ߣ̅ ≤ 0,2 or for ≤ 0,04.
ே ೎ೝ
For a laterally unrestrained member subject to major axis bending should be

verified against lateral-torsional buckling as:
‫ ܯ‬ாௗ
≤ 1.0 (6-51)
‫ ܯ‬௕,ோௗ
Where:
‫ ܯ‬ாௗ : Design value of the moment (Nm)
‫ ܯ‬௕,ோௗ : Design buckling resistance moment (Nm)
The design buckling resistance moment should be considered as:
݂௬
‫ ܯ‬௕,ோௗ = ߯௅் ∙ ܹ ௬ ∙ (6-52)
ߛெ ଵ
Where:
ܹ ௬ : Appropriate section modulus as follows:
ܹ ௬ = ܹ ௣௟,௬ , for Class 1 or 2 cross-sections
247
ܹ ௬ = ܹ ௘௟,௬ , for Class 3 cross-sections
ܹ ௬ = ܹ ௘௙௙,௬ , for Class 4 cross-sections
߯௅் : Reduction factor for lateral-torsional buckling.
ߛெ ଵ : Resistance of members to instability
Estimation of ߯௅் for lateral torsional buckling should be determined as:
1
߯௅் = ≤ 1.0
ߔ ௅் + ଶ
ට ߔ௅் − ߣଶ̅௅் (6-53)
ߔ ௅் = 0,5 ∙ උ1 + ߙ௅் ∙ ൫ߣ̅௅் − 0,2൯+ ߣଶ̅௅் ඏ
ܹ ௬ ∙ ݂௬
ߣ̅௅் = ඨ
‫ ܯ‬௖௥
Where:
ߙ௅் : Imperfection factor (see Table 50)
‫ ܯ‬௖௥ : Elastic critical moment for lateral-torsional buckling

ெ ಶ೏ ଶ
For slenderness ߣ̅௅் ≤ ߣ̅௅்,଴ or for ≤ ߣ̅௅்,଴ lateral torsional buckling effects
ெ ೐ೝ
may be ignored and only cross sectional checks apply.
6.4.1.7 Bending and axial force
In the case of axial force acting on members, its effect on plastic resistance
should be investigated. For class 1 and 2 cross sections, the following criterion
shall be satisfied:
‫ ܯ‬ாௗ ≤ ‫ ܯ‬ே ,ோௗ (6-54)
Where:
‫ ܯ‬ே ,ோௗ : design plastic moment resistance reduced due to the axial force ܰாௗ
For circular cross sections, the criterion applied for bi-axial bending denotes:
248
ଶ ଶ
‫ ܯ‬௬,ாௗ ‫ ܯ‬௭,ாௗ
ቈ ቉ +ቈ ቉ ≤1
‫ ܯ‬ே ,௬,ோௗ ‫ ܯ‬ே ,௭,ோௗ (6-55)
‫ ܯ‬ே ,௬,ோௗ = ‫ ܯ‬ே ,௭,ோௗ = ‫ ܯ‬௣௟,ோௗ ∙ (1 − ݊ଵ,଻)
For Class 3 and class 4 cross sections, in the absence of shear force, the
maximum longitudinal stress shall satisfy the criterion:
݂௬
ߪ௫,ாௗ ≤ (6-56)
ߛெ ଴
Where:
ߪ௫,ாௗ : Design value of the longitudinal stress due to moment and axial force
In the presence of shear, provided that the design value of shear force does not
exceed 50% of the design plastic shear resistance, no reduction to the
resistances defined for bending and axial force should be made. In the opposite
case, a reduction in yield strength should be employed as:
(1 − ߩ) ∙ ݂௬
ଶ
(6-57)
ߩ = ൫2ܸாௗ ⁄ܸ௣௟,ோௗ − 1൯
6.4.1.8 Members in bending and axial compression
For members which are subjected to combined bending and axial compression
should satisfy:
ܰாௗ ‫ ܯ‬௬,ாௗ + ߂ߊ ௬,ாௗ ‫ ܯ‬௭,ாௗ + ߂ߊ ௭,ாௗ

+ ݇௬௬ ∙ + ݇௬௭ ∙ ≤1
߯௬ ∙ ܰோ௞ ‫ ܯ‬௬,ோ௞ ‫ ܯ‬௭,ோ௞
߯௅் ∙ ߛ ߯௅் ∙ ߛ
ߛெ ଵ ெଵ ெଵ
(6-58)
ܰாௗ ‫ ܯ‬௬,ாௗ + ߂ߊ ௬,ாௗ ‫ ܯ‬௭,ாௗ + ߂ߊ ௭,ாௗ
+ ݇௭௬ ∙ + ݇௭௭ ∙ ≤1
߯௭ ∙ ܰோ௞ ‫ ܯ‬௬,ோ௞ ‫ ܯ‬௭,ோ௞
ߛெ ଵ ߯௅் ∙ ߛ ߯௅் ∙ ߛ
ெଵ ெଵ
Where:
ܰாௗ , ‫ ܯ‬௬,ாௗ , ‫ ܯ‬௭,ாௗ : Design values of the compression force and the maximum
moments about the y-y and z-z axis along the member, respectively
249
߂ߊ ௬,ாௗ , ߂ߊ ௭,ாௗ : Moments due to the shift of the centroidal axis (=0 for classes
1, 2, 3; =݁ே ,௬(௭) ܰ௘ௗ for class 4)
߯௬ , ߯௭ : Reduction factors due to flexural buckling
߯௅் : Reduction factor due to lateral torsional buckling
݇௬௬ , ݇௬௭, ݇௭௬ , ݇௭௭ : Interaction factors (App. A and B of standard)
and number of each limit state, as it is referred to in the standard, is included,
as well as the identification number that correspond to this limit state in this
ufEuler Usage factor equal ratio for axial compression / Euler capacity 6-43
ufAxial Usage factor equal ratio for axial load / design resistance for axial loading 6-42
ufTorsion Usage factor due to torsion 6-48
ufShearz Usage factor due to shear in local z direction 6-45
ufSheary Usage factor due to shear in local y direction 6-45
Cross section usage factor according to section 6.2.9 (Bending and Axial Froce) and
ufXSection 6-54/6-56
6.2.10 (Bending Shear and axial Force)
uf646 Usage factor according to equation (6.46) (Buckling Resistance - Compression) 6-49
uf655 Usage factor according to equation (6.55) (Buckling Resistance - Bending) 6-51
uf661 Usage factor according to equation (6.61) (Bending and Axial Compression) 6-58a
uf662 Usage factor according to equation (6.62) (Bending and Axial Compression) 6-58b
Table 51: Limit States according to EN 1993
ufEuler ufAxial ufTorsion ufShearz ufSheary ufXSection uf646 uf655 uf661 uf662
b101 25.00 5.44 22.26 0.00 80.60 0.00 10.54 0.00 7.69 8.18
b102 15.21 6.31 23.19 44.40 44.44 11.21 6.09 0.00 5.16 5.21
b103 15.22 6.32 23.17 44.36 44.45 11.19 6.09 0.00 5.17 5.21
b104 25.01 5.45 22.27 0.00 80.51 0.00 10.53 0.00 7.69 8.18
b201 25.01 5.85 29.45 30.33 53.38 17.81 11.32 0.00 21.35 19.14
b202 18.74 7.69 37.43 56.85 0.00 17.99 7.46 0.00 7.19 7.20
b203 18.72 7.69 37.43 56.78 0.00 17.99 7.45 0.00 7.19 7.20
b204 24.99 5.86 29.50 30.34 53.44 17.77 11.31 0.00 21.34 19.18
250
b301 91.11 10.30 0.00 25.06 37.39 11.57 0.00 0.00 0.00 0.00
b302 22.83 10.67 0.00 53.38 72.05 23.69 10.39 0.00 8.38 8.45
b303 22.86 10.67 0.00 53.39 71.98 23.69 10.39 0.00 8.38 8.44
b304 90.97 10.29 0.00 25.07 37.43 11.59 0.00 0.00 0.00 0.00
b401 0.00 0.00 19.70 31.16 36.07 20.54 0.00 0.00 23.90 24.45
b402 40.06 21.84 19.73 40.55 29.01 21.94 19.20 0.00 11.55 11.58
b403 40.04 21.83 19.76 40.56 29.01 21.97 19.20 0.00 11.54 11.57
b404 0.00 0.00 19.73 31.15 36.04 20.55 0.00 0.00 23.88 24.44
b111 0.00 0.00 0.00 0.00 0.00 68.20 0.00 0.00 0.00 0.00
b112 0.00 0.00 0.00 0.00 0.00 68.20 0.00 0.00 0.00 0.00
b113 0.00 0.00 0.00 0.00 0.00 68.20 0.00 0.00 0.00 0.00
b114 0.00 0.00 0.00 0.00 0.00 68.20 0.00 0.00 0.00 0.00
b115 0.00 0.00 0.00 0.00 0.00 41.17 0.00 0.00 0.00 0.00
b116 0.00 0.00 0.00 0.00 0.00 41.17 0.00 0.00 0.00 0.00
b211 0.00 0.00 80.56 56.72 0.00 20.19 0.00 0.00 0.00 0.00
b212 0.00 13.87 0.00 0.00 33.83 19.49 0.00 0.00 0.00 0.00
b213 0.00 0.00 80.46 56.67 0.00 20.19 0.00 0.00 0.00 0.00
b214 9.25 13.39 0.00 80.26 33.88 46.12 8.30 0.00 10.72 7.78
b215 0.00 0.00 53.41 0.00 56.88 24.63 0.00 0.00 0.00 0.00
b216 0.00 0.00 53.36 0.00 56.83 24.65 0.00 0.00 0.00 0.00
b311 0.00 0.00 0.00 0.00 72.00 13.67 0.00 0.00 0.00 0.00
b312 0.00 16.23 0.00 0.00 30.93 7.00 0.00 0.00 0.00 0.00
b313 0.00 0.00 0.00 0.00 71.97 13.68 0.00 0.00 0.00 0.00
b314 9.54 14.92 0.00 46.27 30.94 8.87 8.74 0.00 9.18 6.81
b315 0.00 0.00 46.35 57.72 72.00 33.86 0.00 0.00 0.00 0.00
b316 0.00 0.00 46.43 57.77 72.07 33.86 0.00 0.00 0.00 0.00
b411 0.00 0.00 0.00 0.00 72.00 10.01 0.00 0.00 0.00 0.00
b412 0.00 21.24 0.00 32.77 22.86 7.65 0.00 0.00 0.00 0.00
b413 0.00 0.00 0.00 0.00 72.01 10.01 0.00 0.00 0.00 0.00
b414 13.15 16.53 0.00 53.24 24.61 7.06 11.94 0.00 4.64 4.18
b415 46.43 0.00 53.36 0.00 20.57 10.29 40.18 0.00 6.26 5.89
b416 46.46 0.00 53.42 0.00 38.81 32.04 40.19 0.00 6.25 5.89
b151 0.00 7.00 96.23 80.22 56.85 10.61 0.00 0.00 0.00 0.00
b152 3.11 6.72 80.52 84.92 80.48 0.00 2.81 0.00 2.53 2.54
b153 8.07 15.07 42.41 0.00 33.87 7.96 7.54 0.00 5.44 5.11
b154 8.07 15.07 42.41 0.00 33.85 7.96 7.53 0.00 5.44 5.11
251
b155 3.11 6.71 80.51 85.04 80.42 0.00 2.81 0.00 2.53 2.54
b156 0.00 7.01 96.18 80.32 56.91 10.60 0.00 0.00 0.00 0.00
b157 12.71 11.91 42.40 0.00 33.88 8.37 11.71 0.00 0.00 0.00
b158 12.70 11.91 42.38 0.00 33.87 8.37 11.72 0.00 0.00 0.00
b251 0.00 6.72 31.86 56.75 71.96 9.18 0.00 0.00 0.00 0.00
b252 3.39 6.61 31.14 0.00 80.47 15.32 3.05 0.00 2.59 2.58
b253 13.51 25.14 50.22 0.00 32.80 8.92 12.64 0.00 7.79 6.94
b254 13.49 25.13 50.14 0.00 32.81 8.92 12.63 0.00 7.78 6.94
b255 3.39 6.61 31.17 0.00 80.47 15.33 3.05 0.00 2.59 2.59
b256 0.00 6.72 31.86 56.75 72.06 9.17 0.00 0.00 0.00 0.00
b257 0.00 35.90 42.46 0.00 30.27 9.20 0.00 0.00 0.00 0.00
b258 0.00 35.90 42.44 0.00 30.33 9.20 0.00 0.00 0.00 0.00
b351 0.00 6.99 28.73 53.31 0.00 9.35 0.00 0.00 0.00 0.00
b352 3.95 6.93 28.77 40.33 0.00 10.55 3.55 0.00 2.90 2.89
b353 18.60 28.23 36.05 0.00 26.36 8.23 16.80 0.00 7.86 6.41
b354 18.56 28.26 36.08 0.00 26.35 8.22 16.81 0.00 7.87 6.41
b355 3.95 6.93 28.76 40.29 0.00 10.56 3.55 0.00 2.90 2.89
b356 0.00 6.99 28.73 53.34 0.00 9.35 0.00 0.00 0.00 0.00
b357 0.00 0.00 36.11 56.80 26.34 8.32 0.00 0.00 11.80 8.26
b358 0.00 0.00 36.03 56.81 26.36 8.32 0.00 0.00 11.79 8.27
b451 0.00 8.11 26.98 40.50 44.43 14.52 0.00 0.00 0.00 0.00
b452 5.02 8.00 28.74 36.07 0.00 11.91 4.57 0.00 3.84 3.82
b453 26.27 40.14 0.00 30.30 20.10 5.83 24.86 0.00 5.22 4.14
b454 26.24 40.10 0.00 30.30 20.09 5.84 24.84 0.00 5.23 4.14
b455 5.02 8.01 28.77 36.04 0.00 11.91 4.57 0.00 3.84 3.82
b456 0.00 8.10 26.97 40.50 44.43 14.52 0.00 0.00 0.00 0.00
b457 0.00 0.00 80.59 53.27 20.76 5.81 0.00 0.00 5.36 4.13
b458 0.00 0.00 80.56 53.34 20.79 5.82 0.00 0.00 5.36 4.13
referring to the leg members, starting from the first elevation, distinguish one set
of members in axial compression and bending (b102-b103) with ߚ = 5.16 and
one set in axial tension (b101-b104) with ߚ = 5.44. Due to the uniform cross
section of the legs, it is expected that the values of ߚ of this elevation will be the
lowest one, sizing the members. Moving to the second elevation, a different
pattern is followed with members (b202-b203) in axial compression and bending
with ߚ = 7.19 having a greater value of reliability index than the second member
252
(b201-b204) subjected in axial tension with ߚ = 5.85. The same pattern as in
the second elevation can be observed in the third elevation with the set of
members (b302-b303) subjected to axial compression and bending with
ߚ = 8.38 and the set of (b301-b304) subjected to axial tension with ߚ = 10.30. A
different behaviour can be observed in the fourth elevation leg members with
the set of members (b402-b403) subjected to axial compression and bending
with ߚ = 11.54 and the set of (b401-b404) subjected to torsion with ߚ = 19.70.
due to the insignificant loads they carry, according to the piling conditions that
have been applied. Members of the first elevation that stand parallel to the flow
(b211-b213) have ߚ = 20.19 due to the cross section usage factor limit state,
while for the members vertical to the flow, the one that meets the flow first
(b214) has a lower value of ߚ = 7.78 due to bending and axial compression and
the second one (b212) has ߚ = 13.87 due to axial tension. The two X-bracing
members (b215-b216) have great values of reliability since they carry minimum
loads, as they have a non functional load to the operation of the structure. In the
third elevation, the two symmetrical members, parallel to the flow (b311-b313)
have ߚ = 13.67 due to the cross section usage factor limit state, the first vertical
to the flow (b314) ߚ = 6.81 due to axial compression and bending, and the
second member (b312) ߚ = 7.00 due to the cross section usage factor limit
state. The members parallel to the flow of the fourth elevation (b411-b413) have
ߚ = 10.01 due to the cross section usage factor limit state, the first member
vertical to the flow (b414) ߚ = 4.18 due to axial compression and bending while
the second member (b412) ߚ = 7.65 due to the cross section usage factor limit
state. At this elevation the X brace members show values of ߚ = 5.89 due to
axial compression and bending criterion, due to the loads transmitted from the
topside loads.
members parallel to the flow, have one of their member (b152-b155) defined
from the axial compression and bending limit state with ߚ = 2.53 and the other
of their members (b151-b156) with ߚ = 7.00 subjected to axial tension. From
253
the sets vertical to the flow, the first (b157-b158) has ߚ = 8.37 due to the cross
section usage factor limit state, while the second one (b153-b154) has ߚ = 5.11
due to axial compression and bending. From the members of the second
elevation that stand parallel to the flow, members (b252-b255) are subjected to
axial compression and bending with ߚ = 2.59 while members (b251-b256) are
subjected in axial tension with ߚ = 6.72. From the sets vertical to the flow, the
first (b257-b258) has ߚ = 9.20 due to the cross section usage factor limit state,
while the second one (b253-b254) has ߚ = 6.94 due to axial compression and
bending. From the members of the third elevation that stand parallel to the flow,
members (b352-b355) are subjected to axial compression and bending with
ߚ = 2.89 while members (b351-b356) are subjected in axial tension with
ߚ = 6.99. From the sets vertical to the flow, the first (b357-b358) has ߚ = 8.26,
while the second one (b353-b354) has ߚ = 6.41 both due to axial compression
and bending. Finally for the members of the fourth elevation, those that stand
parallel to the flow, (b452-b455) are subjected to axial compression and
bending with ߚ = 3.82, while members (b451-b456) are subjected in axial
tension with ߚ = 8.10. From the sets vertical to the flow, the first (b457-b458)
has ߚ = 4.13, while the second one (b453-b454) has ߚ = 4.11 both due to axial
compression and bending.
6.5 ANSI/AISC 360-05: Specification for structural steel

buildings
The design procedure for the design of members under load actions is
described in Chapters A-H of this standard. Axial force and bending (flexure as
it will be referred to herein) about one or both axes, with or without torsion, and
to members subject to torsion only is analytically derived both in a limit state
and in a working state design. In this section, the limit state provisions will be
presented for symmetric members subjected to bending and axial force, since
the members of the structure of reference is of this type. Different sections in
this standard include modified provisions for unsymmetrical members. Finally,
254
provisions for members under combined torsion, flexure, shear and/or axial
force will be presented [4].
The standard defines that design according to its provisions should be satisfied
when ‘the design strength of each structural component equals or exceeds the
required strength determined on the basis of the load combinations considered’.
The typical for LRFD standard basic equation that should be satisfied is:
ܴ௨ ≤ ܴ߮௡ (6-59)
Where:
ܴ௨ : Required strength (MPa)
ܴ௡ : Nominal strength, as specified for each case of loading (MPa)
߮ : Resistance factor, as specified for each case of loading
ܴ߮௡ : Design strength (MPa)
6.5.1.1 Members in tension
For members in tension, the design tensile strength ߮௧ܲ௡ shall be derived by the
lower value of the limit states corresponding to tensile yielding and tensile
rapture:
ܲ௡ = ‫ܨ‬௬ ‫ܣ‬௚ , ߮௧ = 0.90

(6-60)
ܲ௡ = ‫ܨ‬௨ ‫ܣ‬௘, ߮௧ = 0.75
Where:
‫ܣ‬௘ : Effective net area (mm2), calculated as the product of the net area ‫ܣ‬௡ and
the shear lag factor ܷ specified for different connections of tension members.
‫ܣ‬௚ : Gross area of member (mm2)
‫ܨ‬௬ : Specified minimum yield stress (MPa)
‫ܨ‬௨ : Specified minimum tensile stress (MPa)
255
6.5.1.2 Members in compression
For design of members in compression the design compressive strength ߮஼ ܲ௡

shall be determined as the lowest value obtained by the limit states of flexural
buckling, torsional buckling and flexural - torsional buckling. The reduction
factor referring to compression is ߮௧ = 0.90. The nominal compressive strength
ܲ௡ for the case of flexural buckling should be:
ܲ௡ = ‫ܨ‬௖௥‫ܣ‬௚ (6-61)
Where the flexural buckling is determined as:

ಷ೤
‫ܨ‬௖௥ = ቈ0.658 ಷ೐ ቉‫ܨ‬௬ , when ‫ܨ‬௘ ≥ 0.44‫ܨ‬௬
‫ܨ‬௖௥ = 0.877‫ܨ‬௘, when ‫ܨ‬௘ < 0.44‫ܨ‬௬

(6-62)
ߨଶ‫ܧ‬
‫ܨ‬௘ =
‫ ܮܭ‬ଶ
ቀ‫ݎ‬ቁ
Where:
‫ܨ‬௘ : Elastic critical buckling (MPa)
‫ ܮ‬: Laterally un-braced length of member (mm)
‫ݎ‬: Radius of gyration (mm)
‫ ܭ‬: Effective length factor in plane of bending equals 1 unless different analysis
indicates a smaller value
6.5.1.3 Members in Flexure
For the design of flexural strength, ߮஻ ‫ ܯ‬௡ , the corresponding reduction factor
would be ߮௧ = 0.90. The nominal flexural strength ‫ ܯ‬௡ should be determined by
the lower value obtained by the limit states corresponding to yielding and plastic
buckling as follows:
‫ ܯ‬௡ = ‫ ܯ‬௣ = ‫ܨ‬௬ ܼ (6-63)
256
଴.଴ଶଵா
‫ܯ‬௡ = ቆ ವ + ‫ܨ‬௬ ቇܵ, for non-compact sections
೟
଴.ଷଷா
‫ ܯ‬௡ = ‫ܨ‬௖௥ܵ = ವ ܵ, for sections with slender walls
೟
Where:
ܵ : Elastic section modulus (mm3)
ܼ : Plastic section modulus about the axis of bending (mm3)
6.5.1.4 Members in Shear
Design shear strength, ߮௩ܸ௡ should be determined by applying ߮௩ = 0.90 and

for cylindrical cross sections, the nominal shear strength should be determined
according to limit states of shear yielding and shear buckling as:
‫ܨ‬௖௥‫ܣ‬௚
ܸ௡ = (6-64)
2
Where ‫ܨ‬௖௥ should be derived by the maximum of the values:
1.60‫ܧ‬
‫ܨ‬௖௥ = ହ
ට ‫ܮ‬௩ ቀ‫ ܦ‬ቁ
ସ
‫ݐ ܦ‬
(6-65)
0.78‫ܧ‬
‫ܨ‬௖௥ = ଷ
‫ ܦ‬ଶ
ቀ‫ݐ‬ቁ
Where:
‫ܣ‬௚ : Gross area of section based on design wall thickness (mm2)
‫ ܦ‬: Outside diameter (mm)
‫ܮ‬௩ : Distance from maximum to zero shear force (mm)
‫( =ݐ‬0.93 − 1)‫ݐ‬௡ : Design wall thickness (mm)
6.5.1.5 Members subject to Flexure and Axial Force
The fundamental relationship that should constrain the interaction between axial
force and flexure is:
257
௉ ଼ ெ ெ ೝ೤ ௉
ܷ௠ = ௉ೝ + ଽ ∙ ൬ெ ೝೣ + ெ ൰ ≤ 1.0, For ௉ೝ ≥ 0.2
೎ ೎ೣ ೎೤ ೎
(6-66)
௉ೝ ଼ ெ ೝೣ ெ ೝ೤ ௉ೝ
ܷ௠ = ଶ௉ + ଽ ∙ ൬ெ + ெ ൰ ≤ 1.0, For ௉ < 0.2
೎ ೎ೣ ೎೤ ೎
Where:
ܲ௥ : Required axial compressive (tensile) strength, (N)
ܲ௖ = ߮஼ ܲ௡ : Available axial compressive strength, (N)
ܲ௖ = ߮௧ܲ௡ : Available axial tensile strength, (N)
߮஼ = 0.90 : Resistance factor for compression
߮௧ : Resistance factor for tension (as described above)
߮௕ = 0.90 : Resistance factor for flexure
‫ ܯ‬௥ : Required flexural strength (N-mm)
‫ ܯ‬௖ = ߮஻ ‫ ܯ‬௡ : Available flexural strength (N-mm)
‫ ݔ‬: Subscript relating symbol to strong axis bending
‫ ݕ‬: Subscript relating symbol to weak axis bending
6.5.1.6 Members under torsion and combined torsion, shear and/or axial
force
The design torsional strength, ߮ ் ܶ௡ should consider ߮ ் = 0.90 and nominal

torsional strength:
ܶ௡ = ‫ܨ‬௖௥ ∙ ‫ܥ‬ (6-67)
Where:
గ∙(஽ି௧)మ∙௧
‫=ܥ‬ : Torsional constant
ଶ
‫ܨ‬௖௥ : The greatest of (without exceeding 0.6‫ܨ‬௬ ):
1.23‫ܧ‬
‫ܨ‬௖௥ = ହ
(6-68)
ට ‫ ∙ ܮ‬ቀ‫ ܦ‬ቁ
ସ
‫ݐ ܦ‬
258
0.60‫ܧ‬
‫ܨ‬௖௥ = ଷ
‫ ܦ‬ଶ
ቀ‫ݐ‬ቁ
Where:
‫ ܮ‬: Length of the member (mm)
‫ ܦ‬: Outside diameter (mm)
For cylindrical members subjected to combined torsion, shear, flexure and axial
force: when ܶ௥ ≤ 20, provisions of the fundamental relationship between axial
force and flexure should be followed. In the case when ܶ௥ > 20, interaction of
torsion, shear, flexure and/or axial force shall be limited by:
ܲ௥ ‫ ܯ‬௥ ܸ௥ ܶ௥ ଶ
൬ + ൰+ ൬ + ൰ ≤ 1.0 (6-69)
ܲ௖ ‫ ܯ‬௖ ܸ௖ ܶ௖
Where:
ܲ௥ : Required axial strength (N)
ܲ௖ = ߮ ∙ ܲ௡ , design tensile or compressive strength (N)
‫ ܯ‬௥ : Required flexural strength (N-mm)
‫ ܯ‬௖ = ߮௕ ∙ ‫ ܯ‬௡ , Design flexural strength (N-mm)
ܸ௥ : Required shear strength (N)
ܸ௡ = ߮௩ ∙ ܸ௡ : Design shear strength (N)
ܶ௥ : Required torsional strength (N-mm)
ܶ௖ = ߮ ் ∙ ܶ௡ Design torsional strength (N-mm)
6.5.2 Numerical Application
reliability assessment of the reference structure. For this standard, the final
value of the reliability index for each member is determined by dominant limit
state referring to symmetrical members subjected to bending and axial force.
Therefore this case will describe the effect of any action or combination of
259
actions. Following, in Table 56 to Table 58, the results of reliability analysis of
ufH1 Usage factor according to section H1 6-66
Table 55: Limit States according to ANSI/AISC 360-05
ufH1 ufH1
b101 4.64 b301 9.78
b102 4.25 b302 8.65
b103 4.26 b303 8.65
b104 4.64 b304 9.79
b201 7.62 b401 14.91
b202 5.47 b402 10.90
b203 5.47 b403 10.92
b204 7.63 b404 14.88
ufH1 ufH1
b111 61.99 b311 10.26
b112 61.99 b312 8.90
b113 61.99 b313 10.26
b114 61.99 b314 6.06
b115 37.69 b315 17.11
b116 37.69 b316 17.11
b211 10.99 b411 9.01
b212 10.58 b412 6.30
b213 10.97 b413 8.99
b214 7.54 b414 5.42
b215 15.82 b415 8.59
b216 15.81 b416 8.59
ufH1 ufH1
b151 4.90 b351 4.85
b152 2.07 b352 2.64
b153 4.40 b353 5.80
b154 4.40 b354 5.80
b155 2.06 b355 2.64
b156 4.90 b356 4.85
b157 7.52 b357 6.27
b158 7.52 b358 6.28
260
ufH1 ufH1
b251 4.79 b451 7.43
b252 2.26 b452 3.50
b253 7.40 b453 3.82
b254 7.39 b454 3.82
b255 2.26 b455 3.50
b256 4.79 b456 7.43
b257 8.58 b457 3.88
b258 8.58 b458 3.88
Investigating results for the leg members, starting from the first elevation,
distinguish one set of members (b102-b103) with ߚ = 4.25 and one set (b101-
b104) with ߚ = 4.64. Due to the uniform cross section of the legs, it is expected
that the values of ߚ of this elevation will be the lowest one, sizing the members.
Moving to the second elevation, members (b202-b203) have ߚ = 5.47 while
(b201-b204) have ߚ = 7.63. In the third elevation, members (b302-b303) have
ߚ = 8.65 and the set of (b301-b304) has ߚ = 9.78. For members in the fourth
elevation, the set of (b402-b403) is found to have ߚ = 10.90 and the set of
(b401-b404) with ߚ = 14.88.
due to the insignificant loads they carry, according to the piling conditions that
have been applied. Members of the second elevation that stand parallel to the
flow (b211-b213) have ߚ = 10.99, while for the members vertical to the flow, the
one that meets the flow first (b214) has a lower value of ߚ = 7.54 and the
second one (b212) has ߚ = 10.58. The two X-bracing members (b215-b216)
have great values of reliability since they carry minimum loads, as they have a
non functional load to the operation of the structures. In the third elevation, the
two symmetrical members, parallel to the flow (b311-b313) have ߚ = 10.26, the
first vertical to the flow (b314) ߚ = 6.06, and the second member (b312)
ߚ = 8.90. The members parallel to the flow of the fourth elevation (b411-b413)
have ߚ = 9.01, the first member vertical to the flow (b414) ߚ = 5.42 while the
second member (b412) ߚ = 6.30. At this elevation the X brace members show
values of ߚ = 8.59, due to the loads transmitted from the topside loads.
261
members parallel to the flow, have one of their members (b152-b155) with
ߚ = 2.07 and the other of their members (b151-b156) with ߚ = 4.90. From the
sets vertical to the flow, the first (b157-b158) has ߚ = 7.52, while the second
one (b153-b154) has ߚ = 4.40. From the members of the second elevation that
stand parallel to the flow, members (b252-b255) have ߚ = 2.26 while members
(b251-b256) have ߚ = 4.79. From the sets vertical to the flow, the first (b257-
b258) has ߚ = 8.58, while the second one (b253-b254) has ߚ = 7.40. From the
members of the third elevation that stand parallel to the flow, members (b352-
b355) have ߚ = 2.64 while members (b351-b356) have ߚ = 4.85. From the sets
vertical to the flow, the first (b357-b358) has ߚ = 6.27, while the second one
(b353-b354) has ߚ = 5.80. Finally for the members of the fourth elevation, those
that stand parallel to the flow, (b452-b455) have ߚ = 3.50 while members
(b451-b456) have ߚ = 7.43. From the sets vertical to the flow, the first (b457-
b458) has ߚ = 3.88, while the second one (b453-b454) has ߚ = 3.82.
6.6 Discussion
Table 59 to Table 61, summarize the minimum values of reliability indices

based on different limit states that have been investigated based on provisions
of different standards as well as values obtained by the analytical limit state
derived in the previous Chapter (based on the von Mises failure criterion).
Although in general similar trends of figures can be observed, differences in
individual members are found due to the different design considerations that
each standard suggests. The basic difference in those design provisions relates
to the reduction (or resistance) factors that are used in each of the potential
load actions, and are listed in Table 62. EN and AISC, propose more
conservative reduction factors for members under tension (0.8 and 0.9-0.75
respectively), while API LRFD and ISO propose more conservative values for
members in compression (0.85 for both). API LRFD and ISO have propose the
same values for bending, shear and torsion (0.95), and for hoop buckling (0.80).
262
EN’s values for bending, shear and torsion is 1.00 while AISC/ANSI proposes a
value equals to 0.90. Work on the derivation of the strength reduction factors
has been performed in the background documents of the API standards [310].
For every case of action, von Mises criterion has not accounted for any
reduction factor. Further, differences in the expressions for combinations of
actions in standards provide different results, distinguishing from a simple
summation, vectorial summation or weighted summation of action resultants. It
should also be noted here that application of EN and ANSI/AISC is not directly
applicable to offshore structure since they are generic documents for steel
structures and do not have special clauses for actions such as external
pressure on members etc; however useful conclusions can be derived by
including them in this analysis.
For leg members, difference in the pattern of minimum value of reliability is

observed between members in axial tension and compression, combined with
bending, based on reduction of members’ strengths. The lowest elevation
members, which will eventually size the rest of the member’ parts, based on von
Mises criterion, indicate a minimum value of ߚ = 4.79 based on member in
tension, while API and ISO derive values of 4.76 and 4.71 respectively based on
members in compression. Respective values derived by EN and AISC/ANSI are
5.16 and 4.25 based on members in compression. For the second elevation leg
members, von Mises criteria follow the same trend, dominated by members in
tension, with a minimum ߚ = 5.74. The same trend is followed by ISO (6.05), EN
(5.85) and AISC/ANSI (5.47), while API LRFD members are characterized by
their members in axial compression (6.62). For the members of the third
elevation, all standards (8.15, 8.38, 8.65 respectively) and the von Mises limit
state (8.15), are dominated by their members in axial compression, apart from
ISO (7.17) where axial compression prevails. Finally, for the fourth elevation
both the von Mises limit states (10.66) and the standards minimum limit states
(API LRFD: 11.65, ISO: 11.05, EN: 11.54 AISC/ANSI: 10.90) are determined by
their members in axial compression.
263
For the horizontal brace members of the first elevation, all standards agree in
great values of ߚ; those values imply zero value of probability of failure and
therefore great numerical differences are insignificant. From the six members of
the horizontal bracing of the second elevation, the most sever effect is for all
cases derived by the first member located vertical to the flow (API LRFD: 6.50,
ISO: 6.43, EN: 7.78, AISC/ANSI: 7.54, v.Mises: 9.33) suffering from axial
tension. X braces show great values of ߚ due to the lower acting stresses. The
members of the third elevation are dominated by the ones subjected to axial
compression and bending while members of the fourth elevation are dominated
by the ones subjected to axial tension and bending (el3: API LRFD: 6.14, ISO:
6.02, EN: 6.81, AISC/ANSI:6.06, v.Mises: 6.94 / el4: API LRFD: 4.29, ISO: 4.28,
EN: 4.18, AISC/ANSI: 5.42, v.Mises: 5.76). X bracing members on the fourth
elevation, affected by the topside loads, have a lower value or ߚ based on EN
(5.89) while other values derived are (API LRFD: 6.63, ISO: 9.62,
AISC/ANSI:8.41, v.Mises: 8.37).
Vertical X brace members present lower values of reliability index, illustrating

greater differences between standards. For the members of the first elevation,
AISC/ANSI provides the most conservative values of ߚ, apart from the case of
the set of members located vertical to the flow where ISO provides more
conservative values. All values derived, agree that the most critical members
are those subjected in axial compression and bending (API LRFD: 6.10, ISO:
2.20, EN: 2.53, AISC/ANSI:2.06, v.Mises: 5.55). The same pattern is followed for
the second and third elevations where ISO and AISC/ANSI provide the most
conservative results based on the members parallel to the flow, subjected in
axial compression and bending (el3: API LRFD: 6.00, ISO: 2.24, EN: 2.59,
AISC/ANSI:2.26, v.Mises: 5.25 / el4: API LRFD: 6.29, ISO: 2.65, EN: 2.89,
AISC/ANSI: 2.64, v.Mises: 5.33). Finally, the most critical members of the fourth
elevations, are the second vertical to the flow (API LRFD: 3.02, ISO: 2.83, EN:
4.14, AISC/ANSI:3.82, v.Mises: 3.79).
The above analysis concludes that a global characterization between the most
or least conservative standard cannot be made since different standards are
264
more conservative in different classes of members than others. The general
comment that can be concluded from the values obtained is that the ISO and
AISC/ANSI standards provide more conservative results than API LRFD and
EN respectively. As far as the von Mises criterion is concerned, in most of the
cases, the values derived lay between minimum and maximum of the values
derived by design standards, therefore providing a good initial estimate of ߚ.
The benefit of the use of this limit state is that it can account universally for
different actions and combinations of actions of members, allowing simulation of
the response with any non specialized software.
From the application of the derived methodology in the analysis above, both
considering the global von Mises criterion, as well as the different limit states
proposed by each standard, have proven the efficiency of this method to rapidly
calculate reliability in structural design. Further, it can be concluded that the
implied reliability of different standards can be achieved by procedures that do
not necessarily follow their provisions. The procedure that has been derived can
be also applied for the evaluation of the minimum levels of reliability that is
achieved, by assessing the reliability of structures that are designed in a way
that the utilization factor is lower, but as close as possible to unity. This
comparison which could benchmark the performance of design standards,
would verify that certain standards’ provisions are followed, once the provisions
of a different less conservative standards are met.
vMises API ISO EN AISC vMises API ISO EN AISC

b101 4.79 5.97 5.41 5.44 4.64 b301 8.71 9.56 7.17 10.30 9.78
b102 5.22 4.76 4.71 5.16 4.25 b302 8.15 7.75 7.75 8.38 8.65
b103 5.23 4.76 4.71 5.17 4.26 b303 8.14 7.76 7.75 8.38 8.65
b104 4.79 5.97 5.41 5.45 4.64 b304 8.72 9.56 7.17 10.29 9.79
b201 5.74 7.06 6.05 5.85 7.62 b401 14.73 11.74 15.00 15.11 13.14
b202 7.46 6.62 6.56 7.19 5.47 b402 10.66 11.65 11.05 11.55 10.90
b203 7.45 6.62 6.56 7.19 5.47 b403 10.66 11.65 11.05 11.54 10.92
b204 5.74 7.04 6.05 5.86 7.63 b404 14.72 11.73 14.98 15.10 13.14
Table 59: Minimum Reliability indices for Leg Members
265
b111 54.83 59.43 63.25 68.20 61.99 b311 10.46 11.70 11.48 13.67 10.19
b112 54.83 59.43 63.25 68.20 61.99 b312 6.96 8.75 7.57 7.00 8.90
b113 54.83 59.43 63.25 68.20 61.99 b313 10.48 11.69 11.48 13.68 10.26
b114 54.83 59.43 63.25 68.20 61.99 b314 6.94 6.14 6.02 6.81 6.06
b115 35.01 38.75 37.50 41.17 37.69 b315 20.86 35.53 19.59 33.86 15.39
b116 35.01 38.75 37.50 41.17 37.69 b316 20.87 35.58 19.59 33.86 15.38
b211 12.34 13.84 13.77 20.19 10.86 b411 8.98 10.19 10.31 10.01 9.01
b212 9.70 13.24 9.90 13.87 10.58 b412 6.00 7.00 6.75 7.65 6.30
b213 12.33 13.84 13.80 20.19 10.87 b413 8.98 10.18 10.30 10.01 8.99
b214 9.33 6.50 6.43 7.78 7.54 b414 5.76 4.29 4.28 4.18 5.42
b215 18.72 53.05 20.16 24.63 15.47 b415 8.36 6.64 9.63 5.89 8.40
b216 18.71 53.45 20.17 24.65 15.49 b416 8.37 6.63 9.62 5.89 8.41
Table 60: Minimum Reliability indices for Horizontal Brace Members

b151 5.70 8.26 5.67 7.00 4.90 b351 5.30 7.10 5.41 6.99 4.85
b152 5.54 6.11 2.20 2.53 2.07 b352 5.33 6.29 2.65 2.89 2.64
b153 6.93 8.61 5.10 5.11 4.40 b353 6.08 7.33 6.36 6.41 5.80
b154 6.94 8.62 5.09 5.11 4.40 b354 6.09 7.33 6.35 6.41 5.80
b155 5.55 6.10 2.20 2.53 2.06 b355 5.33 6.30 2.65 2.89 2.64
b156 5.70 8.26 5.66 7.01 4.90 b356 5.31 7.11 5.40 6.99 4.85
b157 6.76 7.41 6.08 8.37 7.52 b357 6.16 9.25 4.71 7.35 6.27
b158 6.74 7.40 6.08 8.37 7.52 b358 6.16 9.25 4.71 7.35 6.28
b251 5.27 7.01 5.43 6.72 4.79 b451 6.37 6.68 6.51 8.11 7.43
b252 5.25 6.00 2.24 2.58 2.26 b452 6.33 4.56 3.87 3.82 3.50
b253 7.51 8.99 6.76 6.94 7.40 b453 3.79 3.03 2.82 4.14 3.82
b254 7.51 8.99 6.76 6.94 7.39 b454 3.79 3.02 2.83 4.14 3.82
b255 5.26 6.00 2.24 2.59 2.26 b455 6.34 4.56 3.87 3.82 3.50
b256 5.27 7.01 5.44 6.72 4.79 b456 6.37 6.67 6.52 8.10 7.43
b257 7.81 12.35 6.31 9.20 8.47 b457 3.72 4.10 4.38 4.13 3.84
b258 7.79 12.35 6.32 9.20 8.47 b458 3.72 4.10 4.39 4.13 3.84
Table 61: Minimum Reliability indices for Vertical Brace Members
API ISO EN AISC

Tension 0.95 1/1.05 1/1.25 0.9/0.75
Compression 0.85 1/1.18 1 0.9
Bending 0.95 1/1.05 1 0.9
Shear 0.95 1/1.05 1 0.9
Torsion 0.95 1/1.05 1 0.9
Hoop Buckling 0.8 1/1.25
Table 62: Reduction factors of different standards
266
6.7 Summary
This Chapter has performed a comparison of the ultimate strength reliability

performance of structural members, based on relevant procedures proposed by
design standards. Based on a consistent methodology, the minimum values of
reliability indices have been derived, giving the opportunity for direct
comparison of the obtained results. This study has shown that, although design
standards meet minimum levels of reliability, they do not perform in a uniform
way, in order to classify one of the standards as the more or least conservative.
Further, the von Mises criterion that has been derived earlier, seems to provide
an accurate estimate of ߚ, accounting for multiple actions on members.
267
7 CONCLUSIONS AND RECCOMENTATIONS
This final chapter will summarize what has been presented in the previous
chapters of this Thesis and will identify the contribution of the research that has
been carried out. The main aim of this PhD has been to study the design of
offshore structures through a stochastic perspective and provide a generic
methodology that will efficiently assess reliability levels of structural members,
allowing combination and integration with available design methods and
numerical tools/procedures. Application of the methodology on a hypothetical
jacket structure, deployed in a theoretical site, has provided useful results and
conclusions. Finally, some recommendations for future work will illustrate
aspects of potential improvements and extensions of the method established,
as well as different applications that this method might be related to.
7.1 Summary
The First Chapter of this Thesis has presented the background of structural
reliability and has set the context of structural safety. Through an initial
presentation of the evolution of design methods, development of design
standards was presented. With a view to performance-based-design, which is
freely followed by modern design standards, target reliability requirements are
introduced based on commonly accepted sources, in order to set the threshold
which the reliability performance of structures should aim to meet.
The Second Chapter has presented a review of reliability assessment of steel

and offshore structures, providing fundamental formulations of reliability
calculation and analyzing important decisions for comprehensive reliability
analysis. Selection of the appropriate type of response analysis and integration
from component to system reliability has been discussed. Finally, an initial
review of the Stochastic Methods for reliability assessment will provide the
background for the development of the methodology that will be applied later in
this Thesis.
269
The Third Chapter introduced the numerical procedures for the computation of
structural reliability. First and Second Order Reliability Methods (FORM/SORM)
as well as Simulation Methods are presented, setting the background of the
codes that have been developed for the scope of this Thesis, and are followed
by the established tools that have been used in order to verify the different
steps of the methodology that is proposed. A review of the development of the
Stochastic Response Surface Method (SRSM) and the methodology that has
been applied in the reference application has been analytically discussed.
Regression techniques have been included as fundamentals of the regression
analysis that has been applied in the SRSM, and a variation of the conventional
least square method (LSM) has been proposed for more efficient regression
and prevention form ill conditioned systems of equations. Verification of the
codes for the FEA simulation, the multivariate polynomial regression, and the
FORM/SORM estimation routine that have been developed, based on
commercially available software, have shown that the codes written provide
sufficiently accurate results and can be later used in the analysis. Validation of
the SRSM method, based on results obtained by direct simulations for a
reference structure, have given proof that the SRSM methodology provides
accurate results, and can be confidently used for the later application chapters.
In the Fourth Chapter, the environmental loading of offshore structures has

been presented, considering modelling of wave loads according to different
wave theories and a procedure for the correlation between significant wave
height and peak spectral period based on the joint distribution of statistical
distributions. Modelling of wind and current loading have also been covered.
Further, modelling of capacity of offshore structures has been included, through
an extensive review of literature data for material properties and methodologies
for the consideration of capacity deterioration due to corrosion.
The Fifth chapter presents the assessment of the reliability of structural tubular
members of a typical offshore structure based on ultimate strength under
stochastically represented loads and capacity properties. Analytical derivation of
limit states based on yield failure criteria has been presented for ductile
materials providing the limit state according to which reliability of the structure
270
has been evaluated. An application of the procedure for the correlation between
the significant wave height and peak spectral period for the calculation of the
joint probability distribution function has been also presented. A sensitivity
analysis of the main variables and different models of corrosion has been
included. Finally, integration from component to system reliability, based on the
reliability performance of the structural members and the failure mechanism of
the structure, has been discussed.
Chapter Six, introduces the design requirements of the most widely used
standards for the design of offshore structures API RP 2A [1], [2], and ISO
19902 [3]. Further, ANSI/AISC 360-05 [4], and EUROCODE 3 [5], have been
included as generic codes of the design of steel structures with strong
probabilistic background. Based on their provisions, limit states were formulated
based on different actions or combinations of actions that members are
subjected to. For the same reference case structure, the minimum reliability
index of each member according to those standards was calculated, drawing
conclusions from the results obtained.
7.2 Contribution of this PhD
Design of offshore and steel structures is currently based on provisions of

relevant design standards. This practice can achieve compliance to minimum
levels of reliability; however their generalized character imposes a degree of
conservatism, often overestimating members and structures. Modern standards
derive safety factors based on probabilistic and statistical procedures, therefore
having a reliability based background, but express them deterministically
restricting potential of optimization. This yields for a procedure to be developed
allowing validation of partial factors based on the fundamental requirement of
target reliability. Verification and calibration of design standards is mainly based
on simplistic models considering fewer variables, or even statistical
approximations of resistance and loading performance for a specific set of
loads, representing them with distributions that can explicitly be transformed to
values of reliability index. This method, although significantly easier to
271
implement, lacks flexibility and, especially moving towards limit state formats of
standards, should be expressed with a stronger mathematical formulation.
The contribution of this PhD research can be summarized in the following:
 The methodology that is proposed allows reliability assessment of

structural components through a sequence of individual steps that permit
calculation of reliability, based on easy to program procedures. After
execution of a finite series of simulations, the response of each member
can be identified and a quadratic polynomial response surface based on
the values of the limit states that are examined through data regression
analysis can be formulated. Later, a separate routine can account for the
estimation of the reliability index based on one of the available numerical
techniques. Due to the linear nature of the response surfaces of the
problem investigated, deterministic methods are proven to calculate
adequately values of reliability indices. This methodology was proven to
be efficient, allowing accurate and rapid calculation of reliability.
Those discrete steps, which may employ different tools and procedures
for each task, can handle several problems that are difficult to be
modelled in one unified simulation code. For the problem of the design of
offshore structures that has been studied, reliability analysis took place
using the specialized software DNV SESAM for a confident
representation of the response of structural members through
appropriate modelling of the environmental loads acting on the structure.
Following this methodology, specialized commercial tools for different
applications may be employed for the probabilistic assessment of several
engineering problems.
 Due to combined actions on structural members, different limit states

should be examined in order to derive the minimum reliability index of a
structural member. Analytical derivation of limit states based on widely
accepted failure criteria can account for combinations of actions on a
structural member incorporating multiple limit states into one. This
facilitates representation of the response of structural members since
272
principal stresses or even equivalent von Mises stresses are easy to be
obtained through available commercial tools. Application of such limits
states in this analysis has provided results that can be compared with the
ones derived from analytical limit states following guidelines of design
standards, suggesting a simplified methodology for the consideration of
limit states. Incorporation of a more analytical or stochastic
representation of members’ strength under different actions (tension,
compression, bending, etc.) or employment of resistance factors for the
correction of the stochastic distribution of yield strength could provide
even more accurate results for the calculation of reliability.
 Based on the von Mises limit state that has been analytically derived, a
reference structure has been introduced and based on stochastic
representation of the basic design variables (wave, wind, current, yield
strength), the reliability indices of each member has been calculated for
two different angles of load application (0 and 45 degrees) in order to
obtain the minimum value. Further, a buckling limit state has been
introduced and applied for different values of buckling coefficient. For the
base case of the 0 degrees angle of approach, and based on the
consistent methodology that has been developed, a sensitivity analysis
was executed benchmarking the effect of variation of loads magnitude on
the structure. This analysis has illustrated that for the load cases and
materials studied, the wave load and material yield strength are the most
important factors in the formulation of the reliability index of the structure,
while the current and wind loads play an insignificant role. The effect of
accurate modelling of statistical distributions was also examined and
found to affect the derived results. A separate analysis has shown that
roughness of the surface of members has a considerable effect on the
reliability index of the structure, contributing to the increase of the effect
of environmental loads on members as the structure ages. Different
wave theories have also been examined illustrating that appropriate
selection of the most suitable theory can provide different results. Finally,
capacity deterioration of structural members has been studied for
273
different corrosion models and the corresponding variation of the
reliability index has been illustrated.
 From the procedure that has been established, different limit states were
formulated according to different provisions of widely used design
standards for different potential actions and combinations of actions on
members. API RP 2A, ISO 19902, ANSI/AISC 360-05 and EUROCODE
3 have been studied. For the base case of the reference structure that
has been considered, reliability indices for all different limit states
proposed in each standards’ have been calculated, obtaining the
minimum value of reliability index for each member and each standard.
This has allowed comparison of standards performance, illustrating that
ISO and AISC/ANSI provide more conservative results than API LRFD
and EN. A different conclusion of this study illustrates that although
standards meet relative levels of reliability, they do not perform uniformly
among different classes of members.
 This Thesis illustrates a different approach on the use of design

standards. Towards performance-based-design, where values of target
reliability suggest the main design constraint, this methodology can use
fundamental expressions of standards that incorporate scientific
evidence as well as previous experience information for the calculation of
reliability. Therefore, instead of designing for compliance to maximum
permitted strength requirements, design could aim to achieve additionally
minimum levels of reliability. This approach will provide a better
understanding of the actual structural performance. Further, as it can be
derived from the results of the analysis, transferring from a deterministic
analysis that considers maximum strength to a probabilistic-stochastic
that characterizes members’ performance through reliability indices,
different members are found to be the critical ones. Figure 70, Figure 71
and Table 63, present the stress distribution and the reliability indices of
the reference structure based on the von Mises limit state function,
illustrating different sequence of critical members.
274
5
Stress Distribution x 10
1.5
0.5
Figure 70: Stress distribution
Reliability Distribution
10
Figure 71: Reliability distribution
275
ࢼ vMises ࢼ vMises
(MPa) (MPa)
1 b457 3.72 b458 230278 11 b251 5.27 b414 132287
2 b458 3.72 b457 230277 12 b256 5.27 b356 129416
3 b453 3.79 b453 219414 13 b351 5.30 b351 129415
4 b454 3.79 b454 219414 14 b356 5.31 b251 128804
5 b101 4.79 b102 140430 15 b352 5.33 b256 128804
6 b104 4.79 b103 140430 16 b355 5.33 b412 125771
7 b102 5.22 b252 136741 17 b152 5.54 b357 122132
8 b103 5.23 b255 136741 18 b155 5.55 b358 122132
9 b252 5.25 b352 136389 19 b151 5.70 b152 118106
10 b255 5.26 b355 136389 20 b156 5.70 b155 118106
Table 63: Critical Members
From the above results, excluding the top elevation braces where
extreme values are obtained due to the topside loads, the next elements
in the list follow a different pattern when comparing reliability indices
rather than maximum stresses. This fact implies that throughout the
service life of the structure, the members that will require maintenance
are the ones with lower levels of reliability rather than those that suffer
from greater stresses.
 From the results that have been derived, both using the von Mises limit
state as well as the analytical limit states dictated by the provisions of
design standards, are found to follow a similar trend. A different
conclusion that can be drawn is that the implied reliability of each
standard can be met by procedures that do not necessarily follow their
provisions. The level of minimum reliability that standards aim to achieve
can be estimated using the method described and has been followed in
Chapter 6, by assessing the reliability of structures that are designed in a
way that the utilization factor is lower, but as close as possible to unity.
This comparison, which could benchmark the performance of design
standards, would verify that certain standards’ provisions are followed,
once the provisions of a different but less conservative standard are met.
276
7.3 Future recommendations:
This study has derived some useful conclusions that might initiate further work
on several aspects. Potential topics of interest might be found among the
following:
 Although deterministic methods were found to provide efficient

calculation for the type of limit state functions that were considered in this
application, simulation methods can be also included, allowing handling
of more complicated types of statistical distributions where the
cumulative distribution functions cannot be directly derived as well as
consideration of more complicated limit state functions, including highly
correlated stochastic variables. Subset simulation is a method that could
provide adequate approximation of the reliability index of members
having relatively low computational requirements.
 For the reliability assessment that took place, deterministic values of the
reduction factors of the different actions that members are subjected to
have been used. Those values account for different sources of
uncertainty due to geometrical and structural imperfections, residual
stresses etc. Application of those values on the representative value of
yield strength, in order to derive the design value of resistance of each
action, which is stochastically modelled, transforms the distribution
properties of strength for each different action. Following a fully
probabilistic approach, different statistical distributions should be
considered for different actions adequately describing strength for each
action, as is described in relevant clauses regarding “Design assisted by
testing”.
 Based on the derivation of the limit state functions that were used in this
Thesis, relevant limit states can be derived for brittle materials,
corresponding to appropriate fracture criteria. This practice will allow
assessment of various problems including the continuously evolving
application of composite materials. The special performance of those
277
materials that combine brittle and ductile behaviour before failure can be
adequately assessed in the absence of specific design rules.
 Incorporation of the methodology in one unified code for a specific

application (eg. steel structures) will allow automated iterations of
required simulations for the reliability assessment of members. This will
allow consideration of numerous stochastic variables providing even
more accurate results. Further, combination of such code in a multi-
disciplinary optimization routine (MDO), can automate a procedure that
will lead to a robust design of an optimized structure compromising, but
achieving, minimum acceptable levels of target reliability for the lowest
possible weight. Although the mathematic skills for such an endeavour
are high, the tools and procedures required are already available
constituting incorporation in an iterative design routine feasible. Figure
72, presents a block diagram of the reliability based design approach
concept in structural design. This is an approach that provides significant
advantages for the design of novel and special structures or the
verification of structures in new design environments where data from
previous experience are not available.
 Combining a methodology that allows efficient calculation of reliability of

structural members with techniques that provide real-time data of a
structure (eg. thickness deterioration, surface condition, etc), its reliability
evaluation can be updated for any given time instance. This practice
would provide a more accurate understanding of the actual performance
of the structure, allowing modification of its maintenance schedule,
extending or reducing the intervals between interventions as indicated,
and restoring its safety levels.
 Finally, design standards focus their analysis in a component level,

implying that the minimum system reliability will equal that of the
minimum value of the reliability of the sum of the components. This fact
restricts application of techniques of structural design such as structural
redundancy through alternative load paths. In such cases, the acceptable
278
reliability requirements in a component level may be considered lower
than in the case where the whole structure is considered failed, following
failure of the first member. Therefore, methodical investigation of
structural reliability in a system level can show potential benefits, as it
may prove to have a better reliability performance for the same total cost
of the structure.
CONCEPTUAL DESIGN
DEFINITION OF STRUCTURAL
PROPERTIES
DEFINE TARGET RELIABILITY

AND FAILURE MODES
DEFINITION OF LOADS PROCESSING OF DATA
MODELING OF STRUCTURE
RESISTANCE REQUIREMENTS
DESIGN OPTIMISATION
RESULTS EVALUATION
DESIGN VERIFICATION
DESIGN CERTIFICATION
Figure 72: Reliability Based Optimization
279
REFERENCES
[1] American Petroleum Insitute. Recommended Practice for Planning,

Designing and Constructing Fixed Offshore Platforms - Working Stress Design.
01-Dec-2000. API RP 2A-WSD.
[2] American Petroleum Institute. Recommended Practice for Planning,

Designing and Constructing Fixed Offshore Platforms - Load and Resistance
Factor Design . 01-Jul-1993 . API RP 2A-LRFD.
[3] International Standardization Organization. Petroleum and natural gas

industries-general requirements for offshore structures. 2002. ISO 19000:2002.
[4] American Institute of Steel Construction. Specification for structural steel

buildings. March 2005. ANSI/AISC 360-05.
[5] Comité Européen de Normalisation. Eurocode 3: Design of Steel

Structures. 2005. CEN 1993-1-1:2005 .
[6] International Standardization Organization. General principles on

reliability for structures. 2008. ISO 2394:2008.
[7] Comité Européen de Normalisation. Basis of Structural Design. 2002.

CEN Eurocode 1990.
[8] Det Norske Veritas. Guideline for Offshore Structural Reliability Analysis.
Norway : DNV, 1995. DNV Report No. 95-2018.
[9] T. Bayes. An essey towards solving a problem in the doctrine of chance.

Philosophical Transaction of the Royal Social Society, 1763.
[10] D. V. Lindley. Making Decisions. New York : Wiley-interscience, 1985.
[11] B. D. Finetti. Theory of Probability. New York : Wiley, 1974.
[12] S. Kaplan, and B. J. Garrick. On the quantitative definition of Risk. Risk

Analysis. 1981, Vol. 1.
281
[13] National Physics Laboratory. Manual of Codes of Practice for the
Determination of Uncertainties in Mechanical Tests on Metallic Materials. UK-
EU, 2000.
[14] Det Norske Veritas. Structural Reliability Analusis of Marine Structures.

Norway. Classification No. 30.6. July 1992
[15] T. Onoufriou, Reliability based inspection optimization planning for

offshore structures. Journal of Marine Structures. Vol. 12. 521-539. 1999.
[16] G. Sigurdsson, R. Skjong, B. Skallerud, and J. Amdahl. Probabilistic

Collapse Analysis of Jackets. In: Schuëller, Shinozuka and Yao. Structural
Safety and Reliability. Rotterdam: Balkema. 1994.
[17] P. Thoft-Christensen, and Y. Murotsy. Application of Structural Systems

Reliability Theory. Berlin. Springer-Verlag. 1986.
[18] M. Holicky, and T. Vrouwenvelder. “Chapter I-Basic Concepts of

Structural Reliability”, Leondardo DaVinci Pilot Project CZ/02/B/F/PP-134007,
Handbook 2-Reliability Backgrounds. 2005.
[19] H. Gulvanesian, and M. Holicky. Annex C – Calibration Proceedure, .

Leondardo DaVinci Pilot Project CZ/02/B/F/PP-134007, Handbook 2-Reliability
Backgrounds. 2005.
[20] Det Norske Veritas. DNV-OS-C101 Design of Offshore Steel Structures.

General (LRFD method). Hovik, Norway : Det Norske Veritas, 2008.
[21] T. Moan. Safety of offshore structures. Center for Offhsore Research &
Engineering, University of Singapore. Coder Report No. 2005-04. 2005
[22] F. M. Burdekin. General Principles of the use of safety factors in design

and assessment. Journal of Engineering Failure Analysis. 2006.
282
[23] R. C. Turner, C. P. Ellinas, and G. A. N. Thomas. Worldwide Calibration
of API RP2A-LRFD. Journal of Waterway, Port, Coastal and Ocean
Engineering. Vol. 120. 1994.
[24] BOMEL Ltd, Helth & Safety Executives. System-based calibration of

North West European annex environmental load factors for the ISO fixed steel
offshore structures code 19902. Berkshire : HSE Books. Research Report 087.
2003.
[25] Comité Européen de Normalisation. Eurocode 1990:2002 Basis of

structural design . EU : CEN. 2002.
[26] DG Enterprise and Industry Joint Research Centre. B1: The

EUROCODES: Implimentation and Use. EU. JRC 42745. 2008.
[27] Lloyds Registers. LRS code for offshore Platforms. London. 1998.
[28] Germanischer Lloyd. Guidelines for offshore technology, Part IV –

Industrial Services, Part 6: Offshore Technology. GL. 2007.
[29] Det Norske Veritas. DNV: Rules for the classification of Offshore
Installations. DNV. 1989.
[30] Det Norske Veritas. DNV Offshore Standard OS-C101: Design of offshore
steel structures, General – LRFD Method. DNV. 2008.
[31] Norwegian Oil industry Association and The Federation of Norwegian

Industry. NORSOK N-001 Integrity of offshore structures. Norway. 2008.
[32] A. Kolios, and F. P. Brennan. Reliability Based Design for Novel Offshore
Structures . Porto, Portugal : proceedings of the 3rd International Conference of
Integrity, Reliability and Failure, 20 - 24 July 2009.
[33] T. Moan. Target levels for reliability based reassessment of offshore

structures. Kyoto, Japan : Proceedings of the seventh international conference
on structural safety and reliability. 1997.
283
[34] T. Moan. Current trends in the safety of offshore structures. Honolulu :
Proceedings of the Eighth International Offshore and Polar Engineering
Conference, 1997.
[35] B. Bhattacharya, R. Basu, and K. Ma. Developing target reliabilityfor

novel structures: the case of the Mobile Offshore Base. Marine Structure. 13;
37-58, 2001.
[36] A. E. Mansour. An introduction to structural reliability theory. Washington :

Ship structure committee. Report SSC-351. 1990.
[37] Committee IV. Design Principles and criteria. Trondheim, Norway :

Proceedings of the 13th International Ship and Offshore Structures Congress,
1997.
[38] American Association of State Highway and Transportation Officials.

LRFD breidge design specifications, AASHTO. Washington, DC. 1994.
[39] A. S. Nowak, N. M. Szerszen, and C. H. Park. Target safety levels for

bridges. Kyoto, Japan : Proceedings of the Seventh International Conference on
Structural Safety and Reliability. 1997.
[40] C. G. Ramsaya, A. J. Bolsovera, R. H. Jonesa, and W. G. Medlanda.

Quantitative risk assessment applied to offshore process installations.
Challenges after the piper alpha disaster . Journal of Loss Prevention in the
Process Industries. Volume 7, Issue 4, Pages 317-330. 1994.
[41] R. Whitman. Evaluationg calculated risk in geotechnical engineering. J

Geotech Engng, ASCE. 110; 20:145-88, 1984.
[42] H. N. Cullen, The public inquiry into the Piper Alpha disaster. UK : The
Parliament by the Secratery of the State for Energy, 1990.
[43] R. G. Bea. Reliability Based design criteria for costal and ocean structures.
Australia : The Institution of Engineers. 1990.
284
[44] A. H. S. Ang, and D. De Leon. Determination of optimal target reliabilities
for design and ugrading of structures. Struct Safety. 19:91-103. 1997.
[45] D. L Keese, and W. R. Barton. Risk assessment and its application to

flight safety analysis. SANDIA National Laboratories. SAND 89. 1982.
[46] Construction Industry Research and Information Association.

Rationalization of safety nand serviceaility factors in structural codes. London :
CIRIA. Report No 63. 1977.
[47] D. E. Allen. Criteria for design safety factors and quality assurance
expenditure. Trondheim, Norway : Proceedings of the Third International
Conference on structural safety and reliability. 1981.
[48] International Standardization Organization. ISO/IEC Guide 51:1999 -

Safety aspects - Guidelines for their inclusion in standards. 1999.
[49] S. G. Stiansen, and A. K. Thamyambali. Lessons Learnt from Structural

Reliability Research & Applications in Marine Structures. Marine Structural
Reliability Symposium. 5th – 6th October 1987. New Jersey: Society of Naval
Architects and Marine Eng, 1987.
[50] A. M. Freudenthal. The safety of structures. Transc. ASCE, vol. 112, 1947.
[51] C. A. Cornell. A probability-based structural code. J. Am. Concr. Inst. . 66,

pp. 974–985. 1969.
[52] A. M. Hasofer, and N. C. Lind. Exact and Invariant Second Moment code
format. Journal of the Engineering Mechanics Devision. 100 (EM), ASCE, 1974.
[53] R. Rackwitz. Reliability Analysis-Past, Present and Future. USA: ASCE.

8th ASCE Specialty Conference on Probabilistic Mechanics and Structural
Reliability, 24th – 26th July, 2000.
285
[54] R. Stocki, K. Kolanek, S. Jendo, and M. Kleiber. Study on discrete
optimization techniques in reliability based optimization of truss structures.
Computers & Structures. 79; 2235-2247. 2001.
[55] G.H. Sterling, B.E. Cox, and R.M. Warrington, Shell Oil Co. Design of
the COGNAC PLATFORM for 1025 feet water Depth, GULF OF MEXICO.
Houston, Texas : Proceedings of the Offshore Technology Conference, 30
April-7 May 1979.
[56] D. I. Karsan, Handbook of Offshore Engineering . Houston, TX, USA :

AMEC Paragon, 2005.
[57] W. R. Ulbricht, M. A. Ripping, E. H. Doyle, J. W. Stevens, and J. G.

Mayfield. Design Fabrication and Installation of the Auger TLP Foundation
System. Houston, Texas, USA : Proceedings of the Offshore Technology
Conference (OTC 7626), 2-5 May 1994.
[58] J. J.S.Daniel, R and M. Knight. World trends in major offshore structures:

1970-1999. London: Marine Management (Holdings) Ltd. : Proceedings of
Offshore 1993 IMARE/RINA, Joint Offshore Group 3rd International
Conference. 17th – 18th February, 1993.
[59] J. R. Beck. Worldwide Petroleum Industry Outlook: 2001-2005 projection to

2010. 17th ed. United States of America: Pen Well Publishing, 2000.
[60] Talisman Energy. Beatrice Wind Farm Demonstrator Project Scoping

Report. UK. 2005.
[61] F. Moses. Application of Reliability to Formulation of Fixed Offshore Design

Codes. New Jersey: SNAME. 15-30. : Marine Structural Reliability Symposium.
5th – 6th October, 1987.
[62] J. C. P. Kam, M. Birkinshaw, and J. V. Sharp. Review of the Applications

of Structural Reliability Technologies in Offshore Structural Safety. Proceedings
of the 12th International Conference on Offshore Mechanics and Artic
Engineering. Vol 2. Unit. 1993.
286
[63] A. E. Mansour. Combining Extreme Environmental Loads for Reliability-
Based Designs. Arlington VA: SNAME. : Extreme Loads Response Symposium.
19th – 20th Oct, 1981.
[64] A. E. Mansour, H. Y. Jan, C. I. Zigelman, Y. N. Chen, and S. J. Harding.

Implementation of Reliability Methods to Marine Structures. Transaction
SNAME. 353-380. 1984.
[65] R. Aggarwal, R. G.Bea, B. C Gerwick, C. W. Ibbs, R. B. Reimer, and

G.C. Lee. Development of a Methodology for Safety Assessment of Existing
Steel Jacket Offshore Platforms. Houston : 22th Annual Offshore Technology
Conference. 7th – 10th May 1990.
[66] R. G. Bea. Reliability Based Requalification Criteria for Offshore Platforms.

Proceedings of the 12th International Conference on Offshore Mechanics and
Artic Engineering. Vol 2. 351-361. 1993.
[67] L. Manuel, D. G. Schmucker, C. A. Cornell, and J. E. Carballo. A

Reliability-based design format for Jacket Platforms under Waves Loads.
Journal Marine Structures. Vol (11). 413-428. 1998.
[68] L. Tvedt. Proban – probabilistic analysis . Structural Safety. Volume 28,

Issues 1-2. 2006.
[69] S. Gollwitzera, B. Kirchgäßnerb, R. Fischerb, and R. Rackwitzc.

PERMAS-RA/STRUREL system of programs for probabilistic reliability analysis.
Structural Safety. Volume 28, Issues 1-2. 2006.
[70] R. Skjong, E. B. Gregersen, E. Cramer, A. Croker, O. Hagen, G.

Korneliussen, S. Lacasse, I. Lotsberg, F. Nadim, K. O. Ronold. Guideline for
Offshore Structural Reliability Analysis - General. DNV. Report No. 95-2018.
1995.
[71] K. Bell, B. Hatlestad, O. E. Hansteen, and P. O. Araldsen. NORSAM

User Manual, Part 1, General Description. 1970.
287
[72] N. D. P. Barltrop, and A.J. Adams. Dynamics of Fixed Marine Structures.
Butterwoth-Heinemann Ltd., Reprint. 1991.
[73] D. Faulkner, M. J. Cowling, and P. A. Fierze. Integrity of Offshore

Structures - 4. Applied Science Publishers. 1990.
[74] Society of Naval Architects and Marine Engineers. Site Specific

Assessment of Mobile Jackup Units Guideline, R.P. and Commentaries,
Example calc. S.N.A.M.E. Technical & Research Bulletin 5-5. 1993.
[75] City University. Joint Industry Project. Jackup site Assessment

Procedures - Establishment of an International Recommended Practice. City
University, London, England. Seminar, 1993.
[76] T. Onoufriou, and V. J. Forbes. Developments in Structural System

Reliability Assessments of Fixed Steel Offshore Platforms. Reliability
Engineering and System Safety. Vol. (71). 189-199. 2001.
[77] Y. Murotsu, and H. Okada. Application of Reliability Assessment Methods

to Marine Frame Structures Based on Ultimate Strength Analysis. New Jersey:
S.N.A.M.E. Marine Structural Reliability Symposium. 5th – 6th October, 1987.
[78] N. K. Shetty. Selective enumeration method for identification of dominant

failure paths of large structures. Huston, USA : Proceedings of the International
Conference on Offshore Mechanics and Arctic and Engineering. 1994.
[79] P. Thoft-Christensen, and Y. Murotsy. Application of Structural Systems

Reliability Theory. Berlin : Springer-Verlag. 1986.

Collapse Analysis of Jackets. Rotterdam: Balkema. 535-543. Schuëller,
Shinozuka and Yao. Structural Safety and Reliability. 1994.
[81] P. S. Tromans, and J. W. Van de Graaf. Substantiated Risk Assessment

of Jacket Structure. Journal of Waterway, Port Coastal and Ocean Engineering.
Vol 120 (6). 535-555, 1992.
288
[82] J. W. Van de Graaf, P. S. Tromans, and M. Efthymiou. The Reliability of
Offshore Structures and Its Dependence on Design Code and Environment.
Houston: OTC. 105-118 : 26th Annual Offshore Technology Conference. 2rd –
5th May, 1994.
[83] R. G. Bea, and M. M. Mortazavi. ULSLEA: A Limit Equilibrium procedure

to determine the Limit State Loading of Template-type Platforms. Journal of
Offshore Mechanics and Artic Engineering. Transactions of the ASME. Volume
118(4). 267-275. 1996.
[84] P. A. Frieze, A. C. Morandi, M. Birkinshaw, D. Smith, and A. T. Dixon.

Fixed and Jack-up Platforms: Basis for Reliability Assessment. Journal of
Marine Structure. volume 10. 263-284. 1997.
[85] H. Nordal, C. A. Cornell, and A.Karamchandani. A Structural System

Reliability Care Study of an Eight-leg Steel Jacket Offshore Production
Platform. Marine Structural Reliability Symposium. New Jersey: Society of Navl
Architect. 5-6 October, 1987.
[86] R. S. De, A. Karamchandari, and C. A. Cornell. Offshore Structural

System Reliability under Changing Load Pattern. Applied Ocean Research. Vol.
13 (3). 145-157. 1991.
[87] M. Hohenbichler. An Approximation to the Multivariate Normal Distribution.

Lyngby : DIA-LOG 6-82, Danish Academy of Engineers, pp. 79-100. 1982.
[88] H. Pham. Handbook of reliability engineering . London : Springer. 2003.
[89] O. Ditlevsen. Narrow reliability bounds for structural systems. J. Struct.

Mech. 7, pp. 453–472 1979.

Collapse Analysis of Jackets. In: Schuëller, Shinozuka and Yao. Structural
Safety and Reliability. Rotterdam: Balkema. 535-543. 1994.
289
[91] R. K. Aggarwal, R. G. Bea , B. C. Gerwic. C. W. Ibbs, R. B. Reimer, and
G.C. Lee. Development of a Methodology for Safety Assessment of Existing
Steel Jacket Offshore Platforms. 22th Annual Offshore Technology Conference.
Houston. 7th – 10th May, 1978. 1990.
[92] Health & Safety Executive. Offshore Technology Report-OTO 2000 037
(a Review of Reliability Considerations for Fixed Offshore Platforms). UK:
University of Surrey. HSE. 2000.
[93] S-K Choi, R. V. Grandhi, and R. A. Canﬁeld. Structural reliability under

non-Gaussian stochastic behavior. Computers & Structures. 82 (2004) 1113–
1121. 2004.
[94] G. I. Schueller. A state-of-the-art report on computational stochastic

mechanics. Journal of Probabilistic Engineering Mechanics. 12(4):197-313.
1997.
[95] R. G. Spanos, and P. D. Ghanem. Spectral stochastic finite-element

formulation for reliability analysis. J. Engrg. Mech. ASCE 117 (10), pp. 2351–
2372. 1991.
[96] R. G. Ghanem and W. Brzakala. Stochastic finite-element analysis of soil

layers with random interface. J. Engrg. Mech. ASCE 122 (4), pp. 361–369.
1996.
[97] R. G. Ghanem. Stochastic finite elements with multiple random non-normal

properties. J. Engrg. Mech. ASCE 125 (1), pp. 26–40. 1999.
[98] C. G. Soares, and N.-Z. Chena. Spectral stochastic finite element analysis
for laminated composite plates . Computer Methods in Applied Mechanics and
Engineering. Volume 197, Issues 51-52, Pages 4830-4839. 2008.
[99] M. A. Tatang. Direct Incorporation of Uncertainty in Chemical and

Environmental Engineering Systems, Ph.D. Dissertation. Massachusetts
Institute of Technology, Cambridge, MA. 1995.
290
[100] S. S. Isukapalli. Uncertainty Analysis of Transport-Transformation
Models, Ph.D . Dissertation. Rutgers, The State University of New Jersey, New
Brunswick, NJ. 1999.
[101] C. L. Pettit, R. A. Canfield, and R. Ghanem. Stochastic Analysis of an

Aeroelastic System. New York. 15th ASCE Engineering Mechanics Conference,
Columbia University. June 2-5, 2002.
[102] D. Xiu, and G. Karniadakis,. The Wiener-Askey Polynomial Chaos for

Stochastic Differential Equations. SIAM Journal on Scientific Computing. Vol.
24, (2), pp. 619-644. 2002.
[103] R. Ghanem, and P. D. Spanos. Stochastic Finite Elements: A Spectral

Approach. NY. Springer-Verlag, 1991.
[104] I. Babuska, R. Tempone, and G. E. Zouraris. Galerkin Finite Element

Approximations of Stochastic Elliptic Differential Equations. SIAM Journal on
Numerical Analysis. Vol. 42, pp. 800-825. 2004.
[105] I. Elishakoff. Simulation of space-random fields for solution of stochastic

boundary-value problems. J. Acous. Soc. Am. 65. 399-403. 1979.
[106] B. Ellingwood, and J. Zhang. Orthogonal series expansion of random

fields in first-order reliability analysis. ASCE J. Engrg. Mech. 1994.
[107] M. Grigoriu. Simulation of nonstationary Gaussian processes by random

trigonometric polynomials. ASCE J. Engrg. Mech. 119(2). 328-343. 1993.
[108] H. Hotelling. Analysis of a Comnplex of Statistical Variables into Principle

components. Journal of Educational Psycology. Vol.24, pp.498-520. 1933.
[109] R. L. Iman, and W. J. Conover. A distribution free approach to inducing

Rank Correlation among input variables. Communications in statistics:
Simulation and Computation. B11, pp.311-334. 1982.
291
[110] M. Stein. Large Sample Properties of Simulations Using Latin Hybercube
Sampling. Technometrics. Vol.29, (2), pp. 143-151. 1987.
[111] S. Choi, R. A. Canfield, and R. V. Grandhi. Estimation of Structural

Reliability for Gaussian Random Fields. Structure & Infrastructure Engineering.
Nov. 2005.
[112] S. Choi, R. V. Grandhi, and R. A. Canfield. Robust Design of

Mechanical Systems via Stochastic Expansions. International Journal of
Materials and Product Technology. Vol. 25, pp. 127-143. 2006.
[113] S. Choi, R. V. Grandhi, R. A. Canfield, and C. L. Pettit. Polynomial

Chaos Expansion with Latin Hybercube Sampling for Estimating Response
Variability. AIAA Journal. Vol. 42, pp. 1191-1198. 2004.
[114] E. Cureton, and R. D'Agostinao. Factor Analysis: An applied approach.

NJ : Lawrence Erbaum Associates. 1983.
[115] W. T. Martin, and R.H. Cameron. The orthogonal development of

nonlinear functionals in series of Fourier-Hermite functionals. Ann. Math. 48
385-392. 1974.
[116] S-K. Choi, R. Grandhi, and R. Canfield. Reliability-based Structural

Design. USA. Springer. 2006
[117] Society of Automotive Engineers. Integration of Probabilistci Methods

into the Design Process. Aerospace Information Report 5080. 1997.
[118] A. M. Freudenthal, J. M. Garrelts, and M. Shinozuka. The Analysis of

Structural Safety. Journal od the Structural Division. ASCE. Vol. 92. 1966.
[119] M. Rosenblatt. Remarks on a Multivariate Transformation. The Annals of

Mathematical Statistics. Vol. 23. 1952.
292
[120] M. Hohenbichler, and R. Rackwitx. Non-normal Dependent Vectors in
Structural Safety. Journal of the Engineering Mechanics Division. ASCE. Vol.
107, No. EM6. 1981.
[121] R. E. Merchelers. Structural Reliability Analysis and Prediction. UK. Ellis

Horwood Limited. 1987.
[122] L. Wang, and R. V. Grandhi. Reliability-based structural optimization

using improved two-point adaptive nonlinear approximations . Finite Elements in
Analysis and Design. Volume 29, Issue 1, Pages 35-48. 1998.
[123] K. Breitung. Asymptotic Approximations for Multinormal Integrals. Journal

of the Engineering Mechanics Division. Vol. 110. 1984.
[124] L. Tvedt. Two Second-order Approximations to the Failure Probability -

Section on Structural Reliability. Norwar. DNV. 1984.
[125] L. Tvedt. Distribution of Quadratic Forms in Normal Space Applicationsto

Structural Reliability. Journal of the Engineering Mechanics Division. ASCE.
Vol. 116. 1990.
[126] M. Hohenbichler, and R. Rackwitz. Improvement of Second-order

Reliability Estimates by Importance Sampling. Journal of Engineering
Mechanics Division. ASCE. Vol. 116. 1990.
[127] H. U. Koyluoglu, S. R. K. Nielsen. New Approximations for SORM

Integrals. Structural Safety. Vol. 13. 1994.
[128] G. Q. Cai, and I. Elishakoff. Refined Second-order Approximations

Analysis. Structural Safety. Vol. 14. 1994.
[129] L. P. Wang, and R. V. Grandhi. Improved Two-point Function

Approximation for Design Optimizaation. AIAA Journal. Vol. 32. 1995.
[130] A. Der Kiureghian, H. Z. Lin, and S. J. Hwang. Second Order Reliability

Approximations. Journal of Engineering Mechanics. Vol. 113. 1987.
293
[131] J. Todd. Survey of Numerical Analysis. New York : McGraw Hill. 1962.
[132] K. Breitung. Asymptotic Approximations for Multinormal Integrals. Journal

of the Engineering Mechanics Division. ASCE, Vol. 110, No. 3. pp 357-366.
1984.
[133] Det Norske Veritas. PROBAN Theory Manual. DNV. DNV Software
Report No.: 96-7017. 2004.
[134] J. von Neumann, and S. Ulam. The Monte Carlo Method. Bulletin
AMS,51-0-165. 1945.
[135] M. D. McKay, R. J. Beckman, and W. J. Conover. A comparison of three

methods for selecting valuesm of input variables in the analysis of output from a
computer code. Technometrics. Vol.21, (2), pp. 239-245. 1979.
[136] D. C. Charmpis and G. I. Schuëller. Using Monte Carlo Simulation to

Treat Physical Uncertainties in Structural Reliability. Lecture Notes in
Economics and Mathematical Systems. Volume 581, Part II, 67-83. 2006.
[137] E. C. Anderson, Monte Carlo Methods and Importance Sampling. Notes,

1999.
[138] R. E. Guerra, T. N. Saleem, and T. D. Savitsky. Importance Sampling.

Rice University, Lecture Notes. 2008.
[139] T.S. Doorn, E.J.W. ter Maten, J.A. Croon, A. Di Bucchianico, and O.
Wittich. Importance Sampling Monte Carlo simulations for accurate estimation
of SRAM yield. Eindhoven University of Technology, 2008.
[140] S-K Au, and J. L. Beck. Estimation of small failure probabilities in high
dimensions by subset simulation. Probabilistic Engineering Mechanics Volume
16, Issue 4, Pages 263-277. October 2001.
294
[141] G. I. Schuëller, H. J. Pradlwarter, and P. S. Koutsourelakis. A critical
appraisal of reliability estimation procedures for high dimensions. Probabilistic
Engineering Mechanics Volume 19, Issue 4, Pages 463-474. October 2004.
[142] M. Shinozuka. Basic analysis of structural safety. Journal of Structural

Engineering. ASCE, Vol 109(3), pp. 721-740. 1983.
[143] I. Deak. Three digit accurate multiple normal probabilities. Numerische

Mathematik. Vol. 35, pp. 369-380. 1980.
[144] M. Hohenbichler, and R. Rackwitz. Sensitivity and importance measures

in structural reliability. Civil Engineering Systems. Vol. 3(4), pp. 203-209. 1986.
[145] G. Schall, S. Gollwitzer, and R. Rackwitz. Integration of multinormal

densities on surfaces. Proceedings of the 2nd IFIP Working Conference on
Reliability and Optimization on Structural Systems. P. Thoft-Christensen. ed.
Springer Verlag, 1988.
[146] P. Bjerager, and S. Krenk. Sensitivity measures in structural reliability

analysis. Proceedings of the 1st IFIP Working Conference on Reliability and
Optimization on Structural Systems. P. Thoft-Christensen, ed. Springer Verlag,
Berlin, pp 459-470. 1987.
[147] R. Y. Rubinstein. Simulation and the Monte Carlo Method. s.l. : John
Wiley & Sons Ltd, New York. pp. 115-118. 1981.
[148] R. E. Melcher. Structure Reliability Analysis and Prediction, 2nd ed.

England : John Wiley & Sons Ltd. 1999.
[149] W. C. Broding, F. W. Diederich, and P. S. Parker. Structural

optimization and design based on a reliability design criterion. J. Spacecraft.
1(1), 56–61. 1964.
[150] L. Faravelli. Response Surface Approach for Reliability Analysis. J Eng

Mech. 115(12):2763-81. 1989.
295
[151] M. Lemaire. Finite element and reliability - Combined methods by
response surface PROBAMAT-21st century: Probabilities and materials - Tests,
models and applications for the 21st century; Proceedings of the NATO
Advanced Research Workshop, Perm, Russia; NETHERLANDS; 10-12 Sept.
1997. pp. 317-331. 1998
[152] D. C. Cox, and P Baybutt. Methods for uncertainty analysis: a

comparative survey. Risk Anal . 1(4):251–8. 1981.
[153] S. H. Kim, and S. W. Na. Response surface method using vector

projected sampling points. Struct Saf. 19:3–19. 1997.
[154] K. Breitung, and L. Faravelli. Log-likelihood maximization and response

surface reliability assessment. Nonlinear Dyn. 4:273–86. 1994.
[155] C. G. Bucher, and U. Bourgund. A fast and efficient response surface

approach for structural reliability problems. Struct Saf . 7:57–66. 1990.
[156] S. Gupta, and C. S. Manohar. An improved response surface method for

the determination of failure probability and importance measures. Struct Saf .
26:123–39. 2004.
[157] V. W. Liu, and F. A. Moses. A sequential response surface method and

its application in the reliability analysis of aircraft structural systems. Struct Saf.
16:39–46. 1994.
[158] M. R. Rajashekhar, and B. R. Ellingwood. A new look at the response

surface approach for reliability analysis. Struct Saf . 12:205–20. 1993.
[159] L. Olivi. Response surface methodology in risk analysis. In: Apostolakis

G, Garribba S, Volta G, editors. Synthesis and analysis methods for safety and
reliability studies. Plenum Press, 1980.
[160] H. P. Gavin, and S. C. Yau. High-order limit state functions in the

repsonse surface method for structural reliability analysis. Structural Safety. 30,
162-179. 2008.
296
[161] S. Engelund, and R. Rackwitz. Experiences with experimental design
schemes for failure surface estimation and reliability. In: Proceedings of the
sixth speciality conference on probabilistic mechanics and structural and
geotechnical reliability. ASCE. 1992.
[162] I. Enevoldsen, M. H. Faber, and J. D. Sorensen. Adaptive response

surface technique in reliability estimation. Structural Safety. 1257–64. Schuëller,
Shinozuka, Yao. 1994.
[163] N. Devictor, M. Marques, and M. Lemaire. Utilisation des surfaces de

réponse dans le calcul de la fiabilité des composants mécaniques. Revue
Française de Mécanique. 1:43–51. 1997.
[164] S. H. Kim, and S. W. Na. Response surface method using vector

projected sampling points. Structural Safety . 19:3–19. 1997.
[165] P. K. Das, and Y. Zheng. Cumulative formation of response surface and

its use in reliability analysis. . Probabilistic Engineering Mechanics . 15:309–15.
2002.
[166] N. Impollonia, and A. Sofi. A response surface approach for the static
analysis of stochastic structures with geometrical nonlinearities. Computer
Methods in Applied Mechanics and Engineering . 192:4109–29. 2003.
[167] A. I. Khuri, and J. A. Cornell. Response surfaces designs and analyses.

2nd ed. Marcel Dekker, 1996.
[168] I. Kaymaz, and C. A. McMihon. A response surface mehtod based on

weighted regression for reliability analysis. Probabilistic Engineering Mechanics.
20;11-17. 2005.
[169] A.S.Nguyen, A. Sellier, F. Duprat, and G. Pons. Adaptive response

surface method based on a doublke regression technique. Probabilistic
Engineering Mechanics. 135-143. 2009.
297
[170] B. Sudret, and A. Der Kiureghian. A stochastic Finite Elements methods
and reliability. State of the art report. University of California. 2000.
[171] N. Gayton, J. M. Bourinet, and M. Lemaire. CQ2RS: A new statistical

approach to the response surface method for reliability analysis. Structural
Safety. 25: 99-121. 2003.
[172] R. H. Myers. Response Surface Methodology. Boston. Allyn and Bacon

Inc. 1971.
[173] J. A. Nelder. Inverse polynomials, a useful group of multifactor response

functions. Biometrics. 22, 128-141, 1966.
[174] D. J. J. Toal, N. W. Bressloff, and A.J. Keane. Kriging hyperparameter

tuning strategies. AIAA Journal. vol. 46:pp. 1240–1252. 2008.
[175] D. R. Jones. A taxonomy of global optimization methods based on

response surfaces. Journal of Global Optimization. vol. 21:pp. 345–383. 2001.
[176] A. Forrester, A. Sobester, and A. Keane. Engineering design via

surrogate modelling: a practical guide. Chichester, UK. Wiley, 2008.
[177] A. Sobey, J. Underwood, J. Blake, and & A. Shenoi. Reliability analysis

of damaged steel structures using finite element analysis. Scotland, UK.
Proceeding of 5th International ASRANet Conference, 2010.
[178] O. C. Zienkiewicz, and R. L.Taylor. The Finite Element Method. 5th.

Butterworth-Heinemann. Vol. 2: Solid Mechanics. 2000.
[179] S. S. Rao. The Finite Element Method in Engineering. 4th. Elvesier

Science & Technology Books, Dec. 2004.
[180] R. T. Fenner. Finite Element Methods for Engineers. Imperial College

Press. 1996.
[181] A. J. M. Ferreira. MATLAB code for Finite Element Analysis. Springer.

2009.
298
[182] A. S. Nowak, and K. R. Collins. Reliability of structures. United States of
America. McGraw Hill. 2000.
[183] B. J. Muga, and J. F. Wilson. Dynamic Analysis of Ocean Structure. New

York. Pleanum Press. 1970.
[184] M. Efthymiou, and C. G. Graham. Environmental Loading on Fixed

Offshore Platforms. Netherlands: SUT : Environmental Forces on Offshore
Structures and their Prediction. 1990.
[185] J. Gaythwaite. The Marine Environment and Structural Design. New

York : Van Nostrand Reinhold Co. 1981.
[186] N. P. D. Baltrop, G. M. Mitchell, and J. B. Arkins. Fluid Loading on

Fixed Offshore Structure. London : Department of Energy. 1990.
[187] O. T. Gusmestad, and D. Karunakaran. Wave Current Interaction.

London : Proceedings of the Society of Underwater Technology, 81-110. 1990.
[188] S. K. Chakrabati. Nonlinear Methods in Offshore Engineering. New York :

Elsevier. 1990.
[189] D. H. Peregrine. Interaction of Water Waves and Currents. Advances in

Applied Mechanics. 16: 9-117. 1976.
[190] R. A. Dalrymple. Models for Nonlinear Water Waves in Shear Currents.

Houston. OTC, 843-856 : Proceedings of the 6th Annual Offshore Technology
Conference. 1974.
[191] M. H. Patel. Offshore Structures. N. Morgan. Marine Technology

Reference Book. London. Butterworth. 1990.
[192] P. Pagrut, and M. C. Deo. Extreme Wave Height Estimates. San

Francisco:ISOPE, 186, 193 : Proceedings of the 2nd International Offhsore and
Polar Engineering Conference. 1992.
299
[193] B. Le. Mehaute. An Introductin to Hydraudynamics and Water Waves.
Dusseldorf. Springer-Verlag. 1976.
[194] T. Sarpkaya, and M. Isaacson. Mechanics of Wave Forces on Offshore

Structures. New York. Van Nostrand Reinhold, 1981.
[195] R. G. Dean. Relative Validities of Water Wave Theories. Journal of

Waterways, Harbors and Costal Engineering Division. 96(WWI): 105-119. 1970.
[196] J. H. Vugts. A Review of Hydrodynamic Loads on Offshore Structures

and Their Formulation. London: BOSS, 693-708 : Proceedings of the 2nd
International Conference on Behaviour of Offshore Structures. 1979.
[197] J. R. Morison, M. P. O'Brien, J. W. Johnson, and S. A. Schaaf. The

Force Exerted by Surface Waves on Piles. Petroleum Transactions. 189:149-
157. 1950.
[198] T. H. Dawson. Offshore Structural Engineering. Englewood Cliffs:

Prentice Hall. 1983.
[199] T.Sarpkaya. In-Line and Transverse Forces on Cylinders in Oscillatory

Flow at High Reynolds Numbers. Huston: OTC, 95-108. Proceedings of 8th
Annual Offshore Technology Conference. 1976.
[200] R. L. Wiegel. Oceanographical Engineering. New Jersey. Prentice Hall.

1964.
[201] O. T. Gusmestad, and D. Karunakaran. Wave Current Interaction.

London: Society for Underwater Technology: 81-110 : Proceedings of the
Environmental Forces on offshore structures and their Prediction. 1990.
[202] G. Rodenbusch, and C. Kallstorm. Forces on a Large Cylinder in

Random Two-Dimension Flows. Huston. OTC. Proceedings of 18th Annual
Offshore Technology Conference. 1986.
300
[203] M. de St. Q. Isaacson, and D. J. Maulli. Transverse Forces on Vertical
Cylinders in Waves. Journal of Waterways, Harbors and Costal Engineering
Division. 102 (WW1). 49-60. 1976.
[204] C. Petrauskas, and P. M. Aagaard. Extrapolation of Historical Storm

Data for Estimating Design - Wave Heights. Society of Petroleum Engineers
Jounral. 11(1): 23-37. 1971.
[205] R. E. Haring, and J. C. Heidenman. Gulf of Mexico Rare Wave Return

Periods. Journal of Petroleum Technology. 32(1): 35-47. 1980.
[206] G. P. V. Vledder, and T. J. Zitman. Design Waves: Statistics and

Engineering Practice. San Francisco, USA: The International Society of
Offshore and Polar Engineering : Proceedings of the 2nd International Offshore
and Polar Engineering Conference. 1992.
[207] M. N. Madsen, J. B. Nielsen, P. Klinting, and J. Knudsen. A Design

Load Method For Offshore Structures Based Upon the Joint Probability of
Environmental Parameters. USA: ASME, 75-80 : Proceedings of 7th
International Conference on Offshore Mechanics and Artic Engineering. 1988.
[208] E. M. Bitner-Gregersen, and S. Haver. Joint Environmental Model for

Reliability Calculations. UK. The international Society of Offshore Conference.
Proceedings of the First International Offshore and Polar Engineering
Conference. 1991.
[209] J. C. Heideman, O. Hagen, C. Cooper, and D. Finn-Erik. Joint

Probability of Extreme Waves and Currents on Norwegian Shelf. Journal of
Waterway, Port, Coastal and Ocean Engineering. Vol.115(4). 534-546. 1989.
[210] P. S. Tromas, and L. Vanderschuren. Response Based Design

Conditions in the North Sea: Application of the New Method. Houston. OTC.
287-297. Proceedings of 27th Annual Offshore Technology Conference. 1995.
[211] Fugro GEOS. Wind and wave frequency distributions for sites around the
British Isles. London. Health and Safety Executives. 2001.
301
[212] M. J. Tucker. Waves in Ocean Engineering: Measurements, Analysis,
Interpretation. Chichister. Ellis Horwood Ltd. 1991.
[213] D. Myrhaug, and S. P. Kjeldsen. Parametric Modeling of joint Probability

density distributions for steepness and asymmetry in deep water waves.
Applied Ocean Research. Vol. 6 (4). 207-220. 1984.
[214] M. A. Srokosz. A Note on the Joint Distribution of Wave Height and

Period During the Growth Phase of a Storm. Journal of Ocean Engineering. Vol.
15 (4). 379-387. 1988.
[215] M. A. Skrokosz, and P. G. Challenor. Joint Distribution of Wave Height

and Period: A Critical Comparison. Journal of Ocean Engineering. Vol.14(4).
296-311. 1987.
[216] S. Haver. Wave Climate of Northern Norway. Applied Ocean Research.

Vol. 14 (5). 359-376. 1985.
[217] S. Haver. On the Joint Distribution of Heights and Periods of Sea Waves.
Journal of Ocean Engineering. Vol.14 (5) 359-376. 1987.
[218] M. K. Ochi. On Long-term Statistics for Ocean and Coastal Waves.

Proceedings of 16th Coastal Engineering Conference. Vol.1. 59-75. 1978.
[219] G. P. van Vledder, and T. J. Zitman. Design Waves: Statistics and

Engineering Practice. San Francisco: ISOPE, 170-178 : Proceedings of the 2nd
International Offshore and Polar Engineering Conference. 1992.
[220] J. N. Sharma, and W. G. Grosskopf. Practical Considerations in

Preparing Extreme Wind, Wave and Current Criteria. Houston: OTC, 377-386 :
Proceedings of the 26th Annual Offshore Technology Conference. 1994.
[221] L. R. Muir, and A. H. El-Shaarawi. On the calculation of Extreme Wave

Heights: A Review. Ocean Engineering. 13(1): 93-118. 1986.
302
[222] C. G. Soares, and M. Scotto. Modelling Uncertainty in Long-Term
Predictions of Significant Wave Height. Journal of Offshore Mechanics and Artic
Engineering. 118: 284-291. 1996.
[223] D. J. T. Carter, and P. G. Challenor. Methods of Fitting the Fisher-Tippett

Type I Extreme Value Distribution. Ocean Engineering. 10(3): 191-199, 1983.
[224] G. Q. Yao, B. C. Ding, J. Y. Wang, and Z. X. Ma. Statistic Analysis of

Wave Parameters for the China Sea. Singapore: ISOPE, 22-26 : Proceedings of
the 3rd International Offshore and Polar Engineering Conference. 1993.
[225] L. E. Borgman. Rsik Criteria. Journal of Waterways and Harbors Division.

89(WW3): 1-35. 1963.
[226] R. C. Turner, and M. J. Baker. A Probabilistic Wave Model and Reliability

Analysis. Berlin : Proceedings IFIP Working Conference on Reliability and
Optimization of Structural Systems. In Lecture Notes in Engineering. Springer-
Verlag. 1988.
[227] P. M. Hagemeijer. A Comparison between a Deterministic and

Probabilistic Fluid Loading Model for a Jacket Structure. The Hague: OMAE.
89-97 : Eight International Conference on Offshore Mechanics and artic
Engineering. 1989.
[228] M. K. Ochi. Probabilistic Extreme Values and Their Implication for

Offshore Structure Design. Houston: OTC. 987-989 : Proceedings of 10th
Offshore Technology Conference. 1978.
[229] M. S. Longuet-Higgins. On the Joint Distribution of wave periods and

Amplitudes in a Random Wave Field. London. 241-258 : Proceedings of the
Royal Society of London. 1983.
[230] J. Mathisen, and E. Bitner-Gregersen. Joint distribution for Significant

Wave Height and Wave Zero-Up-Crossing Period. Applied Ocean Research.
Vol.12, (2), 93-103. 1990.
303
[231] D. M. Constantine, and K. Tzanis. Numerical Results of the Joint
Probability of Heights and Periods of Sea Waves. Costal Engineering. Vol. 22.
217-235. 1994.
[232] M. G. Hallam, N. J. Heaf, and L. R. Wootton. Dynamics of Marine

Structures: Methods of calculating the dynamic response of fixed structures
subjected to wave and current action. London : Ciria Underwater Engineering
Group. 1977.
[233] R. G. Bea, and N. W. Lai. Hydrodynamic Loadings on Offshore Platforms.

Houston: OTC, 155-164. Proceedings of 10th Annual Offshore Technology
Conference. 1978.
[234] Norwegian Petroleum Directorate. Regulations Relating to Loadbearing

Structures in the Petroleum Activities. Norway. NPD. 1994.
[235] D. K. Hart, S. E. Rutherford, and A. H. S. Wickham. Structural Reliability

Analysis of Stiffened Panels. UK. The Royal Institute of Naval Architects. 1985.
[236] D. Smith, A. Csenki, C. P. Ellinas. Ultimate Limit State Analysis of

Unstiffened Structural Components. Glasgow, UK. IOS' 87. 1987.
[237] T. V. Galambos, and M. K. Ravindra. Properties of Steel for Use in

LRFD. Journal of Structural Division. ASCE, Vol. 104, No. ST9. 1978.
[238] J. P. Tronskar, A. Karlsen, H. Bratfos, M. Hauge. Probabilitstic Fracture

Mechanics Analysis of Welded K-node: Significance of Local Brittle Zones.
Calgary, Alberta, Canada : Proceedings of 11th International Conference on
Offshore Mechanics and Arctic Engineering, OMAE. 1992.
[239] ECCS. Manual on Stability of Steel Structures. European Convention for

Constructural Steelwork. 1976.
[240] V. Galambos. Guide to stability Design Criteria for Metal Structures. s.l. :
Fourth Edition. John Wiley & Sons. 1988.
304
[241] R. Bjorhovde. Chapter 2.3 - Compression Members. J. E. Harding, and
R. Bjorhovde P. J. Dowling. Constructional Steel Design - an International
Guide. Elsevier Applied Science. 1992.
[242] R. D. Sherman. Chapter 2.4 - Tubular Members. Constructional Steel

Design - An International Guide. P. J. Dowling, J. E. Harding, and R. Bjorhovde.
Elsevier Applied Science. 1992.
[243] M. A. Bonello, M. K. Chryssanthopoulos, and P. J. Dowling. Ultimate

Strength Design of Stiffened Plates under Axial Compression and Bending.
Marine Structures. Vol. 6, Nos. 5&6, pp. 533-552. 1993.
[244] C. P. Ellinas, M. Lalani, M., and J. V. Sharp. Tubular Joint Strength

Formulations. The Implication for Harmonisation. London, England :
Proceedings, SUT Int. Conf. on API RP 2A-LRFD, Its Present and Future Role
in Offshore Safety Cases. 1993.
[245] D. Faulkner. Efficient Design of Orthogonally Stiffened Cylinders. London,

England : Proceedings, BOSS conference. 1992.
[246] M. B. Gibstein. Static Strength of Tubular Joints. Norway : Det Norske

Veritas. Report No. 73-86-C. 1973.
[247] M.B. Gibstein. The Static Strength of T-Joints Subjected to In-plane

Bending Moments. Norway : Det Norske Veritas. Report No. 76-137. 1976.
[248] Department of Energy. Background to New Static Strength Guidance for

Tubular Joints in Steel Offshore Structures. 1990.
[249] MSL Engineering. Static Strength of Tubular Joints in Offshore

Structures. Report Health & Safety Executive. 1992.
[250] J. A. Yura, N. Zettlemoyer, and I. F. Edwards. Ultimate Capacity

Equations for Tubular Joints. Houston, Texas. Proceedings, Offshore
Technology Conference, OTC Paper No. 3690. 1980.
305
[251] I. Lotsberg. Design of Tubular Joints. Norway. NIF Course. 1990.
[252] F. Nadim. Geotechnical Site description using stochastic interpolation.

Oslo, Norway. 10th Geotechnical Meeting (NGM-88). 1988.
[253] J. M. Keaveny, F. Nadim, and S. Lacasse. Autocorrelation function for

offshore geotechnical data. San Francisco. Proceedings of the 5th International
Conference on Structural Safety and Reliability. ICOSSAE '89. 1989.
[254] T. R. Guttormsen. Reliability analysis for offshore foundations —

uncertainty in soil parameters. Oslo. NGI Report No. 51411-3. 1987.
[255] Norweigian Geotechnical Institute. Reliability-based design of offshore

foundations - model uncertainty in axial pile capacity calculations. London : HSE
Books. Offshore Technology Report OTN 95 145. 1994.
[256] K. H. Andersen, , R. Dyvik, R. Lauritzsen, D. Heien, L. Hårvik, and T.

Amundsen. Model tests of gravity platforms: Interpretation. J. Geotech. Engrg.
ASCE, Vol. 115, No. 11, pp. 1550-1568. 1989.
[257] K. H. Andersen, R. Dyvik, K. Schrøder, O.E. Hansteen, and S.

Bysveen. Field tests of nchors in clay - II: Predictions and interpretation. J.
Geotech. Engrg. ASCE, Vol. 119, No. 10. 1993.
[258] Det Norske Veritas. Fatigue Strength Analysis for Mobile Offshore Units.
Høvik, Norway : DNV. DNV Classification Note 30.2. 1984.
[259] British Standards Institution. Guidance on Methods for Assessing the

Acceptability of Flaws in Fusion Welded Structures. UK. PD 6493. 1991.
[260] H. O. Madsen, S. Krenk, and N. C. Lind. Methods of Structural Safety.

N.J. Prentice-Hall Ltd, Englewood Clifs. 1986.
[261] J. G. Kuang, A. B. Potvin, and R. D. Leick. Stress Concentrations in

Tubular Joints. Houston, Texas : Proceedings, Offshore Technology
Conference, OTC Paper No. 2205. 1975.
306
[262] M. B. Gibstein Parametric Stress Analysis of T-Joints. Cambridge,
England : Proceedings, European Offshore Steel Research Seminar. Paper 26.
1978.
[263] M. Efthymiou. Development of SCF Formulae and Generalised Influence

Functions for Use in Fatigue Analysis. Surrey, England : OTJ'88 Recent
Developments in Tubular Joints Technology. 1988.
[264] A. K. Hellier, M. P. Conolly, and W. D. Dower. Stress Concentration

Factors for Tubular T and Y Joints. Int. Journal of Fatigue. Vol. 12, No. 1. 1990.
[265] P. H. Wirsching. Fatigue Reliability of Offshore Structures. Journal of

Structural Engineering. ASCE, Vol. 110, pp. 2340-2356. 1984.
[266] R. E. Melchers. Modeling of marine corrosion of steel specimens. W. T.

Young, and R. M. Kain. Corrosion testing in natural waters. Philadelphia: ASTM
STP 1300, p. 20-33. 1997.
[267] A. Cosham, P. Hopkins, and K. A. Macdonald. Best practice for the

assessment of defect in pipelines - Corrosion. Engineering Failure Analysis. 14
1245-1265, 2007.
[268] R. E. Melchers. The effect of corrosion on the structural reliability of steel

offshore structures. Corrosion Science. 47, 2391-2410. 2005.
[269] R. E. Melchers. Corrosion uncertainty modelling for steel structures.

Journal of Constructional Steel Research. 52, 3-19. 1999.
[270] F. L. LeQue. Marine Corrosion: causes and prevention. New York. Wiley
Interscience. 1975.
[271] B. D. Craig. Handbook of corrosion data. Ohio : ASM International. 1989.
[272] S. Qin, and W. Cui. Effect of corrosion models on the time dependent
reliability of steel plated elements. China. 16, 15-34. 2003.
307
[273] C. R. Southwell, J. D. Bultman, and C. W. Hummer Jr. Estimating of
service life of steel in sea water. M. Schumacher. Seawater corrosion
Handbook. New Jersey. Noyes Data Corporation. 1979.
[274] R. E. Melchers. Probabilistic Modeling of immersion marine corrosion. M.

Shinozuka, Y. K. Wen, and N. Shiraishi. Structural Safety and Reliability.
Rotterdam: Balkema. 1998.
[275] C. G. Soares, and Y. Garbatov. Reliability on maintained, corrosion

protected plates subjected to non linear corrosion and compresive loads. Mar
Struct. 12: 425-45. 1999.
[276] R. E. Melchers. Pitting Corrosion of mild steel in marine immersion

environment-1: maximum pit depth corrosion. NACE. 60 (9) 824-836. 2004.
[277] R. E. Melchers. Pitting Corrosion of mild steel in marine immersion

environment-2: variability of maximum pit dpeth. NACE. 60 (10) 937-944. 2004.
[278] J. Shigley, C. Mischke, and R. Budynas. Shigley’s Mechanical

Engineering Design, 8th Edition. USA : McGraw.Hill Primis, 2006.
[279] G. Tsamasfyros. Mechanics of Deformable Bodies 1 (Greek edition).

Athens, Greece. Symmetria Publications. 1990.
[280] G. Tsamasfyros. Mechanics of Deformable Bodies 2 (Greek Edition).

Athens, Greece. Symmetria Publications. 1990.
[281] J. Marine. Engineering Materials. Englewood Cliffs, N.J. Prentice-Hall.

1952.
[282] J. Mathisen, and E. Bitner-Gregersen. Joint distribution for significant

wave height and wave zero-up-crossing period. Applied Ocean Research.
Vol.12 No.2. 1990.
[283] S. K. Chakrabarti. Nonlinear Methods in Offshore Engineering.

Netherlands. Elsevier Science Publishing Company Inc. 1990.
308
[284] S. Havier. On the Joint Distribution of Heights and Perioods of Sea
Waves. Journal of Ocena Engineering. Vol.14(5), 359-376. 1987.
[285] P. Pagrut, and M. C. Deo. Extreme Wave Height Estimates. 2nd

International Offshore and Polar Engineering Conference. USE: ISOPE. 186-
193. 1992.
[286] J. F.Wilson. Dynamics of Offshore Structures. New Jersey : Johny Wiley

and Sons. 1984.
[287] W. H. Munk. Proposed Uniform Procedure for Observing Waves and

Interpreting Instrument Records. Scripps Institute of Oceanography. Wave
Report 26. 1944.
[288] H. R. Seiwell. Sea Surface Roughness Measurements in Theory and

Practice, Annals. New York Academy of Science. 51 (3). 1949.
[289] N. F. Barber. Ocean Waves and Swell, Proceedings of the Institution of

Civil Engineers. Marine and Waterways Division. 1950.
[290] R. R. Putz. Wave-Height Variability; Prediction of Distribution Function,

Technical Report No. HE 116-318, Series No. 3, Issue No. 318, Institute of
Engineering Research. Berkeley, CA. University of California. 1950.
[291] M. S. Longuet-Higgins. On the Statistical Distribution of the Heights of

Sea Waves. Journal of Marine Research . 2 (3). 1952.
[292] R. R. Putz. Statistical Distribution for Ocean Waves. Transactions,

American Geophysical Union . 33 (5). 1952.
[293] J. Darbyshire. The Generation of Waves by Wind. Proceedings of the

Royal Society Ser. A 215. 1952.
[294] T. Hamada, H. Mitsuyasu, H., and N. Hose. An Experiemental Study of

Wind Effect upon Water Surface. Tokyo : Transportation Technical Research
Institute. 1953.
309
[295] R. L. Wiegel. An Analysis from Data Wave Recoreders on the Pacific
Coast of the United states. Transactions, Americal Geophysical Union. 30,
1949.
[296] A. Kolios, M. Collu, A. Chahardehi, F. P. Brennan, M.H . and Patel.

Multi-Criteria Decision Making Method to Compare Support Structures for
Offshore Wind Turbines. Warsaw : European Wind Energy Conference -
EWEC, April 2010.
[297] W. Musial, S. Butterfield, and B. Ram. Energy from offshore wind.

Houston, Texas. Offshore Technology Conference. NREL/CP-500-39450. May
1–4, 2006.
[298] http://www.dnv.co.uk/services/software/products/sesam/. [Online]
[299] J. C. P. Kam, M. Birkinshaw, and J. V. Sharp. Review of the

Applications of Structural Reliability Technologies in Offshore Structural Safety.
USA : Proceedings of the 12th International Conference on Offshore Mechanics
and Artic Engineering, Vol 2. 1993.

Collapse Analysis of Jackets. Shinozukal, Yao, and Schueller. Structural Safety
and Reliability. Rotterdam: Balkema. 1994.
[301] P. S. Tromans, and J. W. Van de Graaf. Substantiated Risk Assessment

of Jacket Structure. Journal of Waterway, Port Coastal and Ocean Engineering.
Vol 120 (6). 535-555. 1992.
[302] P. A. Frieze, A. C. Morandi, M. Birkinshaw, D. Smith, and A. T. Dixon.

Fixed and Jack-up Platforms: Basis for Reliability Assessment. Journal of
Marine Structure. Vol 10, 263-284. 1997.
[303] Det Norske Veritas. Fatigue Design of Offshore Steel Structures. DNV.
DNV-RP-C203 . April 2010 .
310
[304] Det Norske Veritas. Rules for the Classification of Steel Ships. DNV,
April 2005.
[305] Det Norske Veritas. Design of Offshore Wind Turbine Structures. DNV,
DNV-OS-J101. October 2007.
[306] M-H Tsai. An analytical methodology for the dynamic amplification factor
in progressive collapse evaluation of building structures. Mechanics Research
Communications. Volume 37, Issue 1, Pages 61-66 , January 2010.
[307] M. K. Chryssanthopoulos, and T. D. Righiniotis. Fatigue Reliability of

welded steel structures. Journal of Constructional Steel Research. Vol. 62, pp.
1199-1209. 2006.
[308] E. Ayala-Uraga, and T. Moan. Fatigue reliability-based assessment of

welded joints applying consistent fracture mechanics formulations. International
Journal of Fatigue. Vol. 29, pp. 444-456. 2007.
[309] T. Moan, G. O. Godve, and A. M. Blanker. Reliability based fatigue

design criteria for offshore structures considering the effect of inspection and
repair. Offshore Technology Conference 7189. pp. 591-600. 1993.
[310] F. Moses. Final Reports for API PRAC 79-22 (1980), 80-22 (1981), 81-22
(1982), 82-22 (1983), 83-22 (1985), 85 –22 (1986), 87-22 (1987). Dallas :
American Petroleum Institute.
[311] S. K. Chakrabati. Hydrodynamics of Offhsore Structures. Boston :

Computational Mechanics Publications. 1987.
[312] L. Skjelbreia, and J. A. Hendrickson. Fifth-order Gravity Wave Theory.

Netherlands: ASCE, 184-196. Procedings of the 7th Conference on Costal
Engineering. 1961.
[313] R. G. Dean. Steam Function Representation of Nonlineac Ocean Waves.

Journal of Geophysical Research. 70(18): 4561-4572. 1965.
311
[314] S. N. Rasmussen. Corrosion Protection of Offshore Structures.
Denmark : HEMPEL A/S. 2005.
[315] Norwegian Technology Standards Institution. M-501 - Surface

preparation and protective coating. Norway. NORSOK. June 2004 .
[316] International Standardization Organization. ISO 12944:1998 (part 1 –8)

- Paints and varnishes - Corrosion protection of steel structures by protective
paint systems. Geneve, Switzerland : ISO. 1998.
[317] International Standardization Organization. ISO 20340:2003-Paints

and varnishes - Performance requirements for protective paint systems for
offshore and related structures. Geneve, Switzerland : ISO. 2003.
[318] A. W. Peabody. Peabody's Control of Pipeline Corrosion, 2nd Ed. s.l. :

NACE International. 2001.
[319] American Society for Testing and Materials. B843 - 07-Standard

Specification for Magnesium Alloy Anodes for Cathodic. ASTM . 1998.
[320] American Society for Testing and Materials. B418 - 09 Standard

Specification for Cast and Wrought Galvanic Zinc Anodes. ASTM , 2009.
[321] R. L. Kean, and K. G. Davies. Cathodic Protection. DTI under contract

from NPL. 2000.
312
APPENDICES
APPENDIX A
In this Appendix, complementary to what has been presented in Chapter Four,

the different wave theories will be briefly presented in a way that program
modelling for combination with FEA is feasible.
A.1. Linear Wave Theory
This theory, also known as amplitude, sinusoidal or Airy wave theory is based
on the assumption that the wave height is much smaller than both the wave
length and the water depth. Considering the boundary conditions set, the free
surface boundary conditions can be linearized by neglecting the wave height
term beyond the first order and are satisfied at the mean water level, rather than
at the oscillating free surface. Considering small amplitude waves, the free
surface boundary conditions are reduced to:
߲ߔ ߲ߟ
− = 0 ܽ‫ = ݖ ݐ‬0
߲‫ݐ߲ ݖ‬
(A.1-1)
߲ߔ
+ ݃ ∙ ߟ = 0 ܽ‫ = ݖ ݐ‬0
߲‫ݐ‬
The free surface profile is given by the following equation:
1 ߲ߔ
ߟ= − ∙ ൬ ൰ ܽ‫ = ݖ ݐ‬0 (A.1-2)
݃ ߲‫ݐ‬
Combining the two boundary conditions, eliminating ߟ, gives:
߲ଶߔ ߲ߔ
+݃∙ = 0 ܽ‫ = ݖ ݐ‬0 (A.1-3)
߲‫ݐ‬ଶ ߲‫ݖ‬
Assuming the velocity potential ߔ to be described with the following formula and
substituting to the basic Laplace equation:
ߔ = ߄(‫ ݔ(ߔ)ݖ‬− ܿ‫)ݐ‬ (A.1-4)
313
߲ଶܼ
− ݇ଶܼ = 0
߲‫ݖ‬ଶ
߲ଶߔ
− ݇ଶߔ = 0
߲‫ݔ‬ଶ
Where: ݇ଶ is a constant suitably chosen. Integrating the above differential

equations, the general solutions of ߄ and ߔ are:
ܼ = ‫ܣ‬ଵ ∙ cosh(݇‫ )ݖ‬+ ‫ܣ‬ଶ ∙ sinh(݇‫)ݖ‬ (A.1-5)
ߔ = ‫ܣ‬ଷ ∙ cos[݇(‫ ݔ‬− ܿ‫ ])ݐ‬+ ‫ܣ‬ସ ∙ sin[݇(‫ ݔ‬− ܿ‫])ݐ‬ (A.1-6)
Where: Α1, Α2, Α3 and Α4 are constants to be determined by the boundary

conditions, therefore:
‫ܣ‬ଶ = ‫ܣ‬ଵ = tanh(݇݀)
(A.1-7)
‫ =ݐ‬0 → ‫ = ݔ‬0 → ‫ܣ‬ଷ = 0
Rewriting the above equations, the velocity potential maybe written in the form:
cosh[݇(‫ݖ‬+ ݀)]
ߔ = ‫∙ܣ‬ ∙ sin[݇(‫ ݔ‬− ܿ‫])ݐ‬ (A.1-8)
sinh(݇݀)
Where: ‫ܣ = ܣ‬ଵ ∙ ‫ܣ‬ସ. ‫ ܣ‬can be derived as a function of ‫ ܪ‬, ݇(‫ܮ‬, ܶ), and ܿ(‫ܮ‬, ܶ) as
follows in order to derive ‫ܣ‬ସ:
௚ு ௚ு
‫ = ܣ‬ଶ௞௖ , ‫ܣ‬ସ = ଶ௞௖୲ୟ୬୦(௞ௗ) (A.1-9)
The velocity potential may finally be derived as:
(Error!
Reference
source not
݃‫ ∙ ܪ‬cosh(݇‫)ݏ‬ ߨ‫ ∙ ܪ‬cosh(݇‫)ݏ‬
ߔ= ∙ sin ߠ = ∙ sin ߠ found.Error!
2߱ ∙ sinh(݇݀) ݇ܶ ∙ sinh(݇݀)
Reference
source not
found.-10)
Where: ‫ݖ = ݏ‬+ ݀ and ߠ = ݇ ∙ (‫ ݔ‬− ܿ‫ ݔ݇ = )ݐ‬− ߱‫ݐ‬is the wave phase angle.
314
Having obtained a solution for ߔ , other variables of interest might be calculated
based on formulations presented in Table 64 [194] .
Shallow Water Deep Water

గ ௗ ଵ ௗ ௗ ଵ ௗ
Range of validity ݇݀ < , < , < 0.0025 ݇݀ < ߨ, < , < 0.08
ଵ଴ ௅ ଶ଴ ௚்మ ௅ ଶ ௚்మ
గ௴ ௚௴ ߨ߅ ௞௭ ݃߅ ௞௭
Velocity potential ߮= ∙ ‫= ߠ݊݅ݏ‬ ∙ ‫ߠ݊݅ݏ‬ ߮= ∙ ݁ ∙ ‫= ߠ݊݅ݏ‬ ∙ ݁ ∙ ‫ߠ݊݅ݏ‬
௞మఁௗ ଶఠ ݇ߒ 2߱
߱ଶ ߱ଶ ݃
Dispersion relation ܿଶ = ݃݀ ܿଶ = ܿ଴ଶ ∙ =
݇ଶ ݇ଶ ݇
Wave length ‫ܶ = ܮ‬ඥ݃݀ ‫ܮ = ܮ‬଴ = ݃ܶଶ⁄2ߨ
߅ ߅
Surface elevation ߟ= ∙ ܿ‫ߠݏ݋‬ ߟ= ∙ ܿ‫ߠݏ݋‬
2 2
Horizontal particle ߅ ߅
displacement ߦ= − ∙ ‫ߠ݊݅ݏ‬ ߦ= − ∙ ݁௞௭ ∙ ‫ߠ݊݅ݏ‬
2݇݀ 2݇݀
Vertical particle ߅ ‫ݖ‬ ߅ ௞௭

displacement ߞ= ∙ ቀ1 + ቁ∙ ܿ‫ߠݏ݋‬ ߞ= ∙ ݁ ∙ ܿ‫ߠݏ݋‬
2 ݀ 2
ߨ߅ ߨ߅ ௞௭
Horizontal particle velocity ‫=ݑ‬ ∙ ܿ‫ߠݏ݋‬ ‫=ݑ‬ ∙ ݁ ∙ ܿ‫ߠݏ݋‬
ܶ(݇݀) ܶ
ߨ߅ ‫ݖ‬ ߨ߅ ௞௭
Vertical particle velocity ‫=ݓ‬ ∙ ቀ1 + ቁ∙ ‫ߠ݊݅ݏ‬ ‫=ݓ‬ ∙ ݁ ∙ ‫ߠ݊݅ݏ‬
ܶ ݀ ܶ
Horizontal particle ߲‫ݑ‬ 2ߨଶ߅ ߲‫ ݑ‬2ߨଶ߅ ௞௭

acceleration = ଶ ∙ ‫ߠ݊݅ݏ‬ = ∙ ݁ ∙ ‫ߠ݊݅ݏ‬
߲‫)݀݇( ܶ ݐ‬ ߲‫ݐ‬ ܶଶ
Vertical particle ߲‫ݓ‬ 2ߨଶ߅ ‫ݖ‬ ߲‫ݓ‬ 2ߨଶ߅

acceleration = − ଶ ∙ ቀ1 + ቁ∙ ܿ‫ߠݏ݋‬ = − ଶ ∙ ݁௞௭ ∙ ܿ‫ߠݏ݋‬
߲‫ݐ‬ ܶ ݀ ߲‫ݐ‬ ܶ
1 1
Pressure ‫ = ݌‬−ߩ݃‫ݖ‬+ ߩ݃‫ߠݏ݋ܿ ∙ ܪ‬ ‫ = ݌‬−ߩ݃‫ݖ‬+ ߩ݃‫݁ ܪ‬௞௭ ∙ ܿ‫ߠݏ݋‬
2 2
Group velocity ‫ܥ = ீܥ‬ ‫ = ீܥ‬1⁄2 ‫ܥ‬
Average energy density ‫ = ܧ‬1⁄8 ߩ݃‫ ܪ‬ଶ ‫ = ܧ‬1⁄8 ߩ݃‫ ܪ‬ଶ
Energy flux ܲ = ‫ܧ‬஼ ܲ = 1⁄2 ‫ܧ‬஼
ܵ௑௑ = 3⁄2 ‫ܧ‬, ܵ௑௒ = ܵ௒௑ = 0, ܵ௑௑ = 1⁄2 ‫ܧ‬, ܵ௑௒ = ܵ௒௑ = 0
Radiation stress
ܵ௒௒ = 1⁄2 ‫ܧ‬ ܵ௒௒ = 0
Table 64: Shallow and deep water approximation to linear wave theory
[194]
315
A.2. Stokes Finite Amplitude Wave Theory
The linear wave theory, as it was presented in the previous section, can provide
an initial, rather simplistic approximation of the water particles motion due to
wave loading. For a more accurate and complete solution, the perturbation
technique might be used [311]. Stokes, has proposed an extension of the linear
wave theory for waves of finite height when the following conditions are met:
ு ு
≪ (݇݀)ଶ for ݇݀ < 1, and ≪1 (A.2-1)
ௗ ௅
The velocity potential maybe written as:
ߔ = ෍ ߝ௡ ߔ௡ (A.2-2)
௡ି௜
Where: ࢿ is the perturbation parameter, defined in terms of wave slope (H/L) as

ε = kH/2, and ࢶ ௡ is the nth order solution for ߔ . Substituting the above
expression in the basic Laplace equation and considering the boundary
conditions at the seabed and separation of terms of different polynomial orders,
the following equations may be derived:
߲ଶߔ௡ ߲ଶߔ௡
+ = 0 ݂‫ = ݊ ݎ݋‬1,2, …
߲‫ݔ‬ଶ ߲‫ݖ‬ଶ
(A.2-3)
߲ߔ௡
= 0 ܽ‫ = ݖݐ‬−݀ ݂‫ = ݊ݎ݋‬1,2, …
߲‫ݖ‬
From the above equations set, the difficulty of defining the free surface
boundary conditions due to nonlinear term consisting of product exists, in order
to apply them to the unknown surface elevation rather than mean waterline. In
[312], an expansion of the Stokes wave theory to the fifth order has raised a
widely used method in engineering practice. The velocity potential ߔ and the
free surface elevation ߟ, are written in a series form of fifth order as:
316
ହ
ܿ
ߔ = ෍ ߔ௡ᇱ cosh(݊݇‫ )ݏ‬sin(݊ߠ) (A.2-4)
݇
௡ି௜
ହ
1
ߟ = ෍ ߟ௡ᇱ cos(݊ߠ) (A.2-5
݇
௡ି௜
In the above formulas, the coefficients ߔ௡′ and ߟ௡′ are expressed as functions of
ߣ and a series of coefficients Α and Β, as are presented in Table 65, Table 66
and Table 67. The wave celerity is now:
ܿଶ = ‫ܥ‬଴ଶ ∙ (1 + ߣଶ ∙ ‫ܥ‬ଵ + ߣସ ∙ ‫ܥ‬ଶ) (A.2-6)
Where: ‫ܥ‬଴ is the celerity given by linear wave theory for the same depth and
wave number:
‫ܥ‬଴ଶ = (݃⁄݇) ∙ ‫݊ܽݐ‬ℎ ݇݀ (A.2-7)
Considering appropriate values for the coefficients ‫ ܤ‬and ‫ܥ‬, parameters ߣ and
݇ should be initially defined solving the system:
1 ‫ܪ‬
∙ [ߣ + ‫ܤ‬ଷଷߣଷ + (‫ܤ‬ଷହ + ‫ܤ‬ହହ) ∙ ߣହ] =
݇݀ 2݀
݀ (A.2-8)
݇݀ ∙ tanh(݇݀) ∙ [1 + ‫ܥ‬ଵߣଶ + ‫ܥ‬ଶߣସ] = 4࣊ ଶ ∙ ଶ
݃ܶ
The final values of ࣅ and ݇ are determined from above equation through an
iterative procedure.
߮ଵᇱ = ߣ޿ଵଵ + ߣଷ޿ଵଷ + ߣହ޿ଵହ ߟଵᇱ = ߣ
߮ଶᇱ = ߣଶ޿ଶଶ + ߣସ޿ଶସ ߟଶᇱ = ߣଶ߀ଶଶ + ߣସ߀ଶସ
߮ଷᇱ = ߣଷ޿ଷଷ + ߣହ޿ଷହ ߟଷᇱ = ߣଷ߀ଷଷ + ߣହ߀ଷହ
߮ସᇱ = ߣସ޿ସସ ߟସᇱ = ߣସ߀ସସ
߮ହᇱ = ߣହ޿ହହ ߟହᇱ = ߣହ߀ହହ
Table 65: Φn' and ηn' the coefficients [194]
317
ହ
Velocity potential, ߮ k߮
= ෍ ߮୬ᇱ ∙ cosh(nks) ∙ sin(nߠ)
c
୬ୀଵ
ܿଶ tanh(݇݀)
Wave length, ܿ = ∙ [1 + ߣଶ‫ܥ‬ଵ + ߣସ‫ܥ‬ଶ]
݃݀ ݇݀
ହ
Surface elevation, ߟ ݇ߟ = ෍ ߟ୬ᇱ ∙ cos(nߠ)
୬ୀଵ
ହ
Horizontal particle velocity, ‫ݑ‬ ‫ݑ‬
= ෍ n߮୬ᇱ ∙ cosh(nks) ∙ cos(nߠ)
ܿ
୬ୀଵ
ହ
Vertical particle velocity, ‫ݓ‬ ‫ݓ‬
= ෍ n߮୬ᇱ ∙ sinh(nks) ∙ sin(nߠ)
ܿ
୬ୀଵ
ହ
Horizontal particle acceleration, ߲‫ݑ‬⁄߲‫ݐ‬
߲‫ݑ‬⁄߲‫ݐ‬ = ෍ nଶ߮୬ᇱ ∙ cosh(nks) ∙ sin(nߠ)
߱ܿ
୬ୀଵ
ହ
Vertical particle acceleration, ߲‫ݑ‬⁄߲‫ݐ‬
߲‫ݑ‬⁄߲‫ݐ‬ = − ෍ nଶ߮୬ᇱ ∙ sinh(nks) ∙ cos(nߠ)
߱ܿ
୬ୀଵ
ହ
Temporal derivative of ߮ ߲߮ ⁄߲‫ݐ‬
= − ෍ ݊߮୬ᇱ ∙ cosh(nks) ∙ cos(nߠ)
ܿଶ
୬ୀଵ
Pressure, ‫݌‬ ‫݌‬ ‫ܿ ݏ‬ଶ ߲߮ ⁄߲‫ ݐ‬1 ‫ ݑ‬ଶ ‫ ݓ‬ଶ

=1− − ቊ ଶ + ൤ቀ ቁ + ቀ ቁ ൨ቋ
ߩ݃݀ ݀ ݃݀ ܿ 2 ܿ ܿ
Table 66: Stokes fifth-order wave theory [194]
318
‫ܣ‬ଵଵ = 1⁄ܵ
−‫(ܥ‬5‫ܥ‬ଶ + 1)
‫ܣ‬ଵଷ =
8ܵଶ
−(1184‫ܥ‬ଵ଴ − 1440‫ ଼ܥ‬− 1992‫ ଺ܥ‬+ 2641‫ܥ‬ସ − 249‫ܥ‬ଶ + 18)
‫ܣ‬ଵହ =
1536ܵଵଵ
‫ܣ‬ଶଶ = 3⁄8ܵସ
(192‫ ଼ܥ‬− 424‫ ଺ܥ‬− 312‫ܥ‬ସ + 480‫ܥ‬ଶ − 17)
‫ܣ‬ଶସ =
768ܵଵ଴
(13 − 4‫ܥ‬ଶ)
‫ܣ‬ଷଷ =
64ܵ଻
(512‫ܥ‬ଵଶ − 4224‫ܥ‬ଵ଴ − 6800‫ ଼ܥ‬− 12,808‫ ଺ܥ‬+ 16,704‫ܥ‬ସ − 3154‫ܥ‬ଶ + 107)
‫ܣ‬ଷହ =
4096ܵଵଷ ∙ (6‫ܥ‬ଶ − 1)
(80‫ ଺ܥ‬− 816‫ܥ‬ସ + 1338‫ܥ‬ଶ − 197)
‫ܣ‬ସସ =
1536ܵଵ଴ ∙ (6‫ܥ‬ଶ − 1)
−(2880‫ܥ‬ଵ଴ − 72,480‫ ଼ܥ‬+ 324,000‫ ଺ܥ‬− 432,000‫ܥ‬ସ + 163,470‫ܥ‬ଶ − 16,245)
‫ܣ‬ହହ =
61,440ܵଵଵ ∙ (6‫ܥ‬ଶ − 1) ∙ (8‫ܥ‬ସ − 11‫ܥ‬ଶ + 3)
(2‫ܥ‬ଶ + 1)
‫ܤ‬ଶଶ = ‫∙ ܥ‬
4ܵଷ
(272‫ ଼ܥ‬− 504‫ ଺ܥ‬− 192‫ܥ‬ସ + 322‫ܥ‬ଶ + 21)
‫ܤ‬ଶଶ = ‫∙ ܥ‬
384ܵଽ
3 ∙ (8‫ ଺ܥ‬+ 1)
‫ܤ‬ଷଷ =
64ܵ଺
(88,128‫ܥ‬ଵସ − 208,224‫ܥ‬ଵଶ + 70,848‫ܥ‬ଵ଴ + 54,000‫ ଼ܥ‬− 21,816‫ ଺ܥ‬+ 6264‫ܥ‬ସ − 54‫ܥ‬ଶ − 81)
‫ܤ‬ଷହ =
12,288ܵଵଶ ∙ (6‫ܥ‬ଶ − 1)
‫( ∙ ܥ‬768‫ܥ‬ଵ଴ − 448‫ ଼ܥ‬− 48‫ ଺ܥ‬+ 48‫ܥ‬ସ + 106‫ܥ‬ଶ − 21)
‫ܤ‬ସସ =
384ܵଽ ∙ (6‫ܥ‬ଶ − 1)
(192,000‫ܥ‬ଵ଺ − 262,720‫ܥ‬ଵସ + 83,680‫ܥ‬ଵଶ + 20,160‫ܥ‬ଵ଴ − 7280‫)଼ܥ‬
‫ܤ‬ହହ =
12,288ܵଵ଴ ∙ (6‫ܥ‬ଶ − 1) ∙ (8‫ܥ‬ସ − 11‫ܥ‬ଶ + 3)
(7160‫ ଺ܥ‬− 1800‫ܥ‬ସ − 1050‫ܥ‬ଶ + 225)
+
12,288ܵଵ଴ ∙ (6‫ܥ‬ଶ − 1) ∙ (8‫ܥ‬ସ − 11‫ܥ‬ଶ + 3)
8‫ܥ‬ସ − 8‫ܥ‬ଶ + 9
‫ܥ‬ଵ =
8ܵସ
(3840‫ܥ‬ଵଶ − 4096‫ܥ‬ଵ଴ + 2592‫ ଼ܥ‬− 1008‫ ଺ܥ‬+ 5944‫ܥ‬ସ − 1830‫ܥ‬ଶ + 147)
‫ܥ‬ଶ =
512ܵଵ଴ ∙ (6‫ܥ‬ଶ − 1)
‫ܥ‬ଷ = − 1⁄4ܵ‫ܥ‬
(12‫ ଼ܥ‬+ 36‫ ଺ܥ‬− 162‫ܥ‬ସ + 141‫ܥ‬ଶ − 27)
‫ܥ‬ସ =
192‫ܵܥ‬ଽ
Table 67: The coefficients for Stokes fifth-order wave theory [311]
319
A.3. Stream Function Wave Theory
A different method for the representation of non linear waves is that of the
steam function wave theory [313]. Two different types of this theory distinguish:
 Symmetric or Regular Stream Function Theory that describes periodic

waves of symmetric, permanent form with prescribed period, height and
still water depth.
 Irregular Stream Function Theory that represents a stream function and

associated kinematics of a wave with a predetermined profile. This
theory is suitable in analyzing wave tank or field test data.
The Irregular Stream Function wave theory sets no restrictions on the wave
form; therefore the wave can change form as it propagates through its motion. It
is more suitable in cases were a measured wave surface profile is available and
the water particle kinematics and dynamics are required for this wave profile
and is applied to a single wave in the wave profile (eg. the largest wave). The
Symmetric or Regular Stream Function Theory is the one tackled by other wave
theories which assumed to propagate at a constant speed and without a change
in its form.
This method will be presented for the special case of a constant free surface
pressure, neglecting any underlying current. Moving from a Cartesian to a
relative coordinates system, considering ܿҧto be the speed vector, the problem
is reduced to a steady flow, with the horizontal component of velocity to move in
the u–c plane. Proportionally to the previous theories, for the two dimensional
flow, a steam function ߖ can be formulated, satisfying the Laplace equation as
follows:
߲ଶߖ ߲ଶߖ
+ =0 (A.3-1)
߲‫ݔ‬ଶ ߲‫ݖ‬ଶ
Assuming ܳ to be the Bernoulli’s constant, the corresponding boundary

conditions are:
320
߲ߖ
= 0 ܽ‫ = ݖ ݐ‬−݀
߲‫ݔ‬
‫ݓ‬ ߲ߟ
= ܽ‫ߟ = ݖ ݐ‬ (A.3-2)
‫ ݑ‬− ܿ ߲‫ݔ‬
1 ߲ߖ ଶ ߲ߖ ଶ
ߟ+ ቈ൬ ൰ + ൬ ൰ ቉= ܳ ܽ‫ߟ = ݖ ݐ‬
2݃ ߲‫ݔ‬ ߲‫ݖ‬
In the Cartesian, two dimensional (‫ݔ‬, ‫ )ݖ‬system, the N-order symmetric steam
function ߖ is given as follows, satisfying the bottom boundary condition and the
kinematic free surface condition:
ࢸ (‫ݔ‬, ‫ݖܿ = )ݖ‬+ ෍ ܺ௡ ∙ sinh[݊݇ ∙ (‫ݖ‬+ ݀)] ∙ cos(݊݇‫)ݔ‬ (A.3-3)

௡ୀଵ
Where: the coefficients Xn, the wave number ݇ and the surface value of stream
function are determined using the dynamic free surface boundary condition for a
given ߅ , ߒ and ݀. Differentiation of the steam function can give the individual
horizontal and vertical water particle velocities as:
ே
߲∙ ࢸ
‫ ݑ‬− ܿ= − = −ܿ− ෍ ݊݇ܺ௡ ∙ cosh[݊݇(‫ݖ‬+ ݀)] ∙ cos(݊݇‫)ݔ‬
߲∙ ‫ݖ‬
௡ୀଵ
ே (A.3-4)
߲∙ ࢸ
‫=ݓ‬ = − ෍ ݊݇ܺ௡ ∙ sinh[݊݇(‫ݖ‬+ ݀)] ∙ sin(݊݇‫)ݔ‬
߲∙ ‫ݔ‬
௡ୀଵ
For the Bernoulli constant ܳ, the mean square value ‫ܧ‬, is defined as:
ூ
1
‫ܧ‬ത = ෍ (ܳ௜ ∙ ܳത)ଶ (A.3-5)
‫ܫ‬
௡ୀଵ
Where: i=1,…,l, x takes successive values spanning one complete wave length,
and ܳത represents the average value of the Bernoulli constant. Then, Xn, ݇ and
ߖ are obtained in an iterative way with an initial estimate derived from linear
theory of the wave number, ݇ and the stream function, ߖ such that ‫ܧ‬ത is
minimized.
321
APPENDIX B
B.1. Painting coatings
Coatings and combinations of coatings with cathodic protection (for the

immersion zone of the structure) is the most common method employed for the
protection of structures from corrosive environments. The actual performance of
coating systems and thus the efficiency of the CPS is subjected on several
fundamental parameters that should be considered in the design process. The
most important of those are the following:
 Type and condition of the substrate

 Environment and possible additional stresses
 Surface preparation
 Quality of the coatings
 Selection of the coating systems (Generic types, thickness etc.)
 Application
 Quality control
The substrate of interest in offshore structures is steel. This parameter refers to

the state where the structure is in the fabrication process and should illustrate
the basic specification of the coating system to be applied. The parameter of
environment includes apart from the obvious loading conditions, different
sources of stresses due to thermal, chemical or different nature. The
preparation of the surface is the most important parameter since it can
significantly damage the applied coating. The degree of cleaning (removal of
rust, particles, grease, salts etc.), the final roughness and the preparation of
sharp edges, welding seams and imperfections are critical elements towards
this scope. Selection of coatings should be appropriate and paint of good quality
should be selected and usually certified by appropriate classification bodies.
Coating systems, is the parameter that refers to the combination of different
layers of appropriate coating each providing a different level of protection to the
structure. In conjunction with coatings that target to limit corrosion, for the
322
immersion zone anti-fouling coatings should be used in order to delay or ease
removal of fouling. Table 68, proposes typical coating types and systems for
each of the areas of the offshore structure.
Proper application of the coating or the coating systems is crucial for the
maximization of the efficiency of the protection. It is important to apply the
prescribed thickness of film (within the tolerance range) since smaller thickness
might lead to premature corrosion while thicker layers might result in
undesirable effects such as solvent retention, reduced adhesion etc. [314]. The
problem of variation of the film thickness during application can be avoided by
applying more coats of the same total thickness distributing the variation to all
over the member.
Area Coating types Coating system
Zinc-rich primer, Epoxies

Atmospheric zone Minimum 320 μm/13 mils in minimum 3 coats.
and UV durable topcoat
Splash zone Epoxy or Polyester Minimum 600 μm/24 mils in minimum 2 coats.
Immersion Epoxy Minimum 450 μm/18 mils in minimum 2 coats.
Table 68: Typical coating systems used for offshore structures
Since coatings are applied most of the times manually, areas of the structure
where proper access is not available should be pointed out. Critical points of
this category are welding seams, edges, and corners. For those cases, stripe
coating, corrosion protection schemes or application of CPS within
manufacturing phase should be considered in order to sufficiently protect the
structure.
The wide application and great significance of the painting coatings has raised
an interest for systematic recording of the procedures and the requirements into
design and application standards. NORSOK [315] has published a standard
describing the surface preparation and protective coatings. Within its provisions,
different steps in the coating process are illustrated and the qualifications of the
painters, inspectors and coating systems are included. ISO [316] has published
a standard that illustrates the basic factors that determines the effectiveness of
323
the paint coating systems. These are summarised to the ‘choice and formulation
of the products used in differently classified environments and the standard of
workmanship and execution of their contract’. An additional relevant standard
published by ISO, [317], is a complementary standard to the previous one
referring to the special offshore environment, proposing tougher testing of the
coating systems. Finally, a series of publications from NACA (National
Association of Corrosion Engineers) are available covering different aspects of
application of the method.
B.2. Cathodic protection
Cathodic protection is a widely used technique nowadays to control the

corrosion on a metal surface by transforming it to the cathode of an
electrochemical cell [318]. Initially introduced in the early 1820s by Davy, it
became practically applicable in 1945 and had a rapid development since then.
Application of the method can be found in pipelines, ship hulls, storage tanks,
harbour structures, tubular and foundation pilings and offshore platforms,
floating and subsea structures.
The basic concept of cathodic protection requires connection of the metal to be

protected to another more easily corroded metal that will act as an anode of the
electrochemical cell, as is presented in Figure 73. Three different types of
cathodic protection methods are available:
 Galvanic or Sacrificial Corrosion Protection

 Impressed-Current Corrosion Protection
 Galvanization
For the Galvanic or Sacrificial Corrosion Protection, reactive metals are

employed as anodes that are directly electrically connected to the steel that
needs to be protected. In order to achieve transformation of the structure to a
cathode, it becomes negatively charged by considering the natural potentials
between the anode and the steel, causing a positive current to flow in the
electrolyte, from the anode to the steel. Metals commonly used as sacrificial
324
anodes are zinc, aluminium and magnesium. Standards that describe
application of cathodic protection can be found in [319] and [320].
Impressed-Current Corrosion Protection Systems, that refer mainly to structures

of larger scale, where application of Galvalnic anodes is not economically
feasible, employ inert (zero or low dissolution) anodes and use an external
source of dc power (rectified ac) to impress a current from an external anode
onto the cathode surface. Those anodes are of tubular, rod-type shaped or
continuous ribbons of materials such as high silicon cast iron, graphite, mixed
metal oxide, platinum and niobium wires.
Galvanizing is a technique that accounts for coating steel with a layer of metallic
zinc. This coating is very durable even in harsh environments, since they
combine benefits of coating and cathodic protection. Even in cases with local
damage to the treatment, the method retains effectiveness since the
surrounded areas of the ‘wounded’ one form a galvanic cell with the exposed
steel and protect it from corrosion, as a form of corrosion protection where zinc
acts as an anode.
A very important parameter that should be considered for the more effective
performance of the cathodic protection is the cathodic ‘shielding effect’. This is
based on the fact that in practice solid film backed anti-corrosion coatings such
as polyethylene tapes, shrinkable pipeline sleeves, and factory applied single or
multiple solid film coatings are used. The high electrical resistivity of those
patches protects electrical current to flow from the cathodic protection, blocking
it from reaching to the metal to be protected.
Table 69, adopted by [321], presents the requirements of a Galvanic and an

Impressed-Current Corrosion Protection System.
325
Figure 73: Corrosion Cell
Galvanic System Impressed-Current System
Inert anodes (clusters of which, connected

Sacrificial anodes together often in a backfill, are called the
“groundbed”)
Direct welding to the structure or a conductor

A dc power source.
connecting the anode to the structure
Secure and minimum resistance connections Electrically well insulated, minimum resistance
between conductor and structure, and between and secure conductors between anodes and
conductor and anode. power source
Secure and minimum resistance connections

between power source and structure
Table 69: Requirements of Galvanic and an Impressed-Current Corrosion

Protection System
326
APPENDIX C
In this Appendix, the details regarding the geometry of the reference structure is
given. In Figure 74, the numbering of the structural members is presented.
Further in Table 70, the dimension of each group of members is included.
t51X
bX0X
b45X
b41X
b35X
b31X
b25X
b21X
b15X
b11X
pX00
Figure 74: Numbering of the structural members
327
Type ID Diameter Thickness
(m) (mm)
Legs bX0X 1.25 0.028
Piles pX00 1.5 0.024
V1 b15X 0.6 0.018
V2 b25X 0.6 0.018
V3 b35X 0.6 0.018
V4 b45X 0.6 0.018
H1 b11X 0.6 0.018
H2 b21X 0.6 0.018
H3 b31X 0.5 0.016
H4 b41X 0.5 0.016
Topside t51X 0.6 0.016
Table 70: Dimension of group of members
328
APPENDIX D
% Matlab Code for the Response Surface Method

% This code requires input of a matrix with the geometry of the
% structure,and the corresponding loads and properties as well as a
% complementary FEA code for the execution of simulations and a code
% for regression analysis.
% Output of the code is a vector of PFs and Betas for each of the
% members of the complex structure.
%
% Athanasios Kolios - Cranfield University 2010
clear
clc
format long e
tic
a=open('Data23.mat');
Coord=a.Coords;
Con=a.Connections;
Re=a.BC;
Load=a.Loading;
E=a.Elasticity;
A=a.Area;
er=10^-50;
x1_m=500;
x2_m=21e10;
x3_m=0.5;
x4_m=1000000;
x1_s=200;
x2_s=0.1e10;
x3_s=0.01;
x4_s=100000;
A=[x1_m-3*x1_s x1_m x1_m+3*x1_s];

B=[x2_m-3*x2_s x2_m x2_m+3*x2_s];
329
D=[x3_m-3*x3_s x3_m x3_m+3*x3_s];
E=[x4_m-3*x4_s x4_m x4_m+3*x4_s];
c=num2cell([A; B; D; E],2);
[c{:}] = ndgrid(c{:});
c=reshape(cat(length(c)+1,c{:}),[],length(c));
siz=size(c);
[nel,~]=size(Con);
[n,~]=size(c);
[xL,yL]=size(Load);
for i=1:n
L=Load;
for k=1:xL
for l=1:yL
if L(k,l)~=0
L(k,l)=c(i,1);
end
end
end
[~,~,SAX]=FEA(Coord,Con,Re,L*c(i,1),c(i,2)*ones(nel,1),c(i,3)*on
es(nel,1),1);
S_all=c(i,4)*ones(1,nel);
G(i,:)=abs(SAX)./S_all;
end
for i=1:nel
y_var=G(:,i);
x_var=c;
b=reg_an(x_var,y_var,2);
[Pf_form,b_form]=FORM_SORM(b,x1_m,x2_m,x3_m,x4_m,x1_s,x2_s,x3_s,x4_s);
Pf(i)=Pf_form;
beta(i)=b_form;
end
toc
%end of program
330
% Matlab Code for Adaptive Response Surface Method
% This code requires input of a matrix with the geometry of the
% structure, and the corresponding loads and properties as well as a
% complementary FEA code for the execution of simulations and a code
% for regression analysis.
% Weighted Regression is incorporated after the first iteration in
% order to ‘tune’ the accuracy of the response surface.
% Output of the code is a vector of PFs and Betas for each of the
% members of the complex structure.
%
clear
clc
format long e
tic
a=open('Data23.mat');
Coord=a.Coords;
Con=a.Connections;
Re=a.BC;
Load=a.Loading;
E=a.Elasticity;
A=a.Area;
x1_m=1;
x2_m=21e10;
x3_m=1;
x4_m=100000;
x1_s=0.2;
x2_s=1e10;
x3_s=0.01;
x4_s=10000;
[nel,~]=size(Con);
[xL,yL]=size(Load);
er=10^-40;
331
% Step 1.1
[~,~,SAX]=FEA(Coord,Con,Re,Load,x2_m*ones(nel,1),A,1);
S_all=x4_m*ones(1,nel);
Gmean=abs(SAX)./S_all;
%Step 1.2-1.5 : Load

L=Load;
for k=1:xL
for l=1:yL
if L(k,l)~=0
L(k,l)=(x1_m-3*x1_s)*Load(k,l);
end
end
end
[~,~,SAX]=FEA(Coord,Con,Re,L,x2_m*ones(nel,1),A,1);
GL=abs(SAX)./S_all;
if GL(18)<Gmean(18)
kLoad=-3;
else
kLoad=3;
end
%Step 1.2-1.5 : Elasticity

[~,~,SAX]=FEA(Coord,Con,Re,Load,(x2_m-3*x2_s)*ones(nel,1),A,1);
GE=abs(SAX)./S_all;
if GE(18)<Gmean(18)
kElast=-3;
else
kElast=3;
end
%Step 1.2-1.5 : Area

[~,~,SAX]=FEA(Coord,Con,Re,Load,x2_m*ones(nel,1),(x3_m-3*x3_s)*A,1);
GA=abs(SAX)./S_all;
if GA(18)<Gmean(18)
kArea=-3;
332
else
kArea=3;
end
%Step 1.2-1.5 : Stress Allowed

[~,~,SAX]=FEA(Coord,Con,Re,Load,x2_m*ones(nel,1),A,1);
S_all=(x4_m-3*x4_s)*ones(1,nel);
GSA=abs(SAX)./S_all;
if GSA(18)<Gmean(18)
kSall=-3;
else
kSall=3;
end
%Step 1.6 Calculation of ybest

L=Load;
for k=1:xL
for l=1:yL
if L(k,l)~=0
L(k,l)=(x1_m+kLoad*x1_s)*Load(k,l);
end
end
end
[~,~,SAX]=FEA(Coord,Con,Re,L,(x2_m+kElast*x2_s)*ones(nel,1),(x3_m+kAre
a*x3_s)*A,1);
S_all=(x4_m+kSall'*x4_s)*ones(1,nel);
ybest=abs(SAX)./S_all+er*ones(1,nel);
%Design matrix
c=[x1_m x2_m x3_m x4_m;
x1_m+kLoad*x1_s x2_m x3_m x4_m;
x1_m x2_m+kElast*x2_s x3_m x4_m;
x1_m x2_m x3_m+kArea*x3_s x4_m;
x1_m x2_m x3_m x4_m+kSall*x4_s];
[n,~]=size(c);
%Step 1.6 Calculation of W
W=zeros(n,nel);
for j=1:n
L=Load;
333
for k=1:xL
for l=1:yL
if L(k,l)~=0
L(k,l)=c(j,1)*Load(k,l);
end
end
end
[~,~,SAX]=FEA(Coord,Con,Re,L,c(j,2)*ones(nel,1),c(j,3)*A,1);
S_all=c(j,4)*ones(1,nel);
G(j,:)=abs(SAX)./S_all;
W(j,:)= exp(-(G(j,:)-ybest)./ybest);
end
%Step 2-3 Weighted regression, FORM/SORM

for j=1:nel
x_var=c;
[xs,ys]=size(x_var);
regs=ys+1;
for i=1:xs;
X=x_var;
X(i,regs)=1;
end;
Y=G(:,j);
Weight=diag(W(:,j));
b(:,j)=(X'*Weight*X)\(X'*Weight*Y);
%b(:,j)=inv(X'*Weight*X)*(X'*Weight*Y);
% Definition of Limit State Functions

syms x1 x2 x3 x4
a1=b(1,j)+er;
a2=b(2,j)+er;
a3=b(3,j)+er;
a4=b(4,j)+er;
a5=b(5,j)+er;
%a6=b(6,j)+er;
%a7=b(7,j)+er;
%a8=b(8,j)+er;
%a9=b(9,j)+er;
334
%g=1-(a1*x1^2+a2*x1+a3*x2^2+a4*x2+a5*x3^2+a6*x3+a7*x4^2+a8*x4+a9);
g=1-(a1*x1+a2*x2+a3*x3+a4*x4+a5);
% Calculation of Partial derivatives

dgx1=diff(g,x1);
dgx2=diff(g,x2);
dgx3=diff(g,x3);
dgx4=diff(g,x4);
% Express functions in a amatlab complex form

g_m=matlabFunction(g);
dgx1_m=matlabFunction(dgx1);
% First Iteration
i=1;
%Initial point (mean)

x1_v(i,j)=x1_m;
x2_v(i,j)=x2_m;
x3_v(i,j)=x3_m;
x4_v(i,j)=x4_m;
g_v(i,j)=g_m(x1_v(i,j),x2_v(i,j),x3_v(i,j),x4_v(i,j));
%dgx1_v(i)=dgx1_m();
b_v(i,j)=g_v(i,j)/((((dgx1_m()*x1_s)^2)+((dgx2_m()*x2_s)^2)+((dgx3_m()
*x3_s)^2)+((dgx4_m()*x4_s)^2))^0.5);
a1_v(i,j)=-
(dgx1_m()*x1_s)/((((dgx1_m()*x1_s)^2)+((dgx2_m()*x2_s)^2)+((dgx3_m()*x
3_s)^2)+((dgx4_m()*x4_s)^2))^0.5);
335
a2_v(i,j)=-
3_s)^2)+((dgx4_m()*x4_s)^2))^0.5);
a3_v(i,j)=-
3_s)^2)+((dgx4_m()*x4_s)^2))^0.5);
a4_v(i,j)=-
3_s)^2)+((dgx4_m()*x4_s)^2))^0.5);
%New design point

x1_v(i+1,j)=x1_m+b_v(i,j)*x1_s*a1_v(i,j);
g_v(i+1,j)=g_m(x1_v(i+1),x2_v(i+1),x3_v(i+1),x4_v(i+1));
end
%Step 4 New centre point

Xmean=[x1_v(1,:); x2_v(1,:); x3_v(1,:); x4_v(1,:)];
Xdesign=[x1_v(2,:); x2_v(2,:); x3_v(2,:); x4_v(2,:)];
for i=1:nel
alpha(i)=g_v(1,i)/(g_v(1,i)-g_v(2,i));
for j=1:4
Xnew(j,i)=Xmean(j,i)+(Xdesign(j,i)-Xmean(j,i))*alpha(i);
end
end
for i=1:nel
%Step 5 New design matrix
c2=[x1_m x2_m x3_m x4_m;
x1_m+kLoad*x1_s x2_m x3_m x4_m;
x1_m x2_m+kElast*x2_s x3_m x4_m;
x1_m x2_m x3_m+kArea*x3_s x4_m;
x1_m x2_m x3_m x4_m+kSall*x4_s
Xnew(1,i) x2_m x3_m x4_m;
336
x1_m Xnew(2,i) x3_m x4_m;
x1_m x2_m Xnew(3,i) x4_m;
x1_m x2_m x3_m Xnew(4,i);];
[n,~]=size(c2);
%Step 6 Calculation of W
W=zeros(n,nel);
for j=1:n
L=Load;
for k=1:xL
for l=1:yL
if L(k,l)~=0
L(k,l)=c2(j,1)*Load(k,l);
end
end
end
[~,~,SAX]=FEA(Coord,Con,Re,L,c2(j,2)*ones(nel,1),c2(j,3)*A,1);
S_all=c2(j,4)*ones(1,nel);
G(j,:)=abs(SAX)./S_all;
W(j,:)= exp(-(G(j,:)-
Gmean+er*ones(1,nel)))./(Gmean+er*ones(1,nel)));
end
%Step 7 Reliability Calculations

xexp=c2;
[xs,ys]=size(xexp);
regs=2*ys+1;
for k=1:xs;
X(k,regs)=1;
for j=1:ys;
X(k,2*j-1)=xexp(k,j)^2;
X(k,2*j)=xexp(k,j)^1;
end;
end;
Y=G(:,i);
Weight=diag(W(:,i));
b2=(X'*Weight*X)\(X'*Weight*Y);
337
[Pf_form,b_form]=FORM_SORM(b2,x1_m,x2_m,x3_m,x4_m,x1_s,x2_s,x3_s,x4_s)
;
Pf(i)=Pf_form;
beta(i)=b_form;
end
toc
338
% This code performs multivariate polynomial regression (MPR)
% analysis.
% A scaling of the design parameters is included in order to prevent
% from ill-conditioned systems.
% The vector y_var, includes values of dependent variables (y) while
% the matrix x_var includes the values of independent variables (xi).
%
% An example of 3 variables is included.
%
% Athanasios Kolios - Cranfield University 2010 {R)
clc
clear
format short g
% Input of dependent variables

x_var=[15 15 15 10 10 10 5 5 5 15 15 15 10 10 10 5
5 5 15 15 15 10 10 10 5 5 5;
15 10 5 15 10 5 15 10 5 15 10 5 15 10 5 15 10 5
15 10 5 15 10 5 15 10 5;
600000 600000 600000 600000 600000 600000 600000 600000 600000
700000 700000 700000 700000 700000 700000 700000 700000 700000
800000 800000 800000 800000 800000 800000 800000 800000
800000;]';
%Input of independent variables

y_var=[0.784426667 0.655335 0.527383333 0.652808333 0.522951667
0.394113333 0.522345 0.39159 0.261475 0.672365714 0.561715714
0.452042857 0.55955 0.448244286 0.337811429 0.447724286 0.335648571
0.224121429 0.58832 0.49150125 0.3955375 0.48960625 0.39221375
0.295585 0.39175875 0.2936925 0.19610625]';
% power_f: This matrix is formed as the exponential factor of each of

the % variables for each coefficient.
% For the example of a 4th order regression and 2 variables, the
results will % return:
% power_f=[1,0;0,1;2,0;0,2;3,0;0,3;4,0;0,4;0,0;];
339
% obtain sizes
[n,p] = size(x_var);
% model terms loop
nv=p;
np=2; % degree of the polynomial
% Size of matrix transpose

k=nv*np;
% Formation of matrix with I's

B= repmat(eye(nv),np,1);
% Scaling of variables
std_s = sqrt(diag(cov(x_var)));
x_var_s = x_var*diag(1./std_s);
% Vector of coefficients
F=(1:1:np);
FT=transpose(F);
FTT=zeros(k,1);
% Expand F
for i=1:nv;
for j=1:np;
FTT(i+(j-1)*nv)=FT(j);
end;
end;
% Multiply with B and get BT

for i=1:k;
for j=1:nv;
power_f(i,j)=B(i,j)*FTT(i);
end;
end;
for i=1:nv;
power_f(k+1,i)=0;
340
end;
nt = size(power_f,1);
% build the design matrix

M = ones(n,nt);
sf = ones(1,nt);
for i = 1:nt;
for j = 1:p;
M(:,i) = M(:,i).*x_var_s(:,j).^power_f(i,j);
sf(i) = sf(i)/(std_s(j)^power_f(i,j));
end;
end
% Calculation of Coeff_fact
Coeff= M\y_var;
y_calc = M*Coeff(:);
Coeff_fact=transpose(Coeff)
% Reverse scaling
Coeff_fact=Coeff_fact'*sf;
% variance of the regression parameters

s_var = norm(y_var - y_calc);
% R^2
R2 = 1 - (s_var/norm(y_var-mean(y_var)) )^2;
% RMSE
RMSE = sqrt(mean((y_var - y_calc).^2));
clear B Coeff F FT FTT M i j k n p np nt nv
% End of code
341
% FORM-SORM_v1_0.m
%
% This code calculates the reliability of a four variable function,
% with second order polynomials, where variables are normal, according
% to the Hasofer, Lind, Rackwitz and Fiessler index method.
% A convergence criterion for the FORM is included in the code.
% SORM has been also incorporated.
%
% This version of the code has been verified with DNV PROBAN Software
% Example of 3 variables incorporated
%
tic
clear
clc
er=10^-50;
% Input of dependent variables - Load patterns

xinp=[15 15 15 10 10 10 5 5 5 15 15 15 10 10 10 5
5 5 15 15 15 10 10 10 5 5 5;
15 10 5 15 10 5 15 10 5 15 10 5 15 10 5 15 10 5
15 10 5 15 10 5 15 10 5;
600000 600000 600000 600000 600000 600000 600000 600000 600000
700000 700000 700000 700000 700000 700000 700000 700000 700000
800000 800000 800000 800000 800000 800000 800000 800000
800000;];
x_var=xinp';
%Input of independent variables - Response
yinp=[0.784426667 0.655335 0.527383333 0.652808333 0.522951667
0.394113333 0.522345 0.39159 0.261475 0.672365714 0.561715714
0.452042857 0.55955 0.448244286 0.337811429 0.447724286 0.335648571
0.224121429 0.58832 0.49150125 0.3955375 0.48960625 0.39221375
0.295585 0.39175875 0.2936925 0.19610625];
y_var=yinp';
342
[xs,ys]=size(x_var);
%Definition of Variables
x1_m=12;
x2_m=12;
x3_m=700000;
x4_m=er;
x1_s=2;
x2_s=2;
x3_s=80000;
x4_s=er;
% Expand x_var to X
regs=2*ys+1;
xexp=x_var;
for i=1:xs;
X(i,regs)=1;
for j=1:ys;
X(i,2*j-1)=xexp(i,j)^2;
X(i,2*j)=xexp(i,j)^1;
end;
end;
Y=y_var;
b=X\Y;
%b=inv(X'*X)*X'*Y;
% Definition of iterations
n_iter=40;
err=0.01; % Convergence criterion
% Definition of Limit State Functions

syms x1 x2 x3 x4
a1=b(1);
a2=b(2);
343
a3=b(3);
a4=b(4);
a5=b(5);
a6=b(6);
a7=er;
a8=er;
a9=b(regs);
g=1-(a1*x1^2+a2*x1+a3*x2^2+a4*x2+a5*x3^2+a6*x3+a7*x4^2+a8*x4+a9);
%g=1-(a1*x1+a2*x2+a3*x2+a4*x4+a5*x1^2+a6*x2^2+a7*x3^2+a8*x4^2+a9);
gm=matlabFunction(g);
% U-dimensional space
% Definition of x1
x1_pdf=@(x1f)normpdf(x1f,x1_m, x1_s);
x1_cdf=@(x1f)normcdf(x1f,x1_m, x1_s);
x1_pdf_v=x1_pdf(x1_m);
x1_cdf_v=x1_cdf(x1_m);
x1_s=normpdf(norminv(x1_cdf_v))/x1_pdf_v;
x1_m=x1_m-(norminv(x1_cdf_v))*x1_s;
% Definition of x2
% Definition of x3
344
% Definition of x4
% Calculation of Partial derivatives

dgx1=diff(g,x1);
dgx2=diff(g,x2);
dgx3=diff(g,x3);
dgx4=diff(g,x4);
% Express functions in a matlab complex form

g_m=matlabFunction(g);
% First Iteration
i=1;
x1_v(i)=x1_m;
x2_v(i)=x2_m;
x3_v(i)=x3_m;
x4_v(i)=x4_m;
g_v(i)=g_m(x1_v(i),x2_v(i),x3_v(i),x4_v(i));
dgx1_v(i)=dgx1_m(x1_v(i));
b_v(i)=g_v(i)/((((dgx1_m(x1_v(i))*x1_s)^2)+((dgx2_m(x2_v(i))*x2_s)^2)+
((dgx3_m(x3_v(i))*x3_s)^2)+((dgx4_m(x4_v(i))*x4_s)^2))^0.5);
345
a1_v(i)=-
(dgx1_m(x1_v(i))*x1_s)/((((dgx1_m(x1_v(i))*x1_s)^2)+((dgx2_m(x2_v(i))*
x2_s)^2)+((dgx3_m(x3_v(i))*x3_s)^2)+((dgx4_m(x4_v(i))*x4_s)^2))^0.5);
a2_v(i)=-
a3_v(i)=-
a4_v(i)=-
x1_v(i+1)=x1_m+b_v(i)*x1_s*a1_v(i);
x2_v(i+1)=x2_m+b_v(i)*x2_s*a2_v(i);
x3_v(i+1)=x3_m+b_v(i)*x3_s*a3_v(i);
x4_v(i+1)=x4_m+b_v(i)*x4_s*a4_v(i);
u1_v(i+1)=(x1_v(i+1)-x1_m)/x1_s;
u2_v(i+1)=(x2_v(i+1)-x2_m)/x2_s;
u3_v(i+1)=(x3_v(i+1)-x3_m)/x3_s;
u4_v(i+1)=(x4_v(i+1)-x4_m)/x4_s;
i=0 ;%%%% zero counter

% (2-n_iter) Iteration
for i=2:n_iter
g_v(i)=g_m(x1_v(i),x2_v(i),x3_v(i),x4_v(i));
b_nom_prod(i)=
dgx1_v(i)*x1_s*u1_v(i)+dgx2_v(i)*x2_s*u2_v(i)+dgx3_v(i)*x3_s*u3_v(i)+d
gx4_v(i)*x4_s*u4_v(i);
b_v(i)=(g_v(i)-
b_nom_prod(i))/((((dgx1_m(x1_v(i))*x1_s)^2)+((dgx2_m(x2_v(i))*x2_s)^2)
+((dgx3_m(x3_v(i))*x3_s)^2)+((dgx4_m(x4_v(i))*x4_s)^2))^0.5);
pf_v(i)=normcdf(-b_v(i));
346
a1_v(i)=-
a2_v(i)=-
a3_v(i)=-
a4_v(i)=-
x1_v(i+1)=x1_m+b_v(i)*x1_s*a1_v(i);
x2_v(i+1)=x2_m+b_v(i)*x2_s*a2_v(i);
x3_v(i+1)=x3_m+b_v(i)*x3_s*a3_v(i);
x4_v(i+1)=x4_m+b_v(i)*x4_s*a4_v(i);
u1_v(i+1)=(x1_v(i+1)-x1_m)/x1_s;
u2_v(i+1)=(x2_v(i+1)-x2_m)/x2_s;
u3_v(i+1)=(x3_v(i+1)-x3_m)/x3_s;
u4_v(i+1)=(x4_v(i+1)-x4_m)/x4_s;
e=abs(b_v(i)-b_v(i-1));
if e<err break,
end;
end;
b_form=b_v(i)
Pf_form=pf_v(i)
% Calculation of SORM Pf and beta
x1_MPP=x1_v(i);
x2_MPP=x2_v(i);
x3_MPP=x3_v(i);
x4_MPP=x4_v(i);
u1_MPP=u1_v(i);
u2_MPP=u2_v(i);
u3_MPP=u3_v(i);
u4_MPP=u4_v(i);
347
% Second order partial derivatives
d2gx1=diff(dgx1);
d2gx2=diff(dgx2);
d2gx3=diff(dgx3);
d2gx4=diff(dgx4);
d2gx1_m=matlabFunction(d2gx1);
% Transform to Matlab functions

% Check the variables in each Partial Derivative
d2gx1_v=d2gx1_m();
d2gx2_v=d2gx2_m();
d2gx3_v=d2gx3_m();
d2gx4_v=d2gx4_m();
% Check the variables in each Partial Derivative

dgx1_v=dgx1_m(x1_MPP);
% First order tensor

tens_u=((dgx1_v*(x1_s^2))^2)+((dgx2_v*(x2_s^2))^2)+((dgx3_v*(x3_s^2))^
2)+((dgx4_v*(x4_s^2))^2);
% Second order tensor (table)

tens_u2=zeros(4,4);
tens_u2(1,1)=d2gx1_v;
B_mat=(1/tens_u)*tens_u2;
348
F_mat=[(-(dgx1_v*x1_s)/tens_u) (-(dgx2_v*x2_s)/tens_u) (-
(dgx3_v*x3_s)/tens_u) (-(dgx4_v*x4_s)/tens_u);
0 1 0 0 ;
0 0 1 0 ;
0 0 0 1 ;];
F_11=-dgx1_v*x1_s;
F_22=1;
F_33=1;
F_44=1;
D1=((((F_11)^2)+((F_11)^2)));
e11=1/D1;
g1=e11*F_11;
D2=(((((F_22)^2)+((F_22)^2)))-((abs(((((F_22)^2)+(g1^2))))^2))^0.5);
e12=-(((((F_22)^2)+(g1^2)))^0.5)/D2;
e22=1/D2;
g2=e12*g1+e22*F_22;
D3=(((((F_33)^2)+((F_33)^2)))-abs((((((F_33)^2)+(g1^2))))^2)-
abs((((((F_33)^2)+(g2^2))))^2))^0.5;
e13=-(((((F_33)^2)+(g1^2)))^0.5)/D3;
e23=-(((((F_33)^2)+(g2^2)))^0.5)/D3;
e33=1/D3;
g3=e13*g1+e23*g2+e33*F_33;
D4=(((((F_44)^2)+((F_44)^2)))-abs((((((F_44)^2)+(g1^2))))^2)-
abs((((((F_44)^2)+(g2^2))))^2)-abs((((((F_44)^2)+(g3^2)))))^2)^0.5;
e14=-(((((F_44)^2)+(g1^2)))^0.5)/D4;
e24=-(((((F_44)^2)+(g2^2)))^0.5)/D4;
e34=-(((((F_44)^2)+(g3^2)))^0.5)/D4;
e44=1/D4;
g4=e14*g1+e24*g2+e34*g3+e44*F_44;
H_mat=[g1 g2 g3 g4];
H_mat_fin=[g2 0 0 0;
0 g3 0 0;
349
0 0 g4 0;
0 0 0 g1 ];
H_mat_fin_tr=transpose(H_mat_fin);
% Construction of f matrix
F_mat2=H_mat_fin*B_mat*H_mat_fin_tr;
k11=F_mat2(1,1);
k22=F_mat2(2,2);
k33=F_mat2(3,3);
k44=F_mat2(4,4);
k1=real(k11);
k2=real(k22);
k3=real(k33);
k4=real(k44);
b_form=b_v(i);
Pf_sorm=(normcdf(-b_form))*((1+k1*b_form)^-0.5)*((1+k2*b_form)^-
0.5)*((1+k3*b_form)^-0.5)
b_sorm= -(norminv(Pf_sorm))
%Pf_form
%end of program
toc
350

Athanasios I Kolios Thesis 2010

Uploaded by

Copyright:

Available Formats

Athanasios I Kolios Thesis 2010

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Athanasios I Kolios Thesis 2010

Uploaded by

Copyright:

Available Formats

CRANFIELD UNIVERSITY

ATHANASIOS IOANNIS KOLIOS

A MULTI-CONFIGURATION APPROACH TO RELIABILITY BASED

Thesis Submitted for the Degree of Doctor of Philosophy

Supervisor: Professor Feargal Brennan

Thesis Submitted for the Degree of Doctor of Philosophy

Academic Year 2010

ATHANASIOS IOANNIS KOLIOS

Thesis Submitted for the Degree of Doctor of Philosophy

Supervisor: Professor Feargal Brennan

© Cranfield University 2010. All rights reserved. No part of this

Structural Reliability treats uncertainties in structural design systematically,

Offshore structures should resist complicated and, in most cases, combined

The response of a structure under specific loading conditions can be

Consideration of appropriate limit state functions is an essential decision for

Offshore Structures, Reliability Assessment, Stochastic Response Surface

“Της δικαιοσύνης ήλιε νοητέ “Notional sun of justice

I should express my greatest gratitude to my friend and colleague Dr. Amir

Professor Dimitrios Papantonis (National Technical University of Athens) and

I own appreciation to everyone that has supported this endeavor, especially my

Figure 1: Response Surface Method Flow Chart................................................ 3

Table 1: Examples of Limit States according to DNV ....................................... 16

‫ܣ‬ Coefficient of Reyleigh Parameter

ܾ௜ First Order Regression Coefficients

‫ܥܵܫܣ‬ American Institute of Steel Construction

Current evolution in engineering practices has allowed more critical and

Cases with limited levels of randomness can be treated deterministically,

This contribution will focus on the reliability assessment of offshore structures,

The methodology that is proposed allows reliability assessment of structural

Identification of variables Chapter 4

Construction of Design Matrix Par. 3.3.1

Execution of FEA Simulations

Formulation of Limit States Values Chapter 5&6

Data Regression Analysis Par. 3.4

Reliability Index Calculation Par. 3.2

Figure 1: Response Surface Method Flow Chart

The Second Chapter presents a review of reliability assessment of steel and

In the Fourth Chapter, the environmental loading and capacity modelling of

1.2 Structural Reliability

Structural reliability works on the prediction of the probability of exceedance of

Within different application fields, different methods for reliability assessment

Structural Reliability Analysis is related to the estimation of the probability of

A classification between probability concepts, distinguishes frequentistic to

Statistical modelling of uncertainties is analytically studied in [13], while in [14] a

 Physical (intrinsic or inherent) uncertainty describes the natural

Structural Reliability Methods are classified according to their Level, Moment

 Level, refers to the extent of information that the reliability problem

Offshore structural reliability analysis becomes of particular importance recently

Structural reliability analysis can provide significant benefit to the potential

During the last decades, momentous developments have occurred in the

I. Development of a methodology for the reliability assessment of a

1.4 Design of Offshore Structures

1.4.1 Permissible stresses

The disadvantage of the method is that although R and S are calculated in a

1.4.3 Partial Safety factor

‫ܧ‬ௗ (‫ܨ‬ௗ , ݂ௗ , ܽௗ , ߠௗ ) < ܴௗ (‫ܨ‬ௗ , ݂ௗ , ܽௗ , ߠௗ ) (1-3)

1.4.4 Probabilistic Methods

In addition to the Partial Safety Factor method, concepts of probabilistic

1.4.5 Comments on design methods