(Lecture Notes in Engineering 57) M. G. Donley, P. D. Spanos (Auth.) - Dynamic Analysis of Non-Linear Structures by The Method of Statistical Quadratization-Springer-Verlag Berlin Heidelberg (1990)

Lecture Notes in Engineering
The Springer-Verlag Lectu re Notes provide rap id (approximately six months), refereed publication
of topical items, longer than ordinary journal articles but shorter and less formal than most
monographs and textbooks. They are publ ished in an attractive yet economical format; authors or
editors p rovide manuscripts typed to specifications, ready for photo-reproduction.
The Editorial Board
Managing Editors
C. A Brebbia S. AOrszag
Wessex Institute of 1echnology Applied and Computational Mathematics
Ashurst Lodge, Ashurst 218 Fine Hall
Southampton S04 2AA (UK) Pri nceton, NJ 08544 (USA)
Consulting Editors Materials Science and Computer Simulation:

S, Yip
Chemical Engineering: Dept of Nuclear Engg., MIT
J. H. Seinfeld Cambridge, MA 02139 (USA)
Dept. of Chemical Engg., Spaulding Bldg.
Calif. Inst. of Technology Mechanics of Materials:
Pasadena, CA 91125 (USA) FA Leckie
Dept. of Mechanical Engineering
Dynamics and Vibrations: Univ, of Californ ia
P'Spanos Santa Barbara,
Departm ent of Mechanical and CA 93106 (USA)
Civil Engineering, RiceUniversi ty A. R. S. Ponter
P. O. Box 1892 Dept. of Engineering, The University
Houston, Texas 77251 (USA) Leicester LEI 7RH (UK)
Earthquake Engineering: Fluid Mechanics:

AS. Cakmak K.·P. Holz
Dept. of C ivil Engi neeri ng, Pri nceto n University Inst fU r Stromungsmechanik,
Princeton, NJ 08544 (USA) Universitat Hannover, Call instr. 32
D·3000 Hannover 1 (FRG)
Electrical Engineering:
P. Silvester Nonlinear Mechanics:
Dept. of Electrical Engg., McGill University K.-J. Bathe
3480 University Street Dept. of Mechanical Engg., MIT
Montreal, PO H3A 2A 7 (Canada) Cambridge, MA 02139 (USA)
Geotechnical Engineering and Geomechanics: Structural Engineering:

C.S. Desai J. Connor
College of Engineering Dept. 01 Civil Engineering, MIT
Dept. of Civil Engg. and Engg. Mechanics Cambridge, MA 02139 (USA)
The University of Arizona W, Wu nderlich
Tucson, AZ 85721 (USA) Inst. fU r Konstruktiv en Ingenieurbau
Ruhr-Universitat Bochum
Hydrology: U n vi ersitatsstr. 150,
G. Pinder 0 -4639 Bochum·Ouerenburg (FRG)
School of Engineering, Dept. of Civil Engg.
Princeton University Structural Engineering, Fluids and
Princeton, NJ 08544 (USA) Thermodynamics:
J. Argyris
Laser Fusion - Plasma: Inst fUr Statik und Oynam ik der
R. McCrory Luft- und Raumfahrtkonstruktion
Lab. for Laser Energetics, University of Rochester Pfaffenwaldri ng 27
Rochester, NY 14627 (USA) 0 -7000 Stuttgart 80 (FRG)
Lecture Notes in
Engineering
Edited by C. A. Brebbia and S. A. Orszag
57
M. G. Donley, P. D. Spanos
Dynamic Analysis
of Non-Linear Structures
by the Method of Statistical
Quadratization
Springer-Verlag
Berlin Heidelberg New York London
~~--' Paris Tokyo Hong Kong Barcelona
Series Editors
C. A. Brebbia . S. A. Orszag
Consulting Editors
J. Argyris . K -J. Bathe . A. S. Cakmak . J. Connor· R. McCrory
C. S. Desai· K.-P. Holz . F. A. Leckie· G. Pinder· A. R. S. Pont
J. H. Seinfeld . P. Silvester· P. Spanos· W. Wunderlich· S. Yip
Authors
M.G. Donley
Structural Dynamics Research Corporation
2000 Eastman Drive
Milford, OH 45150-2789
USA
P.D. Spanos
Rice University
Brown School of Engineering
P. O. Box 1892
Houston, TX 77251
USA
ISBN-13: 978-3-540-52743-5 e-ISBN-13: 978-3-642-46715-8

001: 10.1007/978-3-642-46715-8
This work is subject'to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation,
broadcasting, reproduction on microfilms or in other ways, and storage in data banks. Duplication
of this publication or parts thereof is only permitted under the provisions of the German Copyright
Law of September 9, 1965, in its version of June 24, 1985, and a copyright fee must always be paid.
Violations fall under the prosecution act of the German Copyright Law.
© Springer-Verlag Berlin, Heidelberg 1990
The use of registered names, trademarks, etc. in this publication does not imply, even in the
absence of a specific statement, that such names are exempt from the relevant protective laws
and regulations and therefore free for general use.
2161/3020-543210 Printed on acid-free paper.

ABSTRACT
Stochastic linearization is perhaps the most frequently used analytical method for
analyzing the response of many nonlinear systems,. as it provides reasonable estimates of
the mean square response. However, the method is not, in general, well suited for
estimating the power spectra of stationary responses of randomly excited nonlinear
systems. In addition, for a Gaussian excitation, the linearized solution leads to a
Gaussian probability distribution, whereas the true response is non-Gaussian. In this
study, a higher order method termed equivalent stochastic "quadratization" is proposed to
circumvent these shortcomings. The nonlinearity is replaced by a polynomial expansion
up to a quadratic order. In this manner the Volterra series method can be used to
approximate the response of the resulting nonlinear system. The system excitation is
assumed to be Gaussian. However, the response is described by a non-Gaussian
probability distribution. The method is developed for analyzing the stationary response
of single and multi-degree-of-freedom systems; pertinent instructional examples are
included. Further, a useful practical application of the proposed method is pursued for
analyzing the stochastic response of compliant offshore platforms due to nonlinear drag
forces. These are structures used to exploit oil resources in great water depths. The
compliant nature of these platforms introduces nonlinear behavior which can not be
neglected as in conventional offshore platforms. The method is applied for analyzing a
specific three-degree-of-freedom model of a Tension Leg Platform (lLP) subject to wave
and current forces. In addition to nonlinear drag forces, nonlinear potential forces
significantly affect the lLP response. These forces are derived in the form of second
order Volterra series. A stochastic response analysis of the TLP system due to combined
nonlinear drag and nonlinear potential forces is performed to evaluate the relative
significance of these forces.
The analytical results produced by the equivalent quadratization method for the
instructional and practical problems considered, are found in good agreement with
pertinent numerical data generated by Monte Carlo studies.
Clearly, the concept of quadratic, or even higher power, polynomial approximation
of arbitrary nonlinearities and subsequent application of the Volterra series expansion for
determining the random response of the derived equivalent nonlinear system, appears to
be quite promising and meritorious. However, it is noted that the present study is strictly
preliminary in nature, and reporting its findings in the present format conforms with the
objective of the Lecture Notes in Engineering Series. Additional research is required to
address versatility, reliability, and efficiency issues.
Acknowledgement
Funding through a Presential Young Investigator Award from the National Science
Foundation and a consortium of industrial finus is gratefully ackowledged; the partial
support by grants from the National Center For Earthquake Engineering Research at the
State University of New York at Buffalo is appreciated.
TABLE OF CONTENTS
CHAPTER 1: INTRODUCTION ...................................................... 1

1.1 Introduction ................................................... 1
1.2 Aim of Study................................................. 2
1.3 TLP Model .................................................... 3
1.4 Environmental Loads ...... ...................... ............ 5
1.4.1 Methods to Compute Viscous Forces .............. 7
1.4.2 Methods to Compute Potential Forces ............. 8
1.5 Literature Review of TLP Analyses ....................... 11
1.6 Scope of Study............................ ................... 12
CHAPTER 2: EQUIVALENT STOCHASTIC QUADRATIZATION FOR

SINGLE-DEGREE-OF-FREEDOM SySTEMS.................. 14
2.1 Introduction ................................................... 14
2.2 Analytical Method Formulation ............................ 15
2.3 Derivation of Linear and Quadratic Transfer Functions. 21
2.4 Response Probability Distribution ......................... 23
2.5 Response Spectral Density .................................. 26
2.6 Solution Procedure .......................................... 26
2.7 Example of Application .............................. ....... 27
2.8 Summary and Conclusions .................................. 37

MULTI-DEGREE-OF-FREEDOM SySTEMS .................... 41
3.1 Introduction ............................................... ~ ... 41
3.7 Reduced Solution Analytical Method ...................... 49
3.8 Example of Application ..................................... 50
CHAPTER 4: POTENTIAL WAVE FORCES ON A MOORED
VERTICAL CYLINDER ............................................ 58
4.1 Introduction ................................................... 58
4.2 Volterra Series Force Description .......................... 58
4.3 Near-Field Approach for Deriving Potential Forces ..... 63
4.3.1 Fluid Flow Boundary Value Problem ............. 63
4.3.2 Perturbation Expansion .............................. 66
VI
4.4 Linear Velocity Potential.... ....... .... ........... ..... ..... 70

4.5 Added Mass Force ........................................... 73
4.6 Linear Force Transfer Functions ........................... 73
4.6.1 Wave Diffraction Force .............................. 73
4.6.2 Wave Diffraction Moment ........................... 74
4.6.3 Hydrodynamic Buoyancy Force ................... 74
4.6.4 Comparison to Morison's Equation ................ 75
4.7 Quadratic Force Transfer Functions ....................... 75
4.7.1 Wave Elevation Drift Force ......................... 75
4.7.2 Wave Elevation Drift Moment ...................... 77
4.7.3 Velocity Head Drift Force ........................... 78
4.7.4 Velocity Head Drift Moment ........................ 79
4.7.5 Body Motion Drift Forces and Moment ........... 80
4.7.6 Numerical Examples for Fixed Vertical Cylinder 81
4.8 Transfer Functions for Tension Leg Platform ............ 85
4.8.1 Modification of Cylinder Transfer Functions ..... 85
4.8.2 Numerical Example for Tension Leg Platform '" 86

TENSION LEG PLATFORM RESPONSE TO
VISCOUS DRIFT FORCES .................. ... ............... .... 95
5.1 Introduction ................................................... 95
5.2 Formulation of TLP Model ................................. 95
5.7 Axial Tendon Force .......................................... 105
5.9 Numerical Example .......................................... 106
5.10 Summary and Conclusions ................................. 114
CHAPTER 6: STOCHASTIC RESPONSE OF A TENSION LEG PLATFORM

TO VISCOUS AND POTENTIAL DRIFT FORCES ............ 115
6.1 Introduction ................................................... 115
6.3 Numerical Results ........................................... 118
6.3.1 Response to Quadratic Drag Force ................. 118
6.3.2 Response to Quadratic
Wave ElevationNelocity Head Force .............. 122
6.3.3 Response to Quadratic Body Motion Force ....... 127
6.3.4 Response to Combined Viscous and
Potential Quadratic Forces ...... ........... ......... 131
6.3.5 Evaluation of Newman's Approximation .......... 135
6.3.6 High Frequency Axial Tendon Force .............. 137
6.4 Summary and Conclusions ................................. 145
VII
CHAPTER 7: SUMMARY AND CONCLUSIONS................................. 147
APPENDIX A: GRAM-CHARLIER COEFFICIENTS .............................. 151

A.1 Introduction ................................................... 151
A.2 Gram-Charlier Coefficients ................................. 151
APPENDIXB: EVALUATION OF EXPECTATIONS .............................. 153

B.1 Introduction ................................................... 153
B.2 Expectations Involving Quadratic Nonlinearity .......... 153
B.3 High Order Central Moments ............................... 156
APPENDIX C: PIERSON-MOSKOWITZ WAVE SPECTRUM ................... 158
APPENDIX D: SIMULATION METHODS ........................................... 161

D.1 Introduction ................................................... 161
D.2 Linear Wave Simulation ..................................... 161
D.3 Linear Wave Force Simulation ............................. 161
DA Drag Force Simulation ...................................... 162
D.5 Quadratic Wave Force Simulation ......................... 163
REFERENCES: ........................................................................... 167

CHAPTER 1
INTRODUCTION
1.1 Introduction
As offshore oil production moves into deeper water, compliant structural systems
are becoming increasingly important. Examples of this type of structure are tension leg
platfonns (TLP's), guyed tower platfonns, compliant tower platfonns, and floating
production systems. The common feature of these systems, which distinguishes them
from conventional jacket platfonns, is that dynamic amplification is minimized by
designing the surge and sway natural frequencies to be lower than the predominant
frequencies of the wave spectrum. Conventional jacket platfonns, on the other hand, are
designed to have high stiffness so that the natural frequencies are higher than the wave
frequencies. At deeper water depths, however, it becomes uneconomical to build a
platfonn with high enough stiffness. Thus, the switch is made to the other side of the
wave spectrum. The low natural frequency of a compliant platfonn is achieved by
designing systems which inherently have low stiffness. Consequently, the maximum
horizontal excursions of these systems can be quite large.
The low natural frequency characteristic of compliant systems creates new analytical
challenges for engineers. This is because geometric stiffness and hydrodynamic force
nonlinearities can cause significant resonance responses in the surge and sway modes, even
though the natural frequencies of these modes are outside the wave spectrum frequencies.
High frequency resonance responses in other modes, such as the pitch mode of a TLP, are
also possible.
One source of nonlinearity is the hydrodynamic drag force, which is due to flow
separation around a submerged member. This force is frequently modeled mathematically
by empirical equations such as the nonlinear Morison equation. For perfonning stochastic
analyses, linearization methods such as described by Malhotra and Penzien(l970) are often
utilized. However, responses at frequencies outside the wave spectrum frequencies are not
predicted by linearization. Therefore, some response statistics may be significantly
unconservative. ill this study, a higher order method tenned equivalent stochastic
"quadratization" is proposed to circumvent this shortcoming of the equivalent stochastic
linearization method.
2
Another source of nonlinearity is in the wave induced potential forces. These

forces result from potential pressure gradients due to waves. Most compliant platform
analyses in the literature model these forces by numerical methods such as fmite element
methods or sink-source methods, or by analytical methods based on slender member
theory. The numerical methods are good for modeling systems with complex geometries,
but are computationally expensive and more suitable for fmal design analyses. The interest
of this study focuses on analytical methods since they are efficient and provide more insight
into the fundamental behavior of compliant systems, although some accuracy may be
sacrificed. Methods based on slender member theory, however, are inadequate because
they do not consider wave scattering effects. For vertical cylinders, analytical methods that
include wave scattering effects have been published in the literature, but have only been
applied to compliant platform analyses in a limited manner. In this study, a more complete
accounting of the potential forces is made.
1.2 Aim of Study

The purpose of this study is twofold. First, it is to verify the usefulness of the
equivalent stochastic quadratization method as a tool for obtaining the response statistics of
a compliant offshore system subject to nonlinear drag forces. A TLP system is used to
develop and exemplify the proposed method. The verification procedure is presented in a
systematic manner. The method is first developed as a general tool for analyzing nonlinear
single-degree-of-freedom(sdof) systems subject to simple force excitations. The
applicability of the method is then extended to general nonlinear multi-degree-of-
freedom(mdof) systems, before fmally applying it to a TLP system with three degrees of
freedom. The second aim of the study is to analyze the response of the TLP system to
combined nonlinear drag and nonlinear potential forces to evaluate the relative significance
of these forces. Some of the more recent methods for modeling nonlinear potential wave
forces are derived in a form which is more suitable for stochastic analyses of compliant
systems. The estimation of the nonlinear low frequency surge response of a TLP system is
of particular interest to this study. The nonlinear high frequency pitch response and its
effect on axial tendon tensions is also to be investigated. In addition, the non-gaussian
nature of the responses is considered to be an integral part of the analysis. This author is
aware of no other analytical study which is more comprehensive in its modeling of the
nonlinear wave forces and consequent responses of a TLP.
3
The remainder of this chapter is a discussion of modeling TLP systems and the
environmental loads which act on a TLP's structural members. This is followed by a
literature review of TLP response studies. The section on environmental loads is a
somewhat involved review of hydrodynamic wave force theories since the response
analyses of TLP's can not be understood without consideration of the hydrodynamic
forces. The last section of this chapter gives a general scope of this study.
1.3 TLP Model

A TLP has a floating hull which is tied in place by tensioned vertical tendons. The
typical TLP hull shape consists of four cylindrical column members arranged in a
rectangular grid and connected at the base by cylindrical pontoon members. A diagram of
the idealized TLP that is used in this study is shown in Figure 1.1.
The stochastic response of the hull due to wave and current induced forces is the
primary interest of this study. Jefferys and Patel(1981) have shown that the inertia and
wave forces acting on the tendons have a negligible effect on the motion of the hull.
Therefore, the tendons are treated as massless springs which in conjunction with the hull
buoyancy provide the restoring forces on the hull. The geometric nonlinearities inherent in
the restoring forces are neglected since they are less important than the wave force
nonlinearities. This is common assumption used in the literature.
Typically, the hull is considered to be a rigid body with six degrees of freedom.
However, in the presence of a unidirectional flow field, which is parallel to the surge axis
of the TLP, the hull responds in only three degrees of freedom. That is, surge or
horizontal translation, heave or vertical translation, and pitch or rotation. This simple flow
condition can be used to highlight the salient features of TLP responses. Therefore, it is
used in the present analysis for simplicity and clarity of the results. It is noted that the pitch
and heave motion directly influence the force in the tendons, while the surge motion has
only an indirect influence through coupling with the pitch motion. The surge motion is
most important in the analysis and design of the riser system and foundation, which are not
modeled in the present study.
The surge natural period of typical TLP's is on the order of 70 to 120 seconds. The
pitch and heave natural periods are much less and are in the range of 2 to 4 seconds. These
periods are away from the dominant wave periods which are 4 to 6 seconds in normal sea
states and 12 to 20 seconds in severe sea states. It is noted that the surge and pitch degrees
4
x If!
iii
c:
=eo
-
o
()
c:
~
~
! a..
....J
I-
"C
Q)
.~
ca
Q)
-
"C
o
E
ca
~
Q)
ca
(5
5
of freedom are coupled through the added mass matrix and the hydrostatic stiffness matrix.
The natural frequencies, however, are not very different than if the off diagonal terms in the
mass and stiffness matrices are neglected. This indicates that the coupling is small. The
heave degree of freedom is not coupled with the other degrees of freedom.
It is assumed that the total fluid force acting on the TLP hull can be obtained by
summing the fluid forces acting on individual hull members as though other members are
not present. In actuality, the presence of nearby members alters the flow field and, hence,
the force acting on a member. If the members are spaced on the order of five diameters
away from each other, however, the effect is small. This is a reasonable assumption for
most TLP hulls.
1.4 Environmental Loads

The environmental loads acting on a TLP are due to waves, current, and wind.
Only the response due to waves and current is considered in this study. Despite
considering the TLP system to be linear, the response is still nonlinear because the wave
and current induced forces are nonlinear. The linearity of the force depends on its relation
to the wave elevation from linear wave theory. A linear force is linearly related to the wave
elevation, a quadratic force is quadratically related to the wave elevation, and so on. In
offshore systems, it is convenient to express the force and the resulting response as a
Volterra series in which the wave elevation is the input function such as described by
Yamanouchi(1974) and Vassilopoulos(1967). The series is usually truncated after second
order. The linear force is called the first order force while the quadratic force is called the
second order force or drift force. In fue frequency domain, linear and quadratic transfer
functions are needed to describe these forces.
It is well known that the surge response of a TLP subjected to wave and current
loads consists of a wave frequency response, a mean response, and a low frequency or
slowly varying response at the TLP's surge natural frequency. High frequency responses
at the pitch natural frequency can also occur, although this has received less attention in the
literature. The wave frequency response is due predominantly to linear potential forces
acting on the hull and to a much lesser extent to viscous forces. The mean, low frequency,
and high frequency responses are due to higher order wave forces, in particular, quadratic
wave forces. These forces are due to both potential and viscous effects. In general, both
effects contribute significantly to the total drift force. Further, the viscous forces induce a
6
hydrodynamic damping effect on the TLP. This effect substantially reduces the magnitude
of the low frequency response which is a resonance type response.
In the following subsections the analytical methods used to model the fluid induced
forces on cylinders are reviewed. The focus is on methods of modeling the wave forces
acting on vertical surface piercing cylinders, because the predominant structural members
of the TLP hull shape are the cylindrical columns.
There have been many analytical investigations of wave forces acting on vertical
cylinders. However, most have shortcomings for use in TLP analyses. A drawback of
many investigations is that the cylinder is considered fixed. A TLP column, conversely,
experiences large horizontal motion. This motion is the source of quadratic order forces
due to both potential and viscous effects. It is also the source of the viscous hydrodynamic
damping. Secondly, the quadratic forces are often derived only for a regular wave load
case. A general derivation involves the wave load due to two harmonic waves at different
frequencies. The resulting quadratic force is composed of two harmonics, whose
frequencies are the difference and sum of the frequencies of the wave frequencies.
Langley(1987a) gives a simple example which shows this behavior. The regular wave load
case can be viewed as the special case in which the two harmonic waves have the same
frequency. The difference harmonic becomes a mean force and the sum harmonic force has
a frequency of double the wave frequency. Most of the literature on drift forces focuses on
the mean force due to a regular wave. This amounts to solving the quadratic transfer
functions for pairs of equal frequencies. In an irregular sea state composed of many
harmonics, this limited information only gives the mean quadratic force. The full quadratic
transfer function is required to describe the low frequency force, which is due to
combinations of difference frequencies, and the high frequency force, which is due to
combinations of sum frequencies. Newman(1974) proposed an approximation in which
the limited information from the mean force could be used to estimate the low frequency
quadratic forces. This approximation is frequently used in practice although its reliability
has not been completely verified and the results may be unconservative. In addition, the
approximation can not account for any high frequency quadratic forces despite the fact that
recent investigations such as Nielsen and Herfjord(1985) have concluded that they may
influence the fatigue life of offshore structures.
7
The following review of the methods to model hydrodynamic forces is separated

into viscous and potential forces. Based on these methods an approach for modeling the
wave forces on a TLP is proposed.
1.4.1 Methods to Compute Viscous Forces

The viscous or drag force on a cylinder is usually modeled by the Morison equation
drag force term. On a unit length of a submerged member, this force is proportional to the
product of the water velocity and the absolute value of the water velocity. Typically, the
water velocity at the cylinder's centerline, assuming that the cylinder is not present, is used.
If the cylinder moves, a generalized form of Morison's equation, based on the relative
water velocity, is more appropriate to use. In the presence of waves and current, the
relative water velocity is the vectorial summation of the current velocity and the wave field
water velocity minus the cylinder velocity. Including the latter term causes a hydrodynamic
damping effect, which is important to model if resonance responses are expected. This
generalized form of the Morison drag force is adopted in this study. It is noted, however,
that alternative viscous force models, such as Moe and Verley(l980), have also been
proposed.
The analytical form of the Morison equation drag force is nonlinear. In addition, it
is not readily suited to a Volterra series expansion because it does not involve a polynomial
nonlinearity. Nevertheless, it is possible to approximate it by a Volterra series. For this
one may replace the drag nonlinearity by a polynomial expansion of the water velocity.
The unknown polynomial coefficients are found by minimizing the error. In stochastic
analyses, the error is minimized in a mean square sense which requires knowledge of the
probability distribution of the water velocity. If relative velocity effects are considered, the
probability distribution must be obtained iteratively since the statistics of the motion are not
known beforehand. For example, Spanos and Chen(l98 I) use the equivalent linearization
method to analyze the response of a single degree-of-freedom oscillator to drag induced
forces. Although it is not formulated in this manner, it can be shown that the linearization
method amounts to replacing the drag force by a first order Volterra series. Comparison
with simulation shows good agreement for mean responses, wave frequency responses,
and hydrodynamic damping. However, the low and high frequency response due to
quadratic effects can not be obtained. Others have extended the concept of linearization to
higher orders. Gudmestad and Conner(1983) approximate the drag nonlinearity by a
8
polynomial expansion. Quadratic order forces are obtained by including polynomials up to

second order. Cylinder motion is not considered, and the probability distribution of the
water velocity is assumed to be gaussian. Madsen and Jensen(1989) present a similar
analysis, but assume that the water velocity is non-gaussian. Hu and Dixit(1988) use a
Hermite polynomial expansion. Relative velocity effects are not considered and a gaussian
water velocity is used. The response is found to have a non-gaussian distribution.
Olagnon et al.(1988) use a polynomial expansion for the drag nonlinearity to analyze the
response of a free-standing conductor pipe. The wave field water velocity is assumed to be
gaussian and relative velocity effects are considered. The response is assumed to be
gaussian. However, Hu and Dixit(1988) and others such as Langley(1987b),
Naess(1986), and Stansberg(1983) have documented that the response to quadratic order
forces is non-gaussian.
Other methods which do not involve series expansions to approximate the drag
force have also been proposed. Ferretti and Berta(1980) compute the mean drag force due
to a regular wave and current. Newman's approximation, originally developed for
potential drift forces, is proposed to approximate the low frequency force. However, the
wave frequency and high frequency responses, as well as the hydrodynamic damping, are
not modeled by this method.
In this study, a method termed equivalent stochastic "quadratization" is proposed to
analyze the response of a TLP to viscous forces. The generalized drag force is replaced by
a polynomial expansion of the relative velocity up to quadratic order. The wave field water
velocity is assumed to be gaussian. The response, however, is described by a non-
gaussian probability distribution. An iterative procedure is required.
Before applying the equivalent stochastic quadratization method to the TLP
problem, it is developed as a general method to analyze nonlinear systems. Both single and
multi-degree of freedom systems are considered. The method can be viewed as an
extension of the equivalent stochastic linearization method described by Spanos(1981a).
1.4.2 Methods to Compute Potential Forces

Linear potential wave force theory is well established and correlates well with
experimental data. For cylinders with diameters, Dc, small enough (OJ).. < .2) compared to
the wavelength, ).., the use of the Morison equation to model the linear potential force is
justified. This approach is valid for typical TLP columns in severe sea states because the
9
wave energy is concentrated in the low frequency waves which correspond to longer
wavelengths. For sea states that are not severe, wave scattering effects become important,
and the Morison equation is no longer valid. MacCamy and Fuchs(1954) obtained the
linear potential force acting on a cylinder including the wave scattering, or diffraction,
effects. The force is computed by integrating the pressure obtained from the linear
diffracted velocity potential over the mean surface area of the cylinder. This force
converges to the Morison equation inertia force for cylinders with small diameters. The
diffraction approach is used in the present study for two reasons. First, it is applicable to
all sea states and column diameters. Second, the linear diffracted velocity potential is the
source of some quadratic order forces.
Analytical expressions for the quadratic potential forces on a cylinder have been
obtained by several methods. One which is often adopted is based on the near field
approach of Pinkster(1979) for computing forces on floating bodies. This involves
expanding the fluid pressure in a perturbation power series and integrating over the wetted
surface area of the body. This yields a potential force that is also in the form of a
perturbation series which is essentially the same as a Volterra series. The first order or
linear term is the same as the linear force from diffraction theory. The second order terms
are the quadratic forces.
The quadratic order forces that are of most concern for TLP analyses are the wave
elevation drift force, the velocity head drift force, the body motion drift force, and the
quadratic potential drift force. The wave elevation force is the result of integrating the
linear pressure over the fluctuating submerged surface area. The other three forces are
obtained by integrating the quadratic pressure over the mean submerged surface area.
These pressure contributions follow from the second order part of the Bernoulli equation.
The velocity head pressure comes from the square of the fluid velocity associated with the
linear diffracted potential. The body motion pressure is the second order term in the Taylor
expansion of the linear pressure about the mean position of a member and takes into
consideration the displaced position of a member. The quadratic potential pressure is
associated with the quadratic diffracted velocity potential.
Various authors have applied Pinkster's general second order force expressions to
derive some of the quadratic forces on a fixed cylinder. Chakrabarti(l984)(1975),
Chakrabarti and Cotter(1983), Rahman(1984), Rahman and Heaps(1983), and Rahman
and Chakravartty(198I) derived the mean force component for the wave elevation and
10
velocity head drift forces on a fixed cylinder. Herfjord and Nielsen(1986) derived the
wave elevation and velocity head forces on a fixed cylinder due to two harmonic waves.
Their method contains the information to compute the full quadratic transfer function,
although it is not derived as such.
Other analyses such as Olagnon et al.(l988) and Thiagarajan and Baddour(1989)
use Morison's equation to compute the wave elevation drift force. This approach is based
on a slender member assumption, which neglects diffraction effects and the wave height
variation around the cylinder. Other investigations, such as Isaacson(1979), have used a
Morison's equation approach to compute the velocity head drift force. Higher order wave
theories are needed to give a nonzero force. On the other hand, if diffraction effects are
considered, linear theory yields a nonzero force.
The quadratic potential force on a cylinder due to body motion has been addressed
in a few studies. There is some disagreement as to how to apply the Taylor series
expansion. Madsen(1986) expands the pressure about a local coordinate system attached to
the moving cylinder. Pinkster(1979) presented expressions for this force for floating
bodies, but the expansion is somewhat ambiguous and depends on how a gradient is
intetpreted. Spanos and Agarwal(1984) derive a body motion force term intuitively from
Morison's equation. Their approach involves an expansion in a global coordinate system.
Lundgren et al.(1982) discuss the body motion effect in terms of Mathieu instabilities.
Their approach also involves an expansion in a global coordinate system.
Studies by Rahman(1984), Rahman and Heaps(1983), Taylor and Hung(1987),
Kokkinowrachos and Thanos(1988), and Molin(1979) have derived the quadratic forces
due to the quadratic diffracted velocity potential, but doubt remains as to their validity. In
addition, these derivations have been found to be quite complicated and sometime suffer
from computational difficulties. For these reasons, this quadratic force will not be
addressed in this study.
In the present study, the linear and quadratic potential forces are derived for a
moored cylinder using linear diffraction theory. The Volterra series formalism is used to
mathematically describe these forces. The full linear and quadratic transfer functions are
derived. A perturbation approach similar to Pinkster(1979), but in which the body motion
effect is presented more clearly, is used. The resulting quadratic body motion force is
found to agree with Spanos and Agarwal(1984). The wave elevation and velocity head
11
forces agree with Herfjord and Nielsen(1986) if the cylinder is considered fixed, but is
slightly modified if it is not fixed.
I.S Literature Review of TLP Analyses

Many TLP response studies have been conducted utilizing the various approaches
for modeling the wave and current induced forces. The investigations are categorized either
as time domain or frequency domain methods. Time domain methods involve numerical
integration of the equations of motion and are well suited for accounting for many of the
nonlinearities associated with TLP systems. By simulating random wave forces over many
time steps, reliable response statistics can be obtained. However, simulation is an
inefficient and costly procedure and is used in the present study only for evaluating the
reliability of the proposed frequency domain analytical methods. For completeness, the
following list of references which use time domain methods for TLP response analyses is
provided: Natvig and Pendered(1977), Albrecht et al.(1978), Beynet et a1.(1978), Denise
and Heaf(1979), Angelides et al.(1982), Kitami et al.(1982), Salvesen et al.(1982),
Finnigan et al.(1984), Spanos and Agarwal(1984), Datta and Jain(1988), Gidwani(1988).
Frequency domain methods are quite efficient for performing stochastic analyses.
However, they are more difficult than time domain methods to apply for nonlinear
analyses. A review of the frequency domain methods which have been used to analyze
TLP's follows. It is noted that not all of these investigations involve a stochastic analysis.
The early TLP investigations relied on linear frequency domain analyses and
linearization of the nonlinearities. Although mean second order forces could be obtained by
these methods, no low or high frequency responses are accounted for. Kitami et al.(1982),
Paulling and Horton(1970), and Natvig and Pendered( 1977) all assume linear potential
forces and linearize the drag force. Albrecht et al.(1978) also linearize the nonlinear
geometric stiffness. Jefferys and Patel(1981) and Lyons et al. (1983) use linear methods to
investigate the effect of tendon dynamics. Spanos and Agarwal(1984) linearize the
nonlinearity due to the body motion drift force.
Other investigations recognized the importance of the second order response and
sought solutions for the mean, low frequency, and high frequency response. Initial studies
involved the TLP response to a single harmonic wave, which consists of a mean response
and a double wave frequency harmonic response. Many use the mean force with
Newman's approximation to estimate the low frequency response and ignore the high
12
frequency response. Kirlc and Etok(1979) model the linear force with the Morison
equation inertia force and the low frequency response by the Newman approximation. No
explicit method is given for computing the mean response. Mercier et al.(1982) propose a
similar method. Pijfers and Brink(1977) and Bums(1983) analytically compute the mean
viscous force on the TLP hull due to a regular wave and current and use the Newman
approximation for approximating the low frequency viscous force as proposed by Ferretti
and Berta(1980). Yoshida et al.(1981) use Morison's equation to compute the wave
elevation and body motion drift force due to a regular wave; numerical integration is used
and the tendon response at the double frequency is investigated. Denise and Heaf(1979)
use a fmite element method to obtain the linear potential force transfer function and the
mean potential force. Botelho et al.(1984) and Salvesen(1982) use a finite element method
for the potential force and time domain integration for the mean drag force. The mean drag
and potential force are combined and used to approximate the low frequency response.
Faltinsen et al.(1982) use sink-source methods to obtain the linear potential and mean
potential force. Kobayashi(1986), as well, uses the sink-source method for the potential
force and Burns(1983) approach for the viscous force.
Recent investigations have utilized the full quadratic transfer function for the
potential force, but, have linearized the drag force. Tan and deBoom(1981) use the sink-
source method for the potential forces and investigate the low frequency surge response.
deBoorn et al.(1983) use a similar method, but investigate the high frequency axial force in
the tendons. Marthinsen(1989) does a similar analysis. Petrauskas and Liu(1987) use the
method proposed by Herfjord and Nielsen(1986) to analytically model the second order
wave elevation and velocity head drift force. The high frequency force in the tendons is
investigated.
1.6 Scope of Study

In Chapter 1, introductory remarks are made regarding the methods of analyses and
features of dynamic behavior of TLP's.
In Chapter 2, a general development of the equivalent stochastic quadratization
procedure is presented for analyzing the stationary, non-gaussian response of a nonlinear
single-degree-of-freedom oscillator subject to gaussian force excitation. Numerical results
are obtained for examples of damping and stiffness nonlinearities and are compared to
13
linearization and Monte Carlo simulation results. The factors crucial to when quadratic
order effects are important are investigated.
In Chapter 3, the equivalent stochastic quadratization procedure is extended to
nonlinear multi-degree-of-freedom systems. Numerical results are obtained for a simple
two-degree-of-freedom system.
In Chapter 4, the linear and quadratic potential force transfer functions for a moored
vertical surface piercing cylinder are derived. The applicability of these analytical functions
to TLP response analyses is discussed.
In Chapter 5, the equivalent stochastic quadratization procedure is presented for
analyzing the response of a idealized three-degree-of-freedom TLP subject to drag forces
and linear potential forces. The current's effect on the low frequency surge response is
investigated.
In Chapter 6, the equivalent stochastic quadratization procedure is used to analyze
the response of a TLP to first and second order forces due to both viscous and potential
effects. The relative magnitude of the various second order responses is discussed. The
validity of the Newman approximation for low frequency responses is investigated. The
high frequency axial force in the tendons is also addressed.
Finally in Chapter 7, the results are summarized and areas of future research are
recommended.
CHAPTER 2
EQUIVALENT STOCHASTIC QUADRATIZATION
FOR SINGLE-DEGREE-OF -FREEDOM SYSTEMS
2.1 Introduction
The equivalent stochastic linearization method has proven to be a convenient and
efficient analytical tool for computing the response statistics of nonlinear systems. This
method was introduced by Krylov and Bogoliubov(1947) for nonlinear systems subject to
detenninistic excitation. It was first applied to nonlinear stationary systems with random
excitations by Booton(1954) and later Caughey(1963). Later investigators generalized the
method to multi-degree-of-freedom systems, nonstationary responses, and non-gaussian
responses. Pertinent information can be found in Iwan and Yang(1972), Atalik and
Utku(1976), Spanos(1980), Spanos(1981a), Beaman and Hedrick(1981). For a survey on
linearization methods, see Spanos(1981b) and Roberts and Spanos(1989).
In some cases, the accuracy of equivalent stochastic linearization is not adequate.
Specifically, the response spectrum obtained by linearization only spans the same
frequency range as the excitation spectrum. It is well known, however, that for nonlinear
systems, the response spectra can have significant values outside the frequency range of the
excitation. Consequently, some of the response statistical properties may be estimated
erroneously by this method.
The Volterra series method, described by Schetzen(1980) and Rugh(1981), is an
approximate analytical method which does not suffer from this drawback. This method is
best suited for systems with polynomial nonlinearities. However, in many engineering
applications, the nonlinearities are not polynomials and may not even be analytic.
Therefore, an equivalent stochastic "quadratization" method is proposed as an improvement
to the equivalent stochastic linearization method. The nonlinearity is replaced with
"equivalent" polynomials up to quadratic order. The resulting nonlinear "equivalent"
system is solved by the Volterra series method. The response probability distribution,
which in general is non-gaussian, is estimated by a Gram-Charlier expansion. In this
chapter the method is developed for a nonlinear single-degree-of-freedom oscillator subject
to a stationary, gaussian excitation. Response solutions are confmed to stationary
responses.
15
The significance of quadratic order terms is found to depend on the frequency

distribution of the excitation and on the degree of non-symmetry in the nonlinearity. A
description of the symmetry of a nonlinearity is given by way of the examples shown in
Figure 2.1. A symmetric and a non-symmetric nonlinearity are illustrated by the quadratic
stiffness and offset quadratic stiffness nonlinearities, respectively. It should be noted that
although a nonlinearity is symmetric about a particular reference point, it is non-symmetric
if the system is excited about some other reference point, such as a nonzero mean position.
Clearly, reliable analytical methods for dynamical systems which have non-symmetric
nonlinearities are needed.
2.2 Analytical Method Formulation

Consider the following nonlinear equation of motion of a single-degree-of-freedom
oscillator subject to a stationary, gaussian random force, f(t),
rnX + cx + kx + g(x,x) = f(t) (2.1)
Here, x == x(t) is the displacement of the oscillator, m is the mass of the oscillator, c is the
viscous damping coefficient, k is the spring stiffness, and g(x,x) is a nonlinear force which
may contain velocity and displacement terms. The notation, 0, denotes derivative with
respect to time, t.
The force excitation is described mathematically by the sum of a deterministic mean
part, /If, and a random part, which is a filtered white-noise process. The latter part is
expressed mathematically by a time-invariant linear transform involving the convolution of
white noise with the filter impulse response function hf(r). Thus, f(t) can be written as
follows
f(t) = /If + f hc('t) w(t-'t) d't (2.2)
where wet) is a zero-mean, gaussian, white-noise process. That is <w(t)w(t+'t» = 2n/)(r),

where the symbols <.> and /) stand for the operator of mathematical expectation and the
Dirac delta function, respectively.
The spectral density of the excitation is defmed by the equation
Sff(c.o) = 1.. j
2n """
<f(t)f(t+'t» exp( -ic.o't) d't (2.3)
16
g(x) = 'Y 11 +xl(1 +x)

non-symmetric
symmetric
Figure 2.1 Symmetry of Nonlinearity

17
Substituting equation (2.2) into equation (2.3) yields
(2.4)
in which HIm) is the filter transfer function. According to linear system theory, HIm) is
related to hf(t) by the following Fourier transform
HIm) = f hlt) exp( -iOl"c) d't (2.5)
Further, Sww(m) is the spectral density of wet) and is identically equal to one because wet)
is delta-correlated.
In general, the nonlinear function g(x,x) is non-symmetric with respect to (x ,x);
that is
g(x,x) :# - g(-x,-x) (2.6)
Consequently, the solution of equation (2.1) may not have a zero mean even if J.lf is zero.
Therefore, the stationary response is written as
x(t) = Ilx + x(t) (2.7)
where Ilx is the mean of x(t), and x(t) is a stationary, zero-mean process, which in general
is not gaussian. Substituting equation (2.7) into equation (2.1) gives
(2.8)
Ensemble averaging equation (2.8) yields the following equation for the mean response
(2.9)
The response, x(t), will be obtained by solving the response of an "equivalent"

quadratic system. The form of the equivalent system is constructed as follows
(2.10)
where the Clj'S are the quadratization coefficients. The quadratic term associated with the
as coefficient has been simplified to reflect the fact that for stationary responses <ik> = O.
It is also noted that the equivalent stochastic quadratization method reduces to the equivalent
stochastic linearization method for the special case when a3 = a4 = as == O.
18
The quadratization coefficients are computed such that

(2.11)
where the error, E, is defmed by the equation
E == kJlx+ g(~x+xJ) - ~f - (XIx - (X2k - (X3(x2-<x2» - (Xi k2 -<k2» - (Xsxk (2.12)
It can be shown that the necessary condition for minimization is given by the equation
(2.13)
where
(2.14)
(2.15)
and the superscript, T, denotes transposition.

Writing equation (2.13) in matrix form and accounting for the stationarity of the
response, yields the following symmetric system of linear equations
~X2 0 ~x3 ~xk2 0 (Xl <gb

~k2 0 ~k3 ~xk2 (X2 <gb
~x4 - (~x2)2 ~x~2 - ~x2 ~k2 0 (X3 = <gx2>_<g>~x2 (2.16)
~k4 - (~k2)2 ~xk3 (X4 <gk2>_<g>~k2
~x~2 (XS <gxx>
where the ~xmk.n'S are the joint central moments of x(t) and *(t) defined by the equation
(2.17)
The expectations in equation (2.16) will be evaluated by relying on the statistics of the
response of the nonlinear equivalent system.
In order to apply the Volterra series method to approximate the response of the
system, equation (2.10) must first be rearranged. All of the linear terms are kept on the left
hand side of equation (2.10). The quadratic terms are taken to the right hand side of the
equation and are considered to be unknown forces. Also, the scalar coefficient A. is
introduced for bookkeeping purposes and is set equal to one. Without altering the
19
equivalent system, the zero-mean force f-~ is replaced by A(f-~). The nonlinear equivalent
system is thus rewritten as
(2.18)
where
(2.19)
(2.20)
According to the Volterra series method, it is assumed that the solution to equation
(2.18) can be written as an infmite series in the following fOITIl
x(t) = L AjX(j)(t) (2.21)

j=l
As a practicality, the above series must be truncated. TeITIls up to at least j=2 must be
included to account for the effect of the quadratic teITIls. Therefore, teITIlS after j=2 are
neglected. Substituting the truncated fOITIl of equation (2.21) into equation (2.18) and
equating like powers of Aleads to
(2.22)
(2.23)
where
(2.24)
(2.25)
(2.26)
The nonlinear system in equation (2.10) is now approximated by two linear systems which
both have the same differential operator. These two systems are also intertwined since the
right hand side forces in equation (2.23) depend on the response of the system given by
equation (2.22).
Relying on linearity, the systems in equations (2.22) and (2.23) can be combined
into one system, which gives the following approximation for the nonlinear equivalent
system
20
(2.27)
Comparing equations (2.2) and (2.25), it is apparent that the force to)(t) is the linear filtered
white-noise process. It then follows from equations (2.22) and (2.26) that the force P)(t)
is a quadratic transfonn of w(t). Thus, the force f(t) is a second order Volterra series in
which white noise is the input function. This is expressed mathematically by the equation
(2.28)
where the functions hf1)('t) and hf2)('tI,'t2) are respectively linear and quadratic Volterra
kernels for force. They are also called linear and quadratic impulse response functions.
The kernels have the corresponding linear and quadratic transfer functions, Hf1)(co) and
impulse response function and transfer function of the white-noise filter respectively.
Thus, they are related by the linear transfonn in equation (2.5). The quadratic kernel and
quadratic transfer function are related by the following two dimensional Fourier transfonn
00
Hr)(CO I, C0 2) = If hT('tI,'t2) exp(-icoI'tI) exp(-ico2't2) d'tld't2 (2.29)
The forces f(1)(t) and P)(t) are referred to as the linear and quadratic forces respectively.
Since the differential operator in equation (2.27) is linear, the displacement
response of the system is given by the equation
(2.30)
where the tenn xG) is redefmed as the response due to the force to). Furthennore, the
steady-state response must also have a Volterra series representation of the fonn
21
where hi)(t) and h~)(tl,t2) are respectively the linear and quadratic Volterra kernels for
displacement. The associated transfer functions are Hi) (00) and H~)(00 1 ,002).
2.3 Derivation of Linear and Quadratic Transfer Functions

In the ensuing analysis, all numerical computations are carried out in the frequency
domain. Thus, for the Volterra series in equations (2.28) and (2.31), only the transfer
functions need to be computed.
As stated, the force transfer function Hl1)(oo) is the same as the linear white-noise filter
transfer function. The quadratic force transfer function is computed based on equation
(2.26). First, the velocity response, ~(t), must be obtained. Taking the time derivative of
the Volterra series in equation (2.31) leads to the following Volterra series for the velocity
~(t) (2.32)
where hi (t) and h~ (tl,t2) are the Volterra kernels for velocity. It can be shown that the
(I) (2)
associated transfer functions, H~\oo) and H~)(OOl'roz), are related to the displacement
transfer functions by the relations
H (~\oo) = iooH~)(oo) (2.33)
(2.34)
The quadratic kernel for the force is then found by substituting the linear
components of equation (2.31) and (2.32) into equation (2.26) and recalling equation
(2.28) which yields
(2.35)
22
Note that the quadratic kernel has been written in a symmetric fonn. The quadratic transfer
function for the force is found by substituting equation (2.35) into equation (2.29) and
recalling equations (2.33) and (2.34). The result is
(2.36)
The displacement response transfer functions are obtained by writing the steady-
state displacement response of the linear differential system in equation (2.27) alternatively
to equation (2.31) as
i(t) = Jh( t) l(t-t) dt (2.37)
The function h(t) is the impulse response function for the linear differential operator in
equation (2.27). It can be shown that the corresponding transfer function, H(ro), is
computed by the equation
1
H(ro) = -ro 2
m +·lroC eq + k eq
(2.38)
The displacement transfer functions for the Volterra series in equation (2.31), are
then found by substituting equation (2.28) into equation (2.37), which after some
mathematical manipulation yields
H (;A)(,.,)
UI = H( ro)H(1)(
lro ) (2.39)
(2.40)
Inspection of equations (2.36) reveals that the force quadratic transfer function
possesses the following symmetries
(2.41)
(2.42)
where (*) denotes complex conjugation. These properties greatly reduce the computational
effort and storage requirements. It follows from equations (2.34), (2.38), and (2.40) that
the response quadratic transfer functions have the same properties.
23
2.4 R~sponse Probability Distribution

In general, the probability distribution of x(t) is not gaussian. In particular, the
quadratic response, x(2)(t), is non-gaussian because it is a quadratic transfonnation of a
gaussian process. The exact distribution of the response is not known, but there are
methods available to approximate it. One such method is the Gram-Charlier expansion
which assumes that the distribution can be approximated by an expansion of derivatives of
a gaussian distribution, as described by Johnson and Kotz(1972). This method is
particularly well suited to the present problem since part of the response is already
gaussian. Thus, it is assumed that the joint probability distribution, p(xJ), of the
equivalent system can be approximated from the expansion of the joint gaussian
distribution, <I>(xJ), by the following relation
(2.43)
The function <I>(xJ) is given by the equation
(2.44)
in which the state space vector 5 is defmed by the equation
(2.45)
and the matrix S is the covariance matrix for the state space vector.
For a Gram-Charlier expansion, the coefficients rjJ2 are related to the cumulants of
the distribution as described in Appendix A. The relationship between cumulants, means,
and central moments is also given in Appendix A. The approximate probability distribution
in equation (2.43) can be simplified by recalling that the response is stationary which
reduces the joint gaussian distribution ,<I>(x,~), to the product of two one dimensional
gaussian distributions. Thus, the joint distribution given by equation (2.43) and truncated
after third order tenns can be written as
(2.46)
where
24
1
~(i) = exp(- i2) (2.47)
...J21tJ.1i'1 2J.1i2
1
~(i) = exp(- i2) (2.48)
...J 21tJ.1i2 2J.1i2
To completely defme the above joint distribution, the response statistics J.1x2, J.1i2,
J.1i3, J.1l3, and J.1H2 are required. Equations (2.31), (2.32) and (2.17) are used to compute
these joint central moments. For instance, the second moment J.1i2 is obtained by
substituting equation (2.31) into equation (2.17) for m=2 and n=O and taking the expected
value. Recalling the multiplicative properties of gaussian random variables and that the
process w(t) is delta-correlated gives
Equation (2.49) is converted into the frequency domain by using the well known relation
that the autocorrelation function is related to the frequency domain spectral density function
by a Fourier transform. Recalling that the Fourier transform of 21tB(t) is equal to one yields
(2.50)
In a similar manner, the other central moments are computed and are given by the following
equations
(2.51)
(2.52)
25
(2.53)
(2.54)
Note that if the quadratic transfer function is zero, the third order central moments are zero,
and according to equation (2.46), the joint probability distribution of the response is
gaussian. Thus, the distribution obtained by equivalent stochastic linearization is gaussian.
Response moments higher than third order, while not needed for the approximate
probability distribution as given by equation (2.46), are needed to solve for the
quadratization coefficients using equation (2.16). These higher order moments are
computed by using the assumed distribution. This can be done by using the characteristic
function of the distribution, but a simpler way is given in Appendix B and yields the
following expressions for the fourth order central moments
1lx4= 3(l!x2)2 (2.55)
1!i4= 3(l!i2)2 (2.56)
1lx2i2 = J.l,x2 l!i2 (2.57)
llxi2 = 0 (2.58)
The evaluation of expectations in the right hand side vector in equation (2.16) depends on
the nonlinearity, g(x,x).
26
2.5 Response Spectral Density

Besides the response statistics, the response spectral density function, SH(ro), is
often of interest in order to gain insight into the nature of the response. Using the
expression for x(t) given by equation (2.37) and the defmition of spectral density given by
equation (2.3), the following well known result for linear systems is obtained
Sxx(ro) = 1 H(ro) 12 SU(ro) (2.59)
where Srj{ro) is the spectral density of the force 1(t). The force spectral density is computed
similarly by using equation (2.28) for 1(t) which gives the equation
(2.60)
This spectral density function consists of two parts. The first part is the spectral density of
the linear filtered white-noise force, i<l)(t). The second part is the spectral density of the
quadratic force, i<2)(t), and involves a frequency domain convolution. The convolution
introduces spectral contributions in the force 1(t) which are at frequencies that are the
difference and sum of frequencies of the filtered white noise. In the example problems
which follow, it is shown how this phenomenon can cause resonance responses outside the
frequency range of the filtered white noise.
2.6 Solution Procedure

Clearly, for the implementation of the preceding solution scheme, an iterative
procedure is required as the quadratization coefficients are initially unknown. The
procedure is started by assuming initial values for the coefficients. Zero-valued coefficients
may be used if better estimates are not available. The linear transfer function in equation
(2.38) is computed first and is used to compute the linear transfer functions in equations
(2.33) and (2.39). Next, the quadratic transfer function for force is computed using
equation (2.36). The response quadratic transfer functions in equations (2.34) and (2.40)
are then computed. The linear and quadratic transfer functions are used to compute the
mean by equation (2.6) and the central moments by equations (2.50) to (2.54). The
expectations involving the nonlinearity g(x,x) are computed using the defmition of
expectation and the joint probability distribution p(x,k). For relatively simple
nonlinearities, this can be done analytically, otherwise, numerical integration methods are
27
used. A new set of quadratization coefficients is then obtained from equation (2.16). This
procedure is repeated until convergence.
2.7 Example of Application

The equivalent stochastic quadratization method has been used for the following
nonlinear system
x + ~i + x + -yla + il(a + i) = f(t) (2.61)

with
~ =.1 r =1 (2.62)
and
H[(ro) =
{ 0
1
...j 2( rob-roa)
everywhere else
(2.63)
The above linear filter yields a block banded excitation spectral density, as shown in Figure
2.2, with a standard deviation equal to one. The arrow at the zero frequency represents a
delta function and is the spectral contribution of the mean force. In this example, however,
it is assumed that Ilf =O. The parameter a is referred to as the velocity offset.
According to the described quadratization procedure, the mean response is
Ilx = - <ria + ~I(a + ~» (2.64)
and the equivalent quadratic system is

ji + ~~ + x+ (X2i + (X4(~2_<~2» =f(t) (2.65)
Expectations of the following form are evaluated for equation (2.64) and the
elements of the right hand side vector in equation (2.16)
k = 0, 1,2 (2.66)
The evaluation of these expectations is described in Appendix B.

Analytical results have been obtained by both equivalent stochastic linearization and
equivalent stochastic quadratization for this problem. Further, Monte Carlo simulation has
been performed to assess the accuracy of these results. The simulation is performed by
28
O)wij = O)b - O)a
O)a + O)b
=
~ \ J.1~ 15(0))
O)oen
2
0)
o
Filtered White Noise Process
o
White Noise Process
Figure 2.2 Excitation Spectral Density

29
using the sum ofhannonics method, described by Borgman(1969), to produce 300

realizations. In all of the cases the band width frequency, Cllwid' is equal to four.
For a velocity offset, a, equal to .3 and a center band frequency, "\:en, equal to six,
Figure 2.3 shows the one-sided form of the force power spectral density Sff{co) as
computed by equation (2.60). The spectral density of the linear force is the same as the
flltered white-noise spectrum. The spectrum of the quadratic force is a convolution of the
flltered white-noise spectrum, which gives low and high frequency contribution~.:
Figure 2.4 shows the corresponding one-sided form of the displacement response
spectral density obtained by quadratization, linearization, and simulation. The peak at the
excitation frequencies is the response due to the linear force. The low frequency peak is a
resonance response due to the quadratic force. The response due to the high frequency
components of the quadratic force is negligible. It can be seen that the spectra obtained by
simulation and quadratization are in good agreement. The quadratization method, however,
is computationally much more efficient, to the extent that it reduces the requisite
computation time by two orders of magnitude. The spectrum from linearization is virtually
identical to the spectrum from quadratization at the excitation frequencies, but cannot
account for any response frequencies outside that range. It should be noted that even
though the quadratic force spectrum is several orders of magnitude less than the linear force
spectrum, its contribution to the response spectrum is the same order of magnitude because
the dynamic amplification at resonance is several orders of magnitude greater than at the
excitation frequencies.
The probability distribution of the response is shown in Figure 2.5. The
distribution obtained by simulation and quadratization are in good agreement. The vertical
lines mark the computed mean responses and clearly indicate that the simulated and
quadratization solutions are skewed. The linearization solution is gaussian. It is noted that
the correlation between the probability distribution obtained by simulation and
quadratization is better near the mean than in the tails of the distribution. In fact, the tail of
the distribution obtained by quadratization even becomes negative over a small range. This
of course, is physically impossible, but can result mathematically from the Gram-Charlier
expansion. Including higher order moments in the expansion may lessen the degree of this
problem, but negative values for probability distributions are still possible as shown by
Crandall(1985). In addition, higher order moments are costly to compute. It is
emphasized that this is not a problem of inaccurately computing the moments, rather it is a
30
0.4
- - Linear Force
- - - Quadratic Force x 1000
0.3
0.2
\
\
\
\
0.1 \
\, , '
/
/
,
,,--- ..........,
"
/
0.0
"
'~
- --r- - ,
"
......... --- ..
0 2 4 6 8 10 12 14 16
CJJ rad/sec
Figure 2.3 Volterra Series Force Spectral Density

(a = .3)
1.2
1.0 - - Equiv. Quad .

- - Equiv. Lin.
0>
0.8 ........ Sim ulation
~
0.6
j~
0.4
0.2
0.0
0 2 4 6 8 10 12 14 16
CJJ rad/sec
Figure 2.4 Displacement Response Spectral Density

(a = .3)
31
12 c------------------------------------------,
/1',
10
/ I \ - - Equiv. Quad.
I I
I - - Equiv. lin.
8
I •••••••. Simulation
I
I
I
I
I
4
o~~~=-~ ____ ~~ ____ ~~~====~

.(l.3 -0.2 -0.1 0.0 0.1
x
Figure 2.5 Displacement Response Probability Distribution

(a= .3)
0.0
-0.1
-0.2
.:f
·0.3 - - Equiv. Quad.

- - Equiv. lin.
• Simu lation
-0.4
·0.5
0.0 0.2 0.4 0.6 0.8
a
Figure 2.6 Mean Displacement Response

32
problem of not having enough moments to accurately describe the tails of distribution.
From a design viewpoint, this means that the analytical distribution would not be
appropriate for reliability analyses, but would be suitable for fatigue analyses. In any
event, the non-gaussian distribution is a substantial improvement over the gaussian
distribution obtained by linearization.
In Figures 2.6, 2.7, and 2.8, the mean, standard deviation, and skewness of the
displacement are plotted versus the velocity offset. The linearization and quadratization
results are nearly identical to simulation for the mean response. The response standard
deviations obtained by quadratization are also quite reliable. In comparison, linearization
significantly underestimates the simulation, because the resonance response is not
accounted for. The skewness obtained by quadratization overpredicts the simulation, but
has the correct trend. Linearization fails to exhibit any skewness since the probability
distribution is gaussian.
Note that as the velocity offset tends to zero, the statistics computed by linearization
and quadratization converge. This is because for zero velocity offset the nonlinearity is
perfectly symmetric. Thus, the equivalent quadratic system only consists of the linear
terms.
In Figure 2.9, the effect on the response standard deviation by changing the
location of OOeen is illustrated. The standard deviation is plotted versus the velocity offset
for excitation spectra with OOeen equal to 2, 4, and 6, and OOwid equal to 4. To compare the
results, the standard deviation is normalized by the standard deviation of the system with 'Y
=O. Also, the velocity offset is normalized by the velocity standard deviation. Since the
band width of the excitation spectra are the same, the range of the low frequency quadratic
force spectra are also the same and span from -4 to +4. Noting that the natural frequency
of the system is equal to one, a quadratic resonance response occurs for all three excitation
spectra. Because the quadratic force spectra are several orders of magnitude less than the
linear force spectra, as illustrated by Figure 2.3, the relative magnitude of the quadratic
response to the linear response depends on the frequency range of the linear excitation. As
shown in Figure 2.4, for OOeen relatively high compared to the natural frequency, the
quadratic and linear responses are of the same magnitude. This is due to the fact that the
dynamic amplification of the quadratic response is much larger than the dynamic
amplification of the linear excitation. However, for OOeen equal to 2, the linear and quadratic
force fall in the same frequency range. Thus, the quadratic response is negligible compared
33
N
....
0 3
- - Equiv. Quad.
~ 2 - - Equiv. Lin.
• Simulation
o ~ ________ ~ ________ ~ ________ ~ ________ ~
0.0 0.2 0.4 0.6 0.8

a
Figure 2.7 Displacement Response Standard Deviation
o -----------------.
- - Equiv. Quad.
- - Equiv. Lin.
·2
• Simulation
"'~
x
;,..
·4
• •
•
·6
·8 I:....________..J....________...L..________-..I.________-...I
0.0 0.2 0.4 0.6 0.8

a
Figure 2.8 Displacement Response Skewness

34
2.0
- - Equiv. Quad.
- - Equiv. lin.
o 1.5 • Simulation
•
.-!"
1.0
- - - - - ----
] (a
(b
0.5
(c
0.0
0 2 3 4
a/(,fiS>Y
x
.l
Figure 2.9 Normalized Displacement Response Standard Deviation

a) (O",m = 6 b) "ten = 4 c) (O",m = 2
35
to the linear response, and linearization and quadratization give nearly identical solutions.
This force excitation corresponds to clipped white noise.
The quadratization procedure has also been used to analyze the response of the
following oscillator with a stiffness nonlinearity
x + I3x + x + ylxlx = f(t) (2.67)

with
13 = .1 y=5 (2.68)
The linear force fIlter is again given by equation (2.63). The nonlinearity is symmetric if
the mean displacement is zero. However, this would be unlikely for a mean force, Ilf, that
is not equal to zero. Hence, the nonlinearity is considered to be non-symmetric.
Following the quadratization procedure, the mean response is computed by the
equation
Ilx = Ilf - <y1llx + xl(llx + x» (2.69)
and the equivalent quadratic system is

~ + 13~ + x + alx + a3(x 2-<x 2» =f(t) (2.70)
Expectations of the following form are evaluated for equation (2.69) and the
elements of the right hand side vector in equation (2.16)
k =0, 1,2 (2.71)
The evaluation of these expectations is again described in Appendix B.

Numerical results are presented for the nonlinear system in equation (2.67). The
band width frequency, (Owid' is equal to four, and the center band frequency, (Ocen' is equal
to six. Assuming a mean force equal to .1, displacement response spectral densities have
been obtained by quadratization, linearization, and simulation and are shown in Figure
2.10. Again,linearization cannot predict the low frequency resonance response.
Quadratization and simulation are in good agreement except for a slight difference in the
height of the resonance peak. It is noted that the resonance peak occurs at a frequency of
1.34 instead of 1.00 because of the additional linear stiffness.
The probability distribution of the displacement is given in Figure 2.11. There is
very good agreement between the quadratization and simulation and a slight skewness of
the distribution is noticeable.
36
1.6
1.4 - - Equiv. Quad.

- - Equiv. Lin.
1.2
........ Simulation
..,
~ 1.0
x
0.8
~!!
II)
0.6
0.4
0.2
0.0
0 2 4 6 8 10
CJ) radlsec
Figure 2.10 Displacement Response Spectral Density

(~ = .1)
12
10 - - Equiv. Quad.
- - Equiv. Lin.
8 ........ Simulation
i 6
o
-0.10 ·0.05 0.0 0.05 0.10 0.15 0.20 0.25
x
Figure 2.11 Displacement Response Probability Distribution

(~ = .1)
37
In Figures 2.12 to 2.14, the mean, standard deviation, and skewness of the
displacement are plotted versus the mean force. The accuracy of the various methods is
similar to the previous example. Points that are worth mentioning are that for a zero-mean
force the nonlinearity is symmetric. Therefore, linearization and quadratization give the
same results. The relative "unsmoothness" of the quadratization curves in Figures 2.13
and 2.14 is related to the sharpness of the resonance peak and the use of digital means to
perform the analytical integrations. Because the frequency of the resonance peak shifts for
different mean forces, the true resonance peak is usually missed by slightly different
amounts which affects the integration results.
2.8 Summary and Conclusions

As an extension of the equivalent stochastic linearization method, an equivalent
stochastic quadratization method has been developed for stationary response analysis of
nonlinear single-degree-of-freedom systems subject to filtered gaussian white-noise
excitation. The stationary response is written as the sum of a mean time-independent
response and a stationary zero-mean response which is assumed to be non-gaussian. The
mean response is determined from expectations involving the original nonlinear equation.
The zero-mean response is obtained as the solution to an "equivalent" quadratic system
which is constructed by replacing the nonlinearity with polynomials up to quadratic order.
The equivalent system has a form whose solution can be approximated by using the
Volterra series method. The coefficients of the polynomials are determined from a mean
square minimization. A third order Gram-Charlier expansion is used to describe the non-
gaussian response probability distribution, and the system response statistics are
determined in an iterative manner.
The method is applied on two example systems: one with a damping nonlinearity
the other with a stiffness nonlinearity. The analytical results are compared to results
obtained by simulation to verify their reliability. The solution obtained by the proposed
method provides a notable improvement over the solution obtained by the equivalent
stochastic linearization method. The response spectral densities obtained by quadratization
and simulation are in good agreement and exhibit significant resonance responses which
can not be accounted for by linearization because they occur at frequencies outside the
range of excitation frequencies. In addition, the probability distributions obtained by
quadratization are in general non-gaussian and have good agreement with the probability
38
0.10
- - Equiv. Quad.
0.08 - - Equiv. Lin.
• Simulation
0.06
0.04
0.02
0.0 ""-_--'_ _-'-_ _-'-_ _'--_......L_ _......._ _-'--_--'

0.0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
11,
Figure 2.12 Mean Displacement Response
4.0
3.8
•
(II
--_.
~ 3.6
~ 3.4 - - Equiv. Quad.

- - Equiv. Lin.
• Simulation
3.2
3.0
0.0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
11,
Figure 2.13 Displacement Response Standard Deviation
39
0.0 --------------------.
-0.2 - - Equiv. Quad.

- - Equiv. Lin.
-0.4 • Simulation
.,
~
x -0.6
J...
•
-0.8
•
-1.0
-1.2 "'-_-'"_ _--'-_ _-'--_----"'--_---'-_ _-'-_ _-'--_---'

0.0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
Ily

40
distributions obtained by simulation. On the other hand, the probability distributions

obtained by linearization are gaussian. The significance of quadratic order tenns is found
to depend on the frequency distribution of the excitation and on the degree of non-
symmetry in the nonlinearity. When the symmetry of the nonlinearity is symmetric the
quadratization solution reduces to the linearization solution. The quadratization method is
quite efficient to the extent that the requisite computation time is two orders of magnitude
less than that needed by simulation.
CHAPTER 3
EQUIVALENT STOCHASTIC QUADRA TIZATION
FOR MULTI-DEGREE-OF-FREEDOM SYSTEMS
3.1 Introduction
In this chapter the equivalent stochastic quadratization method previously developed
for single-degree-of-freedom systems is extended to multi-degree-of-freedom sy's~ems.
The method is fIrst developed for a system with a general nonlinearity. The computational
effort involved in analyzing this kind of system, however, can be quite expensive.
Therefore, a simplified version of the method is also developed for systems whiCh have
simpler nonlinearities. The reduced method is exemplifIed by considering a system with
two degrees of freedom.

Consider the following nonlinear equation of motion of a multi-degree-of-freedom
system subject to a stationary, gaussian random force vector, f(t),
m X + c X + k x + g(x,x) = f(t) (3.1)
The system has N degrees of freedom, so that the vector x is an Nx 1 displacement

response vector, and m, c, and k are NxN mass, damping, and stiffness matrices,
respectively. The vector g(x,x) is the nonlinear force vector which may contain velocity
and displacement terms. The ith component of this vector is assumed to be of the following
general form
N
gj(x,x) = .2r gj/Xj,Xj) (3.2)
The force excitation is described mathematically by the sum of a deterministic mean

vector, !!-f, and a vector of random fIltered white-noise processes. The latter part is
expressed mathematically by a time-invariant linear transform. Therefore, f(t) can be
written as follows
f(t) = !!-f + JhI<'t) w(t-'t) d't (3.3)

42
As in the previous chapter, wet) is a zero-mean, gaussian, white-noise process. That is

<w(t)w(t+t» =21to(t), where the symbols <.> and 0 stand for the operator of
mathematical expectation and the Dirac delta function, respectively.
The spectral density matrix of the excitation is given by the equation
= 2~ 1
00
Sff{O» <f(t) fT(t+t» exp(-io>t) dt (3.4)
Substituting equation (3.3) into equation (3.4) yields
(3.5)
In this equation, HI: 0» is the filter transfer function vector and is related to filter impulse
response function vector, bc(t), by the following Fourier transform
Hl:o» = f llc(t) exp(-io>t) dt (3.6)
The symbol Sww(O» denotes the spectral density of wet) and is identically equal to one
because wet) is delta-correlated.
In general, the nonlinear force vector, g(x,x), is non-symmetric. Hence, the
solution of equation (3.1) may not have a zero mean. Therefore, the stationary response is
written in the following form
x(t) = Ux + get) (3.7)
where !Ix is the mean displacement vector, and get) is a vector of stationary, zero-mean
processes, which in general are not gaussian. Substituting equation (3.7) into equation
(3.1) gives
(3.8)
Ensemble averaging equation (3.8) leads to the following equation for the mean response
!Ix = k- 1[Uf - <g(Ux+g,g»] (3.9)
The response get) is obtained by solving the response of an "equivalent" quadratic

multi-degree-of-freedom system. Although the equivalent system is also nonlinear, it has a
43
fonn which lends itself to treatment by the Volterra series method. The fonn of the
equivalent system is constructed as follows
m t + c i + k X + UIX + ~i + u3 { [x,xl-<[x,xl> }
(3.10)
where the ~IS are the quadratization coefficient matrices. The notation [.,] denotes a
vector obtained by the following opemtion
N
[x,x] =I,
i=l
(3.11)
where the vector ~i is the ith column of the NxN identity matrix. Note that the quadratic
vector associated with the CIs coefficient matrix has been simplified to reflect the fact that
for stationary responses <[x,i]> = O. It is also noted that the equivalent stochastic
quadratization method reduces to the equivalent stochastic linearization method for the
special case whenClJ = U4 = Us == o.
The quadratization coefficient matrices are computed such that
<r,Tp = minimum (3.12)
where the error vector,.E, is defmed by the equation
.E == ku.x + g(u.x+x,i) - U.f - UIX - U2i - ClJ {[x,xl-<[x,x]> }
- u4 {[i,g]-<[g,g]>} - us[x,g] (3.13)
It can be shown that the necessary condition for minimization is given by the equations
i=ltoN (3.14)
where
(3.15)
and
(3.16)
The notation OJ denotes the ith row of the matrix. The N linear systems of 5N equations
given by equation (3.14) must be solved to obtain the quadmtization coefficients. The left
44
hand side matrix, whose elements consist of the joint central moments of the response, is
the same for each of the N systems, but the right hand side vector is different.
The expectation on the left hand side of equation (3.14) involves joint central
moments of g and g up to fourth order. These moments are evaluated from the response of
the nonlinear equivalent system. Applying the Volterra series method in a manner similar
to the one of Chapter 2 leads to the following approximate system to the nonlinear
equivalent system
(3.17)
where
(3.18)
(3.19)
The force i(1)(t) is the zero-mean part of the linear filtered white-noise process in
equation (3.3). The force i(2)(t) is a quadratic transform of the white-noise process. Thus,
the force i(t) is a second order Volterra series in which white noise is the input function.
This is expressed mathematically by the equation
where the functions bt('t) and ~2)('tI''tZ) are respectively linear and quadratic Volterra
kernel vectors for force. The kernels have the corresponding linear and quadratic transfer
function vectors, ut(ro) and Hj2)(rol'roZ). The linear kernel ~1)('t) and transfer function
Ht(ro) are the same as the linear impulse response function and transfer function of the
white-noise filter, respectively. Thus, they are related by the linear transform in equation
(3.6). The quadratic kernel and quadratic transfer function are related by the following two
dimensional Fourier transform
(3.21)
The forces i(1)(t) and i(2)(t) are referred to as the linear and quadratic force vectors.
45
Because the differential operator in equation (3.17) is linear, the displacement

response of the system is given by the equation
(3.22)
where the tenn gG) is dermed as the response due to the force tG). Furthennore, the steady-
state response can be expressed as a Volterra series in the fonn
where h£)(t) and hi)(tl,t2) are the linear and quadratic Volterra kernels respectively for
displacement. The associated transfer functions are ~)(ro) and Hi)(rol ,ro2).

In the ensuing analysis, all numerical computations are carried out in the frequency
domain. Thus, for the Volterra series in equations (3.20) and (3.23), only the transfer
function vectors need to be computed.
The force transfer function Hi1)(ro) is the same as the linear white-noise fIlter
transfer function. The quadratic force transfer function is computed by the Volterra series
method in the same manner as in Chapter 2, and the result is given by the equation
(3.24)
The displacement response transfer functions are obtained by writing the steady-
state displacement response of the linear differential system in equation (3.17) alternatively
to equation (3.23) as
,&(t) = J h(t) t(t-t) dt (3.25)
The function h(t) is the impulse response function matrix for the linear differential operator
in equation (3.17). It can be shown that the corresponding transfer function, H(ro), is
(3.26)
46
The displacement transfer functions for the Volterra series in equation (3.23), are
then found by substituting equation (3.20) into equation (3.25), which after some
mathematical manipulation yields
(1) (I)
IIi (00) = H(oo)Ht (00) (3.27)
(3.28)
As before, the force quadratic transfer function vector possesses the following
symmetries
(3.29)
(3.30)
where (*) denotes complex conjugation. It follows from equations (3.26) and (3.28) that
the response quadratic transfer functions have the same properties.
3.4 Response Probability Distribution

In general, the joint probability distribution of !(t) is not gaussian. Approximate
joint distributions can be obtained by an expansion of partial derivatives of a joint gaussian
distribution. Thus, the joint probability distribution, p(S), between the N displacement
components and their corresponding velocities of the equivalent system is approximated by
the following relation
(3.31)
where jT =jl + h +.. -+ hN" The symbols denotes the state space vector
(3.32)
clI(S) is a joint gaussian distribution given by the equation
clI(S) = (21t)"N ISrl/2 exp[ - ~ ST S-l 5] (3.33)
where the matrix S is the covariance matrix for the state space vector.
47
For a Gram-Charlier expansion, the coefficients rjljZ"'j2N are related to the

cumulants of the distribution as described in Appendix A. The relationship between
cumulants, means, and central moments is also given in Appendix A. The central moments
of 5are dermed by the equation
(3.34)
Since the computation of moments higher than third order becomes quite costly, the
distribution in equation (3.31) is truncated after third order tenns. For the system in
equation (3.17), the moments up to third order are computed from equation (3.23) and its
time derivative. Thus, an arbitrary second moment is obtained by the following equation
where H~: and H~:(0)1 ,~) are linear and quadratic transfer functions of the state space
variable~. In a similar manner, an arbitrary third order central moment is obtained by the
equation
(3.36)
Response moments higher than third order, while not needed for the approximate
joint probability distribution as given by equation (3.31), are needed to solve for the
quadratization coefficient matrices using equation (3.14). These higher order moments are
computed by using the approximate joint distribution. Using the characteristic function of
this joint distribution it can be shown that a general fourth order central moment is given by
48
(3.37)

In addition to the response statistics, the response spectral density matrix, Sxx(ro) ,
is often of interest. Using the expression for g(t) given by equation (3.25) and the
deftnition of spectral density given by equation (3.4) results in the following well known
equation for linear systems
(3.38)
where Sr/ro) is the spectral density matrix of the force l(t). The force spectral density
matrix is computed similarly by using equation (3.20) for l(t). This leads to
(3.39)
Thus, the force spectral density matrix consists of two parts. The ftrst part is the spectral
density of the linear fIltered white-noise force, l(l)(t). The second part is the spectral
density of the quadratic force, l(2)(t), and involves a frequency domain convolution.

The implementation of the preceding solution scheme follows the same iterative
procedure as for the single-degree-of-freedom system. The expectations in equation (3.14)
are evaluated based on initial values for the quadratization coefftcient matrices and are used
to compute new quadratization coefficient matrices. This procedure is repeated until
convergence. Note that the number of moments needed to solve equation (3.14) is less
than is needed to deftne the truncated fonn of the joint distribution in equation (3.31).
Thus, it makes sense to compute only the moments needed for equation (3.14) dupng the
iteration process and compute other moments after convergence. This saves considerable
computational effort. However, it is still quite costly to compute all the moments and in
particular the third order moments. This effort can be greatly reduced if the components of
the nonlinear force vector, g(~,!), are of the following simplified fonn
(3.40)
A simplifted solution procedure is implemented for this case in the next section.
49
3.7 Reduced Solution Analytical Method

For the simplified fonn of the nonlinearity given by equation (3.40), the linear
system of equations represented by equation (3.14) reduces from a 5Nx5N system to a 5x5
system where the vector X is now given by the equation
-T
X = (x j, x':' j, (xA2j - <xA2j »
A
, (. x.:.2j - <x.:.2j » , xj x':' j )
A
(3.41)
and the vector g is

(3.42)
where alii refers to the ith diagonal element of the respective matrix. Thus, only moments
of the i th degree of freedom are needed to solve for the quadratization coefficients
associated with that degree of freedom. Joint moments of the i th and jth degree of freedom
may still be of interest, but would not need to be computed until after the iteration process
has converged.
The following Gram-Charlier expansion for the joint probability distribution of the
displacement and velocity of the ith degree of freedom is used to solve for the expectations
on the right hand side of equation (3.14) and the mean displacement given by equation
(3.9)
(3.43)
Here the joint central moments, Jlx'1'1 k'!,1 are defmed as
(3.44)
and cp(Xj) and CP(~j) are the following one dimensional gaussian distributions
exp(- XT)
21lXf
(3.45)
(3.46)
Note that the distribution in equation (3.43) has been simplified fo reflect the fact that for
lli.~k.1 and Jlx.1 k.1 = 0
stationary responses .---Ai.
50
The most significant savings for this reduced method is due to the fact that
considerably fewer third order moments need to be computed.
3.8 Example of Application

The discussed quadratization procedure has been used for the two-degree-of-
freedom system shown in Figure 3.1. The nonlinear damping is given by the following
equation
i = 1,2 (3.47)
For this type nonlinearity the reduced solution procedure can be used. As a specific case,
the system properties are: kl = 1, k2 = 1, ml = 1, m2 = 1, cl = .1, c2 = .1,11 = 1,12 = 1,
and al = a2 = a where a is a parameter to be varied. The filtered white-noise excitation
vector, f(t), is obtained from the following filter transfer functions
1
(3.48)
everywhere else
The elements of the covariance matrix for the excitation are all equal to one. The mean
force vector, !H, is assumed to be zero-valued.
According to the described quadratization procedure, the mean response is
lh
(2-1 -11 )-l( <Ia
=_ + xll(a + Xl»
<Ia + x21(a + x2»
)
(3.49)
and the equivalent quadratic system is given by the equation

m i + c g + k g + azg + U4 {[g,g]-<[g,g]>} = Ht) (3.50)
The expectations in equation (3.49) and the right hand side vector in equation
(3.14) are solved using equations derived in the previous chapter and Appendix B.
Numerical results have been obtained for the two-degree-of-freedom system using
both linearization and quadratization for this problem. Further, Monte Carlo simulation has
been performed to assess the accuracy of these results. The simulation is performed by
51
-
f1 (t)
m1
C2
-f2(t)
m2
~) ~2)
LX1 LX2
Figure 3.1 Two-degree-of-freedom nonlinear system

52
using the sum of harmonics method to produce 300 realizations. The following
frequencies parameters have been used for the linear filter
Cil a =4 CillJ =8 (3.51)
For 11 ="f2 =0, the system is linear with natural frequencies, Cill = .618 rad/sec
and ~ = 1.618 rad/sec. Both of these frequencies are well away from the linear excitation
frequencies.
For the velocity offset, a, equal to .3, the displacement spectral densities of mass 1
and mass 2 are shown in Figures 3.2 and 3.3 respectively. A low frequency resonance
peak due to quadratic order forces is present in both masses near the first natural frequency.
For mass 2, this quadratic response is almost the same magnitude as the linear response.
Linearization does not predict the quadratic order response.
The probability distributions for both masses are shown in Figures 3.4 to 3.5. The
vertical lines mark the mean displacements. The simulation and quadratization distributions
are clearly skewed and agree to a reasonable degree. However, as in the sdof oscillator
with the same nonlinearity, the analytical non-gaussian distributions have regions in the tail
which are negative.
In Figures 3.6 to 3.8, the mean, standard deviation, and skewness of the
displacements of mass 1 and 2 are plotted for a range of the velocity offset parameter. As
in the sdof system, the mean responses obtained by all methods are in good agreement. On
the other hand, linearization underpredicts the standard deviation by as much as 100% in
some cases. Linearization and quadratization converge for a =0, because of the symmetry
°
of the nonlinearity. It is noted that the analytical standard deviations at a = disagree
slightly with the simulation standard deviations. This is probably due to higher order
responses which are not considered. The quadratization method gives reasonably good
estimates for the skewness as well.

A stochastic quadratization method has been developed for stationary response
analyses of nonlinear multi-degree-of-freedom systems subject to filtered gaussian white-
noise excitation. The stationary response is written as the sum of a mean time-independent
response and a stationary zero-mean response which is assumed to be non-gaussian. The
mean response is determined from expectations involving the original nonlinear equation.
53
4 - - Equiv. Quad.
- - Equiv. Lin .
..,o
- 3
••....•. Simulation
:§:
'>i:
CI)'" 2
o
o 2 4 6 8 10
(J) rad/sec
Figure 3.2 Displacement Response Spectral Density for Mass 1

(a= .3)
5 c-------------- - - - - - ----,
4 - -- Equiv. Quad.
- - Equiv. Lin .
..,
o .....•.. Simulation
3
o~~~====~
o 2 4
____~==~====d
6 8 10
(J) rad/sec
Figure 3.3 Displacement Response Spectral Density for Mass 2

(a = .3)
54
12
10 - - Equiv. Quad.
- - Equiv. Un .
8 •.....•. Simulation
)(~
c:: 6
0
-0.5 -0.4 -0.3 -0.2 -0.1 0.0
Figure 3.4 Displacement Response Probability Distribution Mass 1

(a= .3)
12
"
II \
I I \
10 - - Equiv. Quad. I I \
- - Equiv. Un. I I \
I I \
8 ......•. Simulation
I I \
I I \
'N I
~
0.
6 I
I
I :
4
l
2
0
·0.7 -0.6 ·0.5 -0.3 -0.2 ·0.1
Figure 3.5 Displacement Response Probability Distribution Mass 2

(a= .3)
55
0.0 IL""--------------------,
-0.1
-0.2
::! -0.3 - - Equiv. Quad.

- - Equiv. Lin.
-0.4 • Simulation
-0.5
-0.6 C-_ _ _-L-_ _ _---L_ _ _ _.L..-_ _ _-1..._ _ _---I
0.0 0.1 0.2 0.3 0.4 0.5

a
Figure 3.6 Mean Displacement Responses
- - Equiv. Quad.
2
- - Equiv. Lin.
• Simulation
o
0.0 0.1 0.2 0.3 0.4. 0.5
a
Figure 3.7 Displacement Response Standard Deviations

56
0.0
-D.5
• • ] m1
-1.0
...0
.-
x
. -1.5
•
c:1" •
-2.0 - -
•
Equiv. Lin.
Simulation
]~
-2.5
-3.0
0.0 0.1 0.2 0.3 0.4 0.5
a

57
The zero-mean response is obtained as the solution to an "equivalent" quadratic system

which is constructed by replacing the nonlinearity with polynomials up to quadratic order.
The equivalent system has a form whose solution can be approximated by using the
Volterra series method. The coefficients of the polynomials are determined from a mean
square minimization. A third order Gram-Charlier expansion is used to describe the joint
non-gaussian response probability distribution, and the system response statistics are
determined in an iterative manner.
The proposed method, developed for systems with general nonlinearities, can be
quite cumbersome since many higher order moments need to be computed. Therefore, a
reduced method for simpler nonlinearities is developed which requires less computational
effort. The reduced method is demonstrated on a system with two degrees of freedom. As
shown for single-degree-of-freedom systems, significant resonance responses which can
not be accounted for by linearization are accurately predicted by quadratization. The
quadratization results are obtained quite efficiently and compare well with time domain
simulation data.
CHAPTER 4
POTENTIAL WAVE FORCES ON A MOORED VERTICAL CYLINDER
4.1 Introduction
In this chapter, analytical expressions are derived for the linear and quadratic
potential wave forces acting on a moored, vertical, and surface-piercing cylinder. The
forces are described by a second order Volterra series with the linear wave elevation as the
input function. The linear and quadratic transfer functions which define the Volterra series
are derived based on linear diffraction theory. As discussed in Chapter 1, several potential
forces are to be derived. These are the linear potential force, the quadratic wave elevation
force, the quadratic velocity head force, and the quadratic body motion force. The resulting
expressions are verified against existing solutions as available. The last section of this
chapter describes how the linear and quadratic transfer functions for potential forces on a
TLP which consists of four cylinders are obtained from the force transfer functions for a
single cylinder. Further pertinent numerical results are presented.
4.2 Volterra Series Force Description

Consider the moored, vertical, and surface-piercing cylinder with radius a and
submerged depth ds in an infmite two-dimensional fluid domain of depth d shown in
Figure 4.1. It is assumed that the cylinder is constrained to move in-plane exhibiting
rotation and horizontal and vertical translations. However, it is assumed that magnitude of
the rotation and vertical translation of the cylinder is small enough to neglect their effect on
the force calculation. Therefore, for the present study only motion in the horizontal x
direction is considered. This is a reasonable assumption for a TLP column member. The
time dependent variable Xl represents the horizontal displacement of the cylinder centerline
in the fixed coordinate system, (O-x-y-z), which has its origin located at the still water line.
A body coordinate system, (Or-Xr-Yr-z), coincides with the fixed system when the cylinder
has zero displacement. A cylindrical body coordinate system (r,e,z) is also introduced
where
xr = r cos e (4.1)
Yr = r sin e (4.2)
59
Figure 4.1 Model of Moored Vertical Cylinder

60
It follows from Figure 4.1 that

x = rcos 9 + xl (4.3)
Y = r sin 9 (4.4)
The elements of the wave force vector, {, are the horizontal force, f l , the vertical
force, f 2, and the moment, f3. The moment is taken about the elevation of the center of
gravity of the cylinder, Zcg, which is somewhere along the cylinder's centerline. The force
vector is expressed as a Volterra series with the linear incident wave elevation, 11(t), as the
input function. Specifically,
(4.5)
where
00
f(1l(t) = J h}ll(t) 11(t-t) dt (4.6)
00
f(2l(t) = II hj>(tl,t2) 11(t-tl)11(t-t2) dtldt2 (4.7)
The force vector f(ol is constant in time, and the vectors t1) and (ll are respectively linear
and quadratic forces in which the functions h}ll (t) and hil(t I't2) are the linear and
quadratic Volterra kernel vectors. These expressions can also be written in the following
frequency domain fonn
00
{(1l(t) = J m1)(ro) ij(ro) exp(irot) dro (4.8)
00
f(2l(t) = II Hil(rol,ro2) ij(rol) ij(ro2) exp(irolt) exp(iro2t) droldro2 (4.9)
The symbols H}l l(ro), Hil(rol,ro2), and ij(ro) are the Fourier transfonns ofh}ll(t),
hil(tl>t2)' and 11(t) given by the equations
00
H}ll(ro) = Jllrl(t) exp(-irot) dt (4.10)

61
00
ej>(ro l,ro2) = If bf>('tl,'t2) exp(-irol'tl) exp(-iro2't2) d'tld't2 (4.11)
il(ro) = 1 jll(t) exp(-irot) dt

-2
1t......
(4.12)
ej>(ro) and e?>(rol'~) are known as linear and quadratic transfer function vectors. An
important feature of quadratic transfer functions is that they can always be put into a
symmetric fonn which has the following properties
(4.13)
(4.14)
These properties greatly reduce the amount of effort it takes to compute the transfer
functions and also reduce the storage requirements.
In offshore engineering, it is typically assumed that the surface elevation is a zero-
mean, stationary, and gaussian process with a known spectral density Sr\,l1(ro). It follows
that the force £(t) is also a stochastic process whose mean vector and spectral density matrix
can be derived using equations (4.5) to (4.7). Recalling that the autocorrelation function of
the surface elevation, R,,1l('t) =<l1(t>1l(t+'t», is related to the spectral density function by a
Fourier transfonn, the mean force vector, !!t, is given by the equation
00
!!t = <f(t» = f(O) + f ei2)(-cr,cr)Sllll(o) dcr (4.15)
The force spectral density matrix, Sm:ro), is obtained from the Fourier transfonn of
the force correlation matrix by relying on the equation
Sff(ro) = 2~I <f(t)C(t+'t» exp(-iro't) d't (4.16)
which yields
(4.17)
62
with
(4.18)
(4.19)
S (ff2)(",)
UJ = 2 Jco H(2)*(
_r a ,ro-a) H(2)T(
_r a ,ro-a) S1111 ()S
a 1111 (ro-a ) da (4.20)
The delta function, aero), in equation (4.18) indicates that the zero order spectral density is
due to the mean force. The first order term in equation (4.19) is the linear force spectral
density. The second order term in equation (4.20) is the quadratic force spectral density
and involves a frequency domain convolution. Using equations (4.13) and (4.14) the
quadratic spectral density can be rearranged to give
00
+ 2 J Hr (a,ro-a) Hr (a,ro-a) S1111(a)S1111(ro-a) da

(2)* (2)T
(4.21)
o
This form of the quadratic force spectral density lends itself to a physical interpretation.
Specifically, the first expression on the right hand side of equation (4.21) represents the
spectral density of the low frequency quadratic force. This force causes surge, sway and
yaw resonance responses in moored vessels. The second expression represents the
spectral density of the high frequency quadratic force.
In dynamic analyses of moored offshore systems, it is common to consider only the
low frequency force component. As mentioned in Chapter 1, Newman(1974) proposed
approximating this force by using the mean force, fm' due to a regular wave. The
analytical form of the approximation is facilitated by defming a reflection coefficient, Rr< ro),
which is related to the mean force by the equation
(4.22)
63
The symbol H is the wave height of the regular wave with frequency co. Relying on
equation (4.9), it can be shown that a quadratic transfer function is related to the reflection
coefficient by the equation
aj>(co,-co) = 2 Rf(co) (4.23)
Thus, the mean force due to random waves can be computed by rewriting equation (4.15)
as
(4.24)
The basis of the Newman approximation is to extend the exact relation in equation
(4.23) to the approximate equation
(4.25)
Note that this relation reduces to equation (4.23) for COl =~. Further note that the exact
quadratic transfer functions are complex valued. However, since the reflection coefficient
is real valued due to equation (4.22) the approximate transfer function is also real valued.
IT the Newman approximation is accepted, the low frequency quadratic force spectral
density can be rewritten as
(4.26)
Pinkster(1980) evaluated this approximation for the low frequency response of a

tanker and semisubmersible. It was found that its accuracy depends on the frequency range
of the excitation. In this study, the need for approximating the quadratic transfer function
is avoided by computing the full quadratic transfer functions for the moored, vertical
cylinder. The full transfer function will be used to evaluate the validity of the Newman
approximation as it relates to TLP's.
4.3 Near-Field Approach for Deriving Potential Forces

4.3.1 Fluid Flow Boundary Value Problem
The boundary value problem for fluid flow around a moving circular cylinder
considering wave scattering effects is briefly discussed. The resulting linear velocity
64
potential is presented in section 4.4. For more details on its derivation, the interested
reader is referred to MacCamy and Fuchs(1954), Sarpkaya and Isaacson(1981), and Dean
and Dalrymple(1984). The linear and quadratic potential forces are derived from the linear
velocity potential using a perturbation expansion.
The dynamic pressure, p, in a fluid is described by the unsteady Bernoulli equation
ael> 1
P = - Pdt - "21 Vel> 12 - pgz (4.27)
where p is the mass density of water, g is the acceleration of gravity, and el> is the velocity
potential. The existence of a velocity potential implies that the fluid is incompressible and
the flow is irrotational. In addition, the velocity potential, el>(r,9,z,xl,t), must satisfy the
following Laplace equation for fluid flow around the cylinder shown in Figure 4.1
(4.28)
Recalling that the fluid velocity components are given by
(4.29)
the following flow boundary conditions must also be satisfied
atz=-d (4.30)
atr= a (4.31)
where the symbol (') denotes derivative with respect to time. The first condition states that
there is no flow n011llal to the bottom boundary. The second condition ensures that the
velocity n011llal to the cylinder surface is equal to the n011llal velocity of the cylinder
surface.
At the free surface, z = ~, the kinematic boundary condition
z=~ r>a (4.32)
must be satisfied. Also the dynamic boundary condition
z =~, r> a (4.33)

65
which is obtained from Bernoulli's equation must be satisfied. Equation (4.33) defines the
zero pressure condition at the free surface, and equation (4.32) defines the zero flow
condition nonna! to the free surface.
In addition to the boundary conditions in equations (4.30) to (4.33), a radiation
condition for r --+00 must also be considered. Specifically,
(4.34)
where ~ = ~i + ~s+~, ~i is the incident velocity potential, ~s is the scattered velocity

potential, and ~ is the body motion velocity potential. The wave number, K, is obtained
from linear wave theory.
Solving the boundary value problem posed above, yields a solution for the velocity
potential which gives the fluid pressure by the Bernoulli equation in equation (4.27).
Unfortunately, the boundary value problem is nonlinear, and cannot be solved readily
without linearizing. This is done by perfonning a Taylor series expansion of the nonlinear
boundary conditions in equations (4.32) and (4.33) at z=0. Linearization of the latter
equation yields the linear free surface elevation, ~(1) ,
~(1) =- i );1)
CJ atz =0, r> a (4.35)
where ~(1) is the superposition of the linear incident, scattered, and body motion linear
velocity potentials. It can be shown that linearization of equation (4.32) yields the
following dispersion relation
ro2 = gK tanh(Kd) (4.36)
where co is the temporal wave frequency and K is the spatial wave frequency, or wave
number.
The linear free surface elevation elevation also consists of incident, scattered, and
body motion components. From the Airy wave theory, it is known that the incident free
surface, ~~1), is a hannonic wave field in which the spatial and temporal frequencies are
related by the same dispersion relation expressed by equation (4.36). Note that the Airy
wave theory, also called linear wave theory, is simply the linearized solution of the fluid
flow problem with no cylinder in the flow field. Dean and Dalrymple(1984) have an
66
excellent presentation of this solution. The linear incident free surface elevation at the fixed
origin is given following special designation
atx=O y=O (4.37)
It is this surface elevation which is used for input function in the Volterra series in
equations (4.6) and (4.7).
4.3.2 Perturbation Expansion

Once the fluid pressure, p, is known, the two-dimensional force vector for the
horizontal and vertical forces is obtained by the surface integral
f(t) = - If p n dS (4.38)
s
where S is the instantaneously wetted cylinder surface, and n is the outward directed
surface normal at the location (r,9,z). The computation of the moment exerted on the
cylinder is discussed later.
Following Pinkster's approach, the velocity potential <I> is written as an expansion in
a power series in terms of E. Thus,
(4.39)
In addition, the surface elevation ~ is expanded in a similar power series as
(4.40)
The perturbation parameter, E, is sometimes defined as the product of the linear wave
amplitude and the wave number. It is more convenient, however, to assume that E is
dimensionless with a value of unity, and to simply use it for bookkeeping purposes. In
this way, the above expansions resemble Volterra series.
Since the cylinder can displace, the horizontal displacement, Xl' is also expanded in
a perturbation series as
Xl
= (0)
Xl +
(I)
EXI +
2
E Xl
(2)
+ ... (4.41)
67
where x~O) is the mean displacement, and xi) and xi) are linear and quadratic displacements
respectively. The velocity potential and free surface elevation in equations (4.39) and
(4.40) respectively are functions of the horizontal position. Thus, they can be rewritten by
performing Taylor series expansions about the mean position of the cylinder to give
cp ( r,e,z,xl>t) = £cp (1)( r,e,z,xI,t

(0) )
+ £2[ cp(2) ( r,e,z,xI,t

(0) ) a (1)(r,e,z,xI,t
+ xl(1) "'L"""CP (0) )]
+ ... (4.42)
oXI
~ (r,e,xI,t) = £'0,..(1)(r,e,xI,t
(0»)
+ £2[ ~(2) (r,e,x (0)

I ,t) + x(1) a ~(1)(r,e,x(0»)]
I "'L""" l,t + ... (4.43)
oXI
Substituting equation (4.42) into Bernoulli's equation yields the following

perturbation expansion of the pressure
(4.44)
where
p(O) = _pgz (4.45)
acp(1)
p(l) = -PT (4.46)
p(2)
a",(2)
= - PT - PII Vcp(1) 12
1
- pXI F.
(1) a",(1)
(4.47)
The wetted surface of the cylinder, S, consists oftwo parts. The term S(O) is the
mean wetted surface area which extends from the bottom of the cylinder to the still water
level. Similarly, S(1) is the oscillating surface area which extends from the still water level
to the instantaneous free surface. Thus, equation (4.38) can be rewritten as
(4.48)
68
The nonna! vector is not expanded in a perturbation series since the cylinder rotation is
neglected.
Finally, by substituting equations (4.45) to (4.47) into equation (4.48) the force
vector is written in tenns of a perturbation power series as follows
f = f(O) + ef(l) + e 2f(2) + ... (4.49)
For the cylinder in Figure 4.1, the following force components are obtained:
Hydrostatic Buoyancy Force:
(4.50)
Linear Wave Diffraction Force:
a
til) = sU)p -t-n dS . ~x = ~J p a~ a cosO do
(1) 0 2'1t (1)
dz (4.51)
Linear Hydrodynamic Buoyancy Force:
(4.52)
Wave Elevation Drift Force :
a.(1)
£2f:~ = sn {pgz + ep T} DdS· ~y
JJ{pgz + ep at } a coso do dz
y(l)l
e.. 'It l)
=
J-e
2'1t
= 2 ~pg ~(1)2 a coso do (4.53)
Velocity Head Drift Force:
(4.54)
69
Body Motion Drift Force(Horizontal):
(4.55)
Quadratic Potential Drift Force:
o 0
sUP ~DdS. ~X = ~J
(2) 0 21t (2)
~= p ~ acosadadz (4.56)
Body Motion Drift Force(Vertical):
(4.57)
where ~x and ~z are unit vectors parallel to the x and z axes respectively, and the symbol (.)
denotes the dot product. The first subscript on the force denotes the vector component, and
the second subscript is a label to distinguish the forces. The linearized free surface
condition in equation (4.35) is used in the derivation of the wave elevation drift force.
The anticipated hydrostatic buoyancy force is not of particular interest. The linear
hydrodynamic buoyancy force is approximated by using the velocity potential of the
incident wave rather than the total velocity potential. This is because the derived scattered
and body motion velocity potentials are not applicable on the bottom of the cylinder since
no boundary condition has been specified there. The linear wave diffraction force is the
same form used by MacCamy and Fuchs, except that body motion effects are included.
The form of the quadratic order forces agrees with Pinkster(1979) with perhaps the
exception of the body motion drift forces as discussed in Chapter 1. Pinkster considered
the case in which the body could exhibit three translational degrees of freedom. Thus, in
place of the term
(1) 0 ~c!l(1) (4.58)
Xl dt Xl
in equation (4.55), Pinkster's work has the term
~(1) • ~ Vc!l(1) (4.59)

70
where x(1) is the displacement vector and V is a gradient operator denoting derivatives with
respect to the fixed coordinates x, y, and z. Since body motion is permitted only in the x
direction, equation (4.59) reduces to equation (4.58) except that the derivative of $(1) is with
respect to x instead of x 1. Recalling the coordinate relations in equations (4.3) and (4.4),
one can see that a derivative with respect to x is ambiguous. That is because the two
coordinates x and yare over-specified by the three coordinates r, 9, and Xl. Thus to take
derivatives with respect to x, one variable must be held constant. If either r or 9 is held
constant, the derivative ax is equal to aXl which is the present result. However, holding Xl
constant amounts to taking the derivative in the local coordinate system which yields a
different result. The present approach involves taking the derivative in the global
coordinate system and results naturally from the formulation. Spanos and Agarwal
obtained a body motion drift force intuitively from Morison's equation. Comparing
equation (4.55) to their equation reveals that they are the same, which gives credibility to
the present formulation. The present approach also agrees with Lundgren et al.(l972) and
Yoshida et al.(1981).
The equations for the wave elevation and velocity head forces agree with the
equations used by other investigators such as Chakrabarti(1984) and Herfjord and
Nielsen(l986) who considered the ftxed cylinder case, except that the linear velocity
potential, $(1), and free surface, ~(l), are evaluated at a nonzero mean position, xiO), of the
cylinder. The quadratic potential drift force is not considered further in this analysis
because of the previously mentioned difftculties.
The expressions for the f3 components of the force vector, that is the moments, are
the same as the horizontal force components, fl' except that the term (z-Zcg) must be
inserted into the integrands.
4.4 Linear Velocity Potential

The linear diffracted velocity potential acting on a ftxed cylinder has been derived
by others. For relevant references, see MacCamy and Fuchs(l954), Sarpkaya and
Isaacson(l981), and Dean and Da1rymple(l984). The equations are summarized here and
put in a form which is convenient for the derivation of the linear and quadratic force
transfer functions. The equations are slightly modifted to compute the velocity potential at
the mean displaced position rather than the undisplaced position in the ftxed cylinder case.
71
According to linear wave theory, the incident velocity potential due to an incident
wave surface elevation of1'\(t) =~exp(-irot) is
~i
(1)
(x,z,t) *
= H+;.(z,ro) *
Hx(x,ro) H .
'2exp(-lrot) (4.60)
where
H .(z ro) = .&.. i cosh K( d+z) (4.61)
UCIli' ro cosh Kd
Hx(x,ro) = exp(-in) (4.62)
Equation (4.61) is the linear transfer function for the velocity potential at x =O. Equation
(4.62) is the horizontal propagation transfer function for any kinematics from linear theory.
Considering the cylinder to be fixed at the mean displaced position, the incident
wave potential in equation (4.60) is rewritten in terms of polar coordinates by using
equation (4.3) and Bessel function identities in Abramowitz and Stegun(l972) to give
~ll)(r,e,z,x\O),t) = H:.(z,ro) H*(xi),ro) (

't'l X
i
m..o
Pm Jm(Kr) cos me J H2 exp(-irot) (4.63)
where
m=O
(4.64)
m~l
and Jm(.) is a Bessel function of the first kind.

It can be shown that the scattered velocity potential is given by the following
equation
* *
(I)
~s
(0)
(r,e,z,xl ,t) = Ht\>;(z,ro) Hixl ,ro)
(0)
x {- i
m..o
Pm JtA~a) H~(Kr) cos me}
H(1) (Ka)
!!2 exp(-irot) (4.65)
m
where H~(.) is a Hankel function.
Combining equations (4.63) and (4.65) yields the following total velocity potential
for a fixed cylinder
72
cp
(1 ) (0)
(r,a,Z,X} ,t)
*
= H.(r,9,z,x} ,t)
(0) H .
'2exp(-lCot) (4.66)
where
(4.67)
(4.68)
For r=a, the function Am(KI') reduces to the following form
(4.69)
It can be shown that the above linear velocity potential satisfies the Laplace equation
in equation (4.28), the bottom boundary condition in equation (4.30), and the linearized
free surface conditions in equations (4.32) and (4.33). The normal flow boundary
condition in equation (4.31) is satisfied if the cylinder is fixed, in which case x}=O. If the
cylinder is not fixed a body motion velocity potential term is needed to satisfy the normal
flow condition. For a cylinder far from the free surface, the following equation for the
body motion velocity potential can be used
= - x} ra
2
CPm cos 9 (4.70)
This equation satisfies the Laplace equation, the bottom boundary condition, and the
normal flow boundary condition in equation (4.31). However, as implied by the restriction
regarding the free surface, the linear free surface condition in equation (4.32) is not
satisfied. The determination of the body motion velocity potential including free surface
effects is quite involved, see for example Hooft(1982), and is beyond the scope of the
present investigation. It suffices to say that the forces which result from the body motion
velocity potential are known as the added mass force and potential damping force. As is
the convention, these forces are taken over to the left hand side of the equation of motion
and included with the structural mass and damping. In general, the added mass and
damping are frequency dependent. In the next section, it is shown that by neglecting the
73
free surface effects and using equation (4.70) for the body motion velocity potential, a
constant value added mass and zero value potential damping are obtained.
The linear and quadratic right hand side forces in the equation of motion are
computed in sections 4.6 and 4.7. These are obtained from equations (4.51) to (4.57) and
are based only on the linear incident and scattered velocity potential.
4.5 Added Mass Force

The added mass force for the cylinder is derived by substituting equation (4.70)
into equation (4.51). Expanding Xl in a perturbation expansion, the following linear added
mass force, til~, is obtained
,.(1)
I la = - p7ta2dS Xl
.. (1) (4.71)
This is the familiar result stating that the added mass is equal to the mass of the displaced
fluid. A quadratic order added mass also results and is identical to equation (4.71) except
that the superscript (2) is used instead of (1). The added mass force is of no further
concern, except that it should be included in the equation of motion to obtain reliable natural
frequencies.
4.6 Linear Force Transfer Functions

In this section, the linear force transfer functions are derived for the horizontal wave
diffraction force, wave diffraction moment, and vertical dynamic buoyancy.
4.6.1 Wave Diffraction Force

The velocity potential on the cylinder wall is written as a linear transformation of the
incident surface elevation by the equation
= f Hcp(a,e,z,xI
co
(1) (0) (0) - •
cp (a,e,z,xI,t) ,00) 11(00) exp(loot) doo (4.72)
where the transfer function Hcp(a,e,z,xiO),oo) is given by equation (4.67). Taking the time
derivative of this equation and substituting into equation (4.51) yields the following
expression for the linear wave diffraction force
74
DO
f(:)(t) = I H~)(ro) 'ij(ro) exp(irot) dro (4.73)
where
0271:
= ~J i pro H,(a,o,z,xi),ro) a cosO do dz (4.74)
For simplicity, the dependence of the force transfer function on the mean displacement is
not shown explicitly. This will also be true of the remaining force transfer functions which
are derived. Using equation (4.67) to perform the integrations in equation (4.74) leads to
(4.75)
where
t1) = sinhted - sinhte(d-ds)
(4.76)
ZI te coshted
4.6.2 Wave Diffraction Moment

The linear moment transfer function for the cylinder is derived in a similar manner
with
(4.77)
and
1(1) _ d s sinhte(d-ds) coshte(d-ds) - coshted
Zz - te coshted + te2 coshted (4.78)
4.6.3 Hydrodynamic Buoyancy Force

The hydrodynamic buoyancy force is computed from equation (4.52).
Representing the incident wave potential by equation (4.63) and evaluating for z=-dg and
r=O, gives the following transfer function for the vertical dynamic buoyancy force
H~12)(ro) = 2 cosh te(d-d s) H «0) ) (4.79)

I' pg1ta cosh ted x Xl ,ro
75
4.6.4 Comparison to Morison's Equation

A comparison of the linear wave diffraction force in equation (4.73) is made with
the Morison equation inertia force on a cylinder. This comparison is similar to one pursued
by Sarpkaya and Isaacson(1981). Taking the Fourier transform of an assumed harmonic
surface elevation of 11(t) = H/2 cos(rot), substituting into equation (4.73), and rearranging
gives
f (ll)(t) = Cm ('.') p1ta2 g

\AI
sinhKd - sinhlC(d-ds) H
coshKd
( (0) ())
2 cos cot - KXl - '" ro (4.80)
where
(4.81)
(4.82)
Equation (4.80) has the same form of Morison's equation where Cm is the inertia
coefficient and '" is the phase angle. Unlike Morison's equation, however, these variables
have a frequency dependency and are not constant. By using Bessel function identities, the
above equations for Cm and '" can be shown to be identical to those given by Sarpkaya and
Isaacson(1981). In Figure 4.2, Cm and '" are plotted versus the nondimensional
frequency, Ka, to illustrate the frequency dependency. As Ka tends to zero, these variables
converge to the values used in Morison's equation. That is, Cm = 2 and", = - 1t/2. The
general rule of thumb for the frequency range for which Morison's equation is valid is:
lCa < .21t.
4.7 Quadratic Force Transfer Functions

In this section, the full quadratic transfer functions are obtained for the wave
elevation, velocity head, and body motion drift forces and moments.
4.7.1 Wave Elevation Drift Force

The quadratic transfer function for the horizontal wave elevation drift force is
obtained from equation (4.53). First, the linear free surface elevation, ~(1), is written as the
following linear transform of the incident free surface elevation
76
3 .----------------------------------------, 0
o L -____________ ~ _____________ L_ _ _ _ _ _ _ _ _ _ _ _ ~ -~
o 2 3
Figure 4.2 Inertia Force Coefficient and Phase Angle

from Linear Diffraction Theory
77
00
~(1)(r,9,xi),t) = I H~(r,9,xi),ro) ii(ro) exp(irot) dro (4.83)
The linear surface elevation transfer function is obtained from the linear free surface
condition, equation (4.35), and the velocity potential transfer function, equation (4.67). It
is given by the equation
(4.84)
Substituting equation (4.83) with r=a into equation (4.53) yields the following quadratic
transfer function for the wave elevation drift force
(4.85)
where
(4.86)
m=O
(4.87)
m~l
The subscripts on the wave numbers indicate the corresponding frequency.
4.7.2 Wave Elevation Drift Moment

The moment due to the wave elevation drift force is found by inserting (z-Zcg) into
the integrand in equation (4.53). Keeping only terms of quadratic order yields
(2)
f 31(t) = - ZcgIll(t)
"(2)
(4.88)
Therefore,
(4.89)
78
4.7.3 Velocity Head Drift Force

The quadratic transfer function for the horizontal velocity head drift force is
obtained from equation (4.54). For a two dimensional flow around a cylinder, the velocity
head, 1Vcp(l) 12, is given by the equation
(4.90)
where the terms on the right are velocity components in polar coordinates. Using equation
(4.29) and the velocity potential in equation (4.72), the following linear transfer functions
for the velocity components on the cylinder surface are obtained
HUs(a,e,z,xI(0» ,ro ro . cosh K(d+z) H «0) ) ; A ( ) . e

= - tea I cosh Kd x Xl ,ro ~o m Ka sm m (4.91)
(0)
Huz(a,e,z,xI ,ro) = ro i
sinh K(d+z) (0)
sinh Kd Hx<xI ,ro) l-o Am(Ka) cos me
-
(4.92)
(4.93)
Using these linear transfer functions and equation (4.54), the following quadratic
transfer function for the velocity head drift force is obtained
(4.94)
where
(4.95)
(4.96)
sinh(KI+K2)d - sinh(KI+K2)(d-ds)
s+ = 2(KI+K2)sinhK l d sinhK2d (4.97)
79
sinh(lCl-lC2)d - Sinh(lCl-lC2)(d-ds)
s- = 2( lCl-lC2)sinhlCl d sinhlC2d (4.98)
4.7.4 Velocity Head Drift Moment

Similarly, the velocity head moment quadratic transfer function is found to be
(4.99)
where
..(2)
1Z3 = sd+ _ +
+ sd - c - c- (4.100)
+ _ +
= sd
..(2)
17.4 - sd - c + c- (4.101)
+ ds sinh(lCl+ lC2)(d-d s)
sd = 2(lCl+lC2)sinhlCld SinhlC2d (4.102)
d s sinh( lCl-lC2)( d-d s)

sd = 2(lCl-lC2)sinhK 1d sinhK2d (4.103)
COSh(lCl +lC2)d - COSh(lCl +lC2)(d-ds)

c+ = 2(lCl+ lC2)2 sinhlCld sinhlC2d
(4.104)
c- = cosh(lCl-lC2)d - COSh(lCl-lC2)(d-ds)
(4.105)
2(lCl-lC2)2 sinhlCl d sinhlC2d
80
4.7.5 Body Motion Drift Forces and Moment

The quadratic transfer function for the horizontal body motion drift force is
computed from equation (4.55). Taking time and space derivatives of the velocity potential
in equation (4.72) and perfonning the surface integrations, gives the following equation for
the horizontal body motion drift force
(4.106)
In equation (4.106)
00
gl(t) = f H~:(oo) ,,(00) exp(ioot) doo (4.107)
and
(1)( 00) --
H gl . H(1)(
- IIC fl 00
) (4.108)
The linear horizontal response, xil)(t), is written as a linear transform of the surface
elevation by the equation
(4.109)
Here H~:(00) is the horizontal response transfer function. It is the first component of the
response transfer vector, H~)(oo), obtained by the equation
(4.110)
where H(oo) is the linear response transfer function matrix for the moored dynamical
system.
Substituting equations (4.107) and (4.109) into equation (4.106) and rearranging
leads to the body motion quadratic transfer function
(4.111)
81
Note that the transfer function is written in a fonn which maintains the symmetries in
equations (4.13) and (4.14).
In a similar manner, the body motion drift moment and vertical force are derived as
(2)
Hf33(CDl,CD2
)
= 2'1 [0) (I) (I) (1)
HXI(CDl)Hg3(CD2) + H~(CDl)Hxl(CDV
]
(4.112)
(4.113)
where
. H(I)( )
= - IIC f3 CD (4.114)
. H(1)(
= - IIC f2 CD
) (4.115)
4.7.6 Numerical Examples for Fixed Vertical Cylinder

The derived expressions for the quadratic force are compared to existing solutions
for verification. The fixed cylinder example is used since it has been the subject of several
investigations.
A fixed cylinder with dJa=5, d/ds=l, and moments taken about Zcg= -d is
considered first. In Figure 4.3, the horizontal force reflection coefficient from the wave
elevation and velocity head drift forces normalized by pga, is plotted versus the normalized
frequency Ka. These plots agree well with Chakrabarti(1984) and Herfjord and
Nielsen(1986). The wave elevation and velocity head forces have similar magnitudes and
are in opposite directions. Thus, it is very important that both forces be included in an
analysis so that the quadratic force is not grossly overestimated. The reflection coefficient
for the moment normalized by pgad is plotted in Figure 4.4. Herfjord and Nielsen(1986)
do not consider moment. Therefore, results are only compared to Chakrabarti(1984). The
wave elevation drift moment has good agreement with Chakrabarti's results. Note that for
the wave elevation force, the nonnalized moment is the same as the normalized force
according to equation (4.88). The reflection coefficient for the nonnalized velocity head
moment qualitatively agrees with Chakrabarti's result, but is about a factor of two smaller.
It is believed that he must have made a scaling error, however, because his result is
82
,,-~----------------------
r
,,..----
I
ttl
Ol
0-
0
]: "-
cr...:"
' - ...... ,
-1
--- --- --- - - - - -
- - Combined Force
- - - Wave Elevation Force
- - Velocity Head Force
-2
0 2 4 6
lCa
Figure 4_3 Normalized Reflection Coefficient for Horizontal

Potential Drift Force on a Fixed Cylinder
(d/a = 5, d"/d = 1, Zeg = -d,,)
,,-
~ .. ---------------------
.,----
,/-
/
)
"0
ttl
Ol
......
0-
0
]: "-
cr.-.. ' - ...... ,
--- --- --- ----
-1 - - Combined Moment
- - - Wave Elevation Moment
- - Velocity Head Moment
-2
0 2 4 6
lCa
Figure 4.4 Normalized Reflection Coefficient for

Potential Drift Moment on a Fixed Cylinder
(d/a = 5, d"/d = 1, Zeg = -d,,)
83
physically not possible. This is demonstrated by noting that the maximum moment due to
the velocity head drift force has an upper bound equal to the horizontal drift force times the
maximum moment arm, d. This upper bound is obtained if the mean force is concentrated
at the still water line (i.e. z =0), rather than distributed along the submerged depth of the
cylinder. Thus, the normalized moment can be no larger than the normalized force.
Chakrabarti's normalized moment is about twice as large as the normalized force which is
impossible. In the present results, the normalized velocity head moment shown in Figure
4.4 is slightly less than the normalized force. This suggests that the force is concentrated
near the SWL. This feature is reasonable since the force per unit length decreases from the
SWL as the square of an exponential function according to equations (4.53) and (4.67).
In Figure 4.5, the linear and quadratic horizontal force spectral densities for a fixed
cylinder with the following dimensions are shown: d = 340 m, d s = 325 m, a =7.5 m.
The wave spectrum is given by a Pierson-Moskowitz spectrum with significant wave
height, H s ' equal to 9.86 meters and an average period, T avg , equal to 11.51 seconds. The
analytical expression for the Pierson-Moskowitz wave spectrum is given in Appendix C.
The linear, low frequency quadratic, and high frequency quadratic components of the
horizontal force spectrum are plotted separately. The quadratic force includes both the
wave elevation and velocity head drift force and is referred to as the wave elevation/velocity
head force. Quite good agreement between these spectra and the spectra for the same case
illustrated by Herfjord and Nielsen(1986) is seen. It is noted that the quadratic spectra in
this reference are somewhat jagged, while the present spectra are actually quite smooth as
should be expected. The explanation for this is not known. It may be related to how their
spectral densities were computed which is not clearly discussed. Specifically, the quadratic
force is derived as the deterministic result from two harmonic waves, rather than from a
general Volterra series description with a quadratic transfer function. However, overall the
agreement is quite good and enhances the confidence in the reliability of the present results.
A comparison between the low frequency quadratic force spectrum obtained
"exactly" using the full quadratic transfer function and the low frequency quadratic force
spectrum obtained using the Newman approximation is shown in Figure 4.6. The cylinder
dimensions and wave spectrum are the same as in the previous case. Therefore, the exact
spectrum is the same as shown in Figure 4.5. Good agreement for frequencies very close
to zero and frequencies greater than 1.0 is observed, but at other frequencies there is
significant disagreement. The validity of the approximation ultimately depends on the
84
300
250 - - Linear Force

- - - Low Frequency Quadratic Force x 1000
&l
'? 200 - - High Frequency Quadratic Force x 100
'"
~
150
~
en- 100
50
0
0.0 0.5 1.0 1.5 2.0 2.5
(J) rad/sec
Figure 4.5 Unear and Quadratic Potential Force Spectral Densities

for Fixed Cylinder
(a z 7.5m, d,. z 325m, d . 340m, H,." 9.8m, T""II = 11 .51 sec)
0.05 c - - -- - - - - - - - - - - - - - - - - - - ,
0.04 - - - Exact
&l I '" \ - - - Newmans Approximation
'? I \
'"Z~ 0.03 I
I \
\
,
I \
\
I \
I \
I \
0.02
-g
~= ,
I
I \
\
\
en I \
I \
0.01 ~~--- - -~------~,
".: ....... ..
::-:,.......... ........
.... .....
0.0
- .... .:.'":::-::..-:::."'"-.-.,..- - -- - -
0.0 0.5 1.0 1.5 2.0 2.5
(J) rad/sec
Figure 4.6 Low Frequency Quadratic Potential Force Spectral Density

for Fixed Cylinder
(a" 7.5m, d,. .. 325m, d .. 340m, H,. = 9.8m, Ta'll" 11 .51 sec)
85
natural frequency of the system. This will be investigated further in Chapter 6 for a TLP
model.
4.8 Transfer Functions for Tension Leg Platform

4.8.1 Modification of Cylinder Transfer Functions
The linear and quadratic wave force transfer functions derived for the single
cylinder can be used to obtain the transfer functions for the wave force acting on the
columns ofa TLP. In the idealized TLP shown in Figure 1.1, the vertical columns are
located at the comers of a rectangle with the center of gravity of the TLP located at the
geometric center of the rectangle. The horizontal position of a column's centerline in the
fixed coordinate system is XI ±b l , where x I is the horizontal position of the TLP's center of
gravity, and the plus or minus depends on whether it is an up-wave or down-wave column.
Thus, the mean horizontal position of a column centerline is xiO)±b l .
Summing the linear wave forces acting on the mean position of each of the four
columns gives a linear force transfer function vector for the TLP that is equal to the transfer
function vector for a single cylinder multiplied by the factor 4 cos mi'
Similarly for the wave elevation and velocity head drift forces, the quadratic transfer
function vectors for a TLP are obtained from equations (4.85), (4.89), (4.94), and (4.99)
by mUltiplying by 4 cos (lCIH':2)bl.
The quadratic transfer functions for the body motion effects do not need to be
multiplied by any factor if the linear transfer function vector H~)(ro) is computed from the
linear force transfer function vector for the TLP.

An additional moment, not present for a single cylinder arises for the TLP, because
the vertical forces acting on the up-wave and down-wave columns at a given instant of time
are not identical. It can be shown that the linear transfer function for this moment is
obtained from the linear transfer function for the vertical force on a single column, given by
equation (4.79), multiplied by 4b l sin lCb l .
The potential forces acting on the pontoon members are computed using Morison's
equation. This is a valid approximation as the pontoon diameters are usually small. Kirk
and Etok(1979) give a good description of the analytical method to compute these forces.
The quadratic potential wave forces acting on the pontoons are neglected.
86
4.8.2 Numerical Example for Tension Leg Platfonn

The reflection coefficients and force spectra are obtained for a TLP in a water depth
of 600 m and for the dimensional, structural, and hydrodynamic conditions given in Table
4.1. This TLP system has natural frequencies of .057 rad/sec, 3.32 rad/sec, and 3.24
rad/sec for surge, heave, and pitch, respectively. A Pierson-Moskowitz wave spectrum is
used with parameters: Hs= 19.2 m, Tavg= 16.9 sec. The current velocity is assumed to be
zero.
In Figures 4.7 to 4.9, the reflection coefficients for the quadratic surge force, heave
force, and pitch moment are shown respectively. The reflection coefficients due to the
wave elevation, velocity head, and body motion drift force are shown separately. For the
surge force and pitch moment, the wave elevation and velocity head drift force counteract
each other. Both of these forces are an order of magnitude greater than the body motion
drift force. It should be mentioned that the body motion drift force uses the linear surge
response xi> due to the linear potential force and a linearized drag force. Details on this
procedure is given in the following chapter. These results agree qualitatively with
Pinkster's results for a semisubmersible. The heave drift force only consists of body
motion effects which are shown.
The linear force spectral densities for surge, heave, and pitch are shown in Figures
4.10 to 4.12. The low frequency quadratic force spectral densities are shown in Figures
4.13 to 4.15 and the high frequency quadratic force spectral densities are shown in Figures
4.16 to 4.18. The important features to notice are that the quadratic force spectral densities
are of the order of 100 times less than the linear force spectral densities. In the region of
the surge natural frequency, however, the low frequency quadratic force spectral density
dominates. For this particular wave spectrum, the high frequency quadratic force spectral
densities are not close to the heave or pitch resonance frequencies. Therefore, it is
anticipated that no resonance response in these degrees of freedom will occur. However,
for wave spectra which correspond to operational sea states, the high frequency component
of the quadratic force is at higher frequencies and pitch or heave resonances may be
possible. This would be important for analyzing the fatigue life of the tethers.
87
Table 4.1 Particulars for Idealized TLP
Description Symbol Value
7
Structural Mass - 4.0 x 10 kg
Structural Rotational Mass - 8.24 x 1010 kg-rrf

Surge Added Mass - 3.38 x 107 kg
7
Heave Added Mass - 1.41 x 10 kg
Pitch Added Mass - 4.11 x 1010 kg - rrf

8
Surge/Pitch Coupled Mass - -7.73 x 10 kg-m
Column Diameter Dc 16.0 m
Pontoon Diameter Dp 7.0 m
Column Spacing b1 ' b2 45.0 m
Operating draft ds 35.0 m
Elevation of C.G. zqJ 3.0 m
No. of Tendons - 16
Total Pretension - 1.37 x 108 N
Cross Sectional Area Tendon - .1042 rrf

Water Depth d 600.0 m
5
Surge Stiffness k11 2.43 x 10 N/m
8
Heave Stiffness k22 5.98 x 10 N/m
Pitch Stiffness k33 1.21 x 1012 N -m/rad

6
Surge/Pitch Coupled Stiffness k13 -9.24 x 10 N
Drag Coefficient Co 1.0

88
0.6 c----- -- ---------------,
0.4
.,,-
""",- - - --_- -- -_._--
...
'"E / ' / - ---

Z
::e 0.2
/ ,/
/ ---
:§:
II:
-~
0.0
..........
-
/:..............
......
....... _-
--- --- ----
- - Combined Force - ......
.......
·0.2 - - - Wave Elevation Force
......
- - Velocity Head Force
........ Body Motion Force x 10
·0.4
0.0 0.5 1.0 1 .5 2.0 2.5
CJ) rad/sec
Figure 4.7 Reflection Coefficient for Potential Surge Drift Forces

on a Tension Leg Platform
0.015
0.010 ........ Body Mot io n F orce
'"......E
z
::e
0.005
:§:
~ . 0.0
·0.005 '--_ _ _-'-_ _ _ _..I..-_ __ - - '_ _ __ ....L.._ _ _- - - '
0.0 0.5 1 .0 1.5 2 .0 2 .5

CJ) rad/sec
Figure 4.8 Reflection Coefficient for Potential Heave Drift Force

89
1.5 - - Comblned Moment

- - - Wave Elevation Moment
-------
-_ .....
- - Velocity Head Moment
1.0
-
........ Body Motion Moment x 10 ,/ .....
E ,/
/
z 0.5 /
::!
--:::-..
/
/
0.0 ..
..... ,! ........ -
g ~-~~-~~
;'-~
:' ~,
~
.
ce'" .0.5
.... . ... ~: \ ~\ f -
\ '~
\../ \,
· 1.0 '-- .. ---........
· 1.5 " -_ _ _-'-_ _ _ _..L.._ _ _- - '_ _ _ _- ' -_ _ _----' ------------------
0.0 0.5 1.0 1.5 2.0 2.5
CJ) rad/sec
Figure 4.9 Reflection Coefficient for Potential Pitch Drift Moments

3000
2500
~
'"Z 2000
::!
1500
-gen
:c .
_-:
1000
\J\
500
0 I-J
0.0 0.5 1.0 1.5 2.0 2.5
(j) rad/sec
Figure 4.10 Spectral Density of Unear Potential Surge Force

, on a Tension Leg Platform
(He .. 19.2m, T""" = 16.9sec)
90
3000
2500
&!
'!'
'"Z 2000
~
1500
g
- ..S"
:c ..S"
(/) 1000
500
0 -.J
0.0 0.5 1.0 1.5 2.0 2.5
(Jl rad/sec
Figure 4.11 Spectral Density of Linear Potential Heave Force

(H" = 19.2 m, Tavg = 16.9 sec)
120000
100000
~
"'~
i 80000
Z
~
60000
"S' 40000
E::=
(/)
20000
0
0.0 0.5 1.0 1.5 2.0 2.5
(Jl rad/sec
Figure 4.12 Spectral Density of Linear Potential Pitch Moment

on a TenSion Leg Platform
(H" = 19.2m, TaYg = 16.9sec)
91
30
25 - - Combined Force
- - Wave ElevNe!. Head Force
N 20 ..•...•. Body Motion Force
~
15
0.0 0.5 1.0 1.5 2.0 2.5

CJ) rad/sec
Figure 4.13 S~ectral Densities of Low Frequency Quadratic

Potential Surge Forces on a Tension Leg Platform
(H" • 192m. T""9 = 16.9sec)
2.0
1.5 .•.•• ... Body Motion
1.0
0.5
0.0 '~ . • -"- ". ~ . " . . . ...... ... .. . . . .. . ··· • ·· ·•·••• ••· .. .... t .... - -_...... • .. • .. · .. • · • .. , ···· · · · · · ··· · ···_·
0.0 0.5 1.0 1.5 2.0 2.5

CJ) rad/sec
Figure 4.14 Spectral Density of Low Frequency Quadratic

Potential Heave Force on a Tension Leg Platform
(H" = 192 m. T8"'<1 = 16.9 sec)
92
2000
.'.
: ~.
; l
&l ; !
~ 1500 f t - - Combined Moment
N
! :
- - Wave Elev.Nel. Head Moment
~ ! 1
: ! ..••.... Body Motion Moment
z ! \
:::E
1000 : !
i \
! ~
g i l
-~..J"
..J"
(J) 500 f \
o
0.0 0.5 1.0 1.5 2.0 2.5
w rad/sec
Figure 4.15 Spectral Densities of Low Frequency Quadratic
Potential Pitch Moments on a Tension Leg Platform
(H,." 19.2m. Ta"1l = 16.9sec)
60
- - Combined Force
r
50 I \ - - Wave Elev.Nel. Head Force
I \ ....... Body Motion Force
~ I \
40
N
I \
Z I \
::!:
I \
30
I \
I \
g 20
I \
Ii -~
- (J).r
\
\
\
10
0
0.0 0 .5 1.0 1.5 2.0 2.5
w rad/sec
Figure 4.16 Sj:>ectral Densities of High Frequency Quadratic
Potential Surge Forces on a Tension Leg Platform
(H,." 19.2m. Ts"1l = 16.9sec)
93
2.0 c-------------------,
1.5 •....... Body Motion Force
..z
~
::::i!
1.0
'$" l\"
.'
J\
~-~
II)
0.5
0.0 . . .••........• / ··••·· ...... ••...... ...... . .. T =..................

................
0.0 0 .5 1.0 1.5 2 .0 2 .5

ro radlsec
Figure 4.17 Spectral Density for High Frequency Quadratic

Potential Heave Force on a Tension leg Platform
(H" = 19.2m, T""'9. 16.9sec)
2000 c----------------------------------------,
( \ - - Combined Moment
:
! \ - - Wave Elev.Ne!. Head Moment
~
1500 j \ ........ Body Motion Moment
j \
1000 :f/~
:
, \\:\:
j \\
I \
i \
500
f \\
f \\\ .........
\.- \.
o
0.0 0.5 1.0 1.5 2.0 2.5
ro radlsec
Figure 4.18 Spectral Densities of High Frequency Quadratic

Potential Pitch Moments on a Tension Leg Platform
(H" • 192m, Tavg. 16.9sec)
94

Linear and quadratic potential forces acting on a vertical, moored, and surface-
piercing cylinder have been derived. The derived forces are the linear potential force, the
quadratic wave elevation force, the quadratic velocity head force, and the quadratic body
motion force. The forces are described in the fonn of second order Volterra series with the
linear wave elevation as the input function. The linear and quadratic transfer functions
which defme the Volterra series are derived based on linear diffraction theory. The spectral
densities of the second order Volterra series forces are obtained as the sum of linear and
quadratic spectral densities. The linear spectral densities have the same frequency range as
that of the wave elevation spectrum. The quadratic spectral densities are detennined from a
frequency domain convolution of the wave elevation spectrum. This leads to low
frequency and high frequency spectral components. It is anticipated that the low and high
frequency quadratic spectra may significantly affect die response of a compliant offshore
platfonn. The resulting expressions are verified against existing solutions as available.
It is also shown how the derived Volterra series for the potential forces acting on a
single vertical cylinder can be modified to give the Volterra series for the potential forces
acting on a 1LP consisting of four vertical columns. Pertinent numerical results are given.
CHAPTER 5
EQUIV ALENT STOCHASTIC QUADRATIZATION FOR
TENSION LEG PLATFORM RESPONSE TO VISCOUS DRIFT FORCES
5.1 Introduction
The equivalent stochastic quadratization method is used in this chapter to analyze
the response of a TLP to wave and current induced forces. The method is applied on an
idealized TLP model with three degrees of freedom. To clearly evaluate the reliability of
the equivalent stochastic quadratization method, only higher order wave forces due to
viscous effects are considered in this chapter. In Chapter 6, the response of the TLP
subject to both potential and viscous higher order forces is obtained.
5.2 Formulation of TLP Model

A diagram of the two dimensional TLP model used in this analysis is shown in
Figure 1.1. The fixed coordinate system, (O-x-z), has its origin located at the still water
line (SWL). The distances d and ds are the water depth and the TLP submerged depth in
still water, respectively. The term Zcg is the elevation of the center of gravity of the TLP
hull and topside equipment. The distances 2b l and 2b2 are the center-line to center-line
distances between the columns.
The wave elevation at the origin, l1(t), is a is zero-mean, stationary, and gaussian
random process. Its second order statistics are defmed by a two-sided spectral density
function Sl1l1(ro). The fluid kinematics is based on linear wave theory and hence can be
studied by linear transformations of the wave elevation. In addition, it is assumed that the
waves propagate to the right. The current velocity, U(z), is an arbitrary function of depth
and could be in the same or in an opposite direction to the wave propagation.
The vector X represents the in-plane displacements of the center of gravity relative to
the fixed system (O-x-z). The symbols Xl' x2' and x3 represent surge, heave, and pitch,
respectively. It is assumed that the mass of the tendons and the hydrodynamic forces
acting on them have a negligible effect on the motion of the hull. Thus, the resulting
analytical model is a rigid body with three degrees of freedom. The equation of motion for
the three degree-of-freedom system is
96
(5.1)
where m, C, and k are the 3x3 mass, damping, and stiffness matrices, respectively. The
forces fp(t) and fo(x,t) are the vectors for the total linear potential force and total nonlinear
viscous force acting on the hull, respectively. For simplification, the hydrodynamic forces
are computed from the undisplaced position of the TLP.
The mass and damping matrices are composed of structural components and
hydrodynamic components due to potential effects. In reality, the hydrodynamic
components are frequency dependent. However, as discussed in Chapter 4, the assumed
velocity potential yields only constant valued hydrodynamic added mass and no
hydrodynamic damping. This is suitable for the present purposes, but a more precise
description of these terms could be incorporated into the present procedure. The restoring
forces are provided by the tendon stiffness and hull buoyancy and are assumed to be linear.
The linear potential wave force, fp(t), is often referred to as the inertia force. As
shown in Chapter 4, this force can be viewed as the output of a linear system whose input
function is the wave elevation. For a time invariant system, this relationship is expressed
in the form
(5.2)
where bi't) is the impulse response function vector of the linear time invariant system. The
inertia force transfer function vector, l:M(i), is the frequency domain counterpart of hp('t) to
which it is mathematically related by a Fourier transform.
The total viscous force, fo(x,t), is also referred to as the drag force. It is computed
analytically by the Morison equation drag force with relative velocity effects. In this study,
only the drag force resulting from horizontal motion is accounted for. The total drag force
is computed by integrating the Morison equation drag force per unit length over the mean
wetted length of each member. This yields
Nm
fo(x,t) = L[ f c..on(z)1 U(z) + v(x,z,t) I (U(z) + v(x,z,t») dLn] (5.3)
n=l Ln
where
v(X,z,t) = u(x,z,t) - xl(t) - (z-Zcg)X3(t) "'" u(x,z,t) - xl(t) (5.4)

97
Here Nm is the nwnber of submerged structural members and Ln is the mean wetted length
of the nth member. The vector ~(z) contains the Morison drag parameters for the nth
member. It is equal to the vector of the drag force acting on a unit length of that member
due to a unit relative velocity.
The velocity quantity U(z} + v(x,z,t} is the relative horizontal water particle velocity
at the position (x-z). The random part of the relative velocity, v(x,z,t}, is due to both wave
motion and rigid body motion. The horizontal velocity at the location (x-z) due to rigid
body motion is approximated by the horizontal surge velocity of the hull, xI(l). This is a
good approximation because the horizontal motion caused by the pitch rotation is very
small. The horizontal water particle velocity due to wave motion, u(x,z,t), is based on
linear wave theory. Like the inertia force, it can be expressed as linear transform of the
wave elevation by the following equation
u(x,z,t) = f hu(x,z,t) 11(t-t) dt (5.5)
where hu(x,z,t) is the impulse response function for u(x,z,t}. According to linear wave
theory, the corresponding linear transfer function, Hix,z,O), is given by the equation
cosh lC(d+z) .
Hu(x,z,O)} = 0) sinh lCd exp(-In) (5.6)
where the wave nwnber, lC, is related to the frequency, 0), by the linear dispersion
equation.
To illustrate the computation of the drag force, the drag force acting on one of the
downwave columns of the TLP is considered. Relying on equation (5.3), the drag force
acting on this member is computed by
where
(5.8)
98
In equation (5.8), Dc is the diameter of the column, p is the mass density of water, and Co
is the Morison drag force coefficient Note that there is no drag force contribution in the
heave direction because only the drag force due to horizontal motion is considered.

The equivalent stochastic quadratization method of Chapters 2 and 3, is used to
obtain the response statistics of the nonlinear system given by equation (5.1). Due to the
fact that the drag force nonlinearity is non-symmetric, the stationary response of the system
in equation (5.1) is written as
x(t) = Ux + x(t) (5.9)
where Ux is the mean displacement vector, and x(t) is a vector of stationary, zero-mean
processes, which in general are not gaussian. Substituting equation (5.9) into equation
(5.1) gives
(5.10)
where the drag force reflects the approximation made in equation (5.4). Ensemble
averaging equation (5.10) yields
Ux = k-l<io(kl,t» (5.11)
where <.> denotes mathematical expectation.

The response, x(t), is obtained by solving the response of an "equiValent" quadratic
system. The equivalent system is also nonlinear, but can be solved approximately by the
Volterra series method. The form of the equivalent system is constructed as follows
(5.12)
where
Nm
to(kl,t) =~[ J Qon(z){ 2CXl (x,z)U(z)v(x,z,t)

n=l Ln
+ CX2(X,z)[ v 2(x,z,t) - <v2(x,z,t»]} dLn] (5.13)
The <lj(x,z)'s are nondimensional quadratization coefficients and
v(x,z,t) = u(x,z,t) - il(t) (5.14)

99
For the special case when ~(x,zF-O, the method reduces to the equivalent stochastic
linearization method.
The error vector is defmed by the equation
£ == k u.x - f o(i1,t) + t o(i1,t) (5.1S)
Further, combining equations (S.3), (S.11), and (S.13), equation (S.lS) can be rewritten as
Nm
£ =L [ J Qon { <IU+vl(U+v» - IU+vl(U+v)

n=1 Ln
(S.16)
where the dependence on x, z, and t is understood.

The quadratization coefficients are found by minimizing £ in a mean square sense.
The simplest approach is to minimize the mean square of the integrand in equation (S.16) at
(x,z) positions that are spaced to allow numerical integration of the force in equation
(S.13). Based on this type of minimization procedure, the analytical expressions for the
quadratization coefficients are
(S.17)
(S.18)
where
~m = <vm> (S.19)
and
Ek = < IU+vl(U+v)vk > k= 0,1,2 (S.20)
The statistics of the relative velocity vare obtained from the response of the
nonlinear equivalent system and are used to evaluate the expectations in equations (S.19)
and (S.20). The Volterra series method is used to approximate the response of this system.
For this, equation (S.12) is rearranged. The linear surge velocity term in to is taken over to
the left hand side of equation (S.12) to form a viscous hydrodynamic damping. The
to
remaining linear and quadratic terms in are kept on the right hand side of the equation
100
and are considered to be unknown forces. The scalar coefficient A is introduced for
bookkeeping purposes and is set equal to one. Without altering the equivalent system, the
inertia force vector, fp(t), is replaced by Afp(t) and the water particle velocity, u(x,z,t), is
replaced by A.u(x,z,t). The nonlinear equivalent system in equation (5.12) is thus rewritten
as
where
Nm
= ~ [ I Qon(z)2al(x,z)U(z)~] eI (5.22)
n=l Ln
Nm
tg)(t) =~[ I QDn(z)2al (x,z)U(z)u(x,z,t) dLn] (5.23)
n=l Ln
Nm
t~)(il,t) = ~[ I QDn(z)a2(x,z)(u(x,z,t) - il(t )A- 1)2 ~] (5.24)
n=l Ln
The matrix Ceq is the viscous hydrodynamic damping matrix, the vector el is the first
column of the identity matrix, and the superscript T denotes transposition.
Implementing the Volterra series method, the solution of equation (5.21) is written
in the form
00
i(t) = ~ Aji(j)(t) (5.25)

j=l
Substituting equation (5.25) into equation (5.21) and equating like powers of Aup to
quadratic order leads to
(5.26)
(5.27)
Due to linearity, equations (5.26) and (5.27) can be combined into one equation, which
gives the following approximation for the nonlinear equivalent system
101
m i + (c + Ceq)! + kg = let) = l(O) + l(l)(t) + F(t) (5.28)
where
(5.29)
(5.30)
(5.31)
Recalling that fp(t) and u(x,z,t) are linear transforms of the wave elevation, the
force i(l)(t) must be a linear transform of the wave elevation, and the force ~)(t) must be a
quadratic transform of the wave elevation. Thus, the force i(t) can be expressed as the
following second order Volterra series
The vectors ~l)(t) and hf)(tl>t2) are respectively linear and quadratic Volterra kernels for
force and have the corresponding linear and quadratic response transfer function vectors,
Ul1)(oo) and uf)(OOl,002). The forces i(1)(t) and £(2)(t) are referred to as the linear and
quadratic forces. In this chapter, the latter force is also referred to as the viscous drift
force.
Since the differential operator in equation (5.28) is linear, the steady-state displacement
response can also be written in a Volterra series form. Specifically,
where b~)(t) and b~)(tl,t~ are the linear and quadratic Volterra kernel vectors for
displacement, respectively. The associated transfer function vectors are ui)(oo) and
U~)(OOl'~. It follows from equations (5.29) and (5.31) that the zero order response
102
vector, g(O), is the negative of the expected value of g(2). Therefore, <get»~ =Q as
formulated.

All numerical computations are carried out in the frequency domain. Thus, only
transfer functions and not the kernels themselves need to be computed. The linear force
transfer function vector m1)(ro) is obtained by using the linear Fourier transform relation
and equations (5.2), (5.5), (5.23), and (5.30). That is,

Nm
Hl)(ro) = Hp(ro) + L [ J QDn(z)2cx l(x,z)U(z)H (x,z,ro)
u dLn] (5.34)
n=l Ln
Similarly, the quadratic force transfer function is obtained by using the linear and
quadratic Fourier transform relations and equations (5.5), (5.24), and (5.31). This
procedure gives
(5.35)
where Hk~ (ro) is the linear surge vel?city transfer function.
The derivation of the displacement and velocity response transfer function vectors
from Ht(ro), Ht2)(rol,ro2)' and the linear response transfer function matrix, H(ro), for the
linear differential operator in equation (5.28) follows the procedure described in Chapter 3
and is not be repeated here. One difference to note, however, is that the linear response
transfer function matrix is now computed by relying on the equation
(5.36)
It is noted that the force quadratic transfer function vector in equation (5.35)
possesses the previously discussed symmetry conditions. It follows that the response
quadratic transfer function vectors have the same properties.
103
5.5 Response Probability Distribution

The probability distribution of the response given by equation (S.33) is, in general,
not gaussian because it involves a quadratic transfonnation of a gaussian process. The
Gram-Charlier expansion, which is an expansion of derivatives of a gaussian distribution,
is utilized to approximate the probability distribution. Truncating the expansion after the
third derivative, the probability distribution for the displacement response of the ith degree
of freedom is
i = 1,2,3 (5.37)
where cIl(xi) is a gaussian distribution defmed by the equation
(5.38)
Further, ~i'"1 is the m th central moment of Xi which is expressed mathematically as
~iT = <xf> (S.39)
For the nonlinear equivalent system, these moments are estimated using equation
(5.33). For instance, the second moment of the ith degree of freedom is obtained by
substituting the i th vector component of equation (5.33) into equation (S.39) for m=2 and
taking the expected value. Recalling the multiplicative properties of gaussian variables it
can be shown that
(5.40)
where ~'I\('t) is the autocorrelation function of the wave elevation process which is defmed
by
(S.41)
104
Equation (5.40) is converted into the frequency domain by relying on the well
known relation
~T\(t) = J ST\T\(O» exp(io>t) do> (5.42)
Substituting equation (5.42) into equation (5.40) and using the symmetry relations
of the quadratic transfer functions gives
(5.43)
Proceeding in a similar manner it can be shown that .
(5.44)
Note that according to equation (5.44), the probability distribution in equation (5.37)
reduces to a gaussian distribution if the response quadratic transfer function is zero. Thus,
the distribution obtained by equivalent stochastic linearization is gaussian.
In addition to computing the displacement response statistics, the statistics for
v(x,z,t) must be computed to calculate the quadratization coefficients using equations
(5.17) to (5.20). The non-gaussian probability distribution and central moments of v(x,z,t)
can be computed by equations analogous to equations (5.37), (5.43), and (5.44). The
Volterra series forv(x,z,t) is obtained from equations (5.4) and (5.5) as
= 1[hu(x,z,t) - h~l(t)] l1(t-t) dt

00 (I)
v(x,z,t)
(2)
- JJh~1(tl,t2) 11(t-tl)11(t-t2) dtl dt2
00
(5.45)
_00
105
The expectation given in equation (5.20) is detennined using the non-gaussian probability
distribution for v(x,z,t) and the equations in Appendix B.

In addition to the probability distribution of the response, it is also often of interest
to know the power spectral density of the response. By definition, the response spectral
density matrix, Su(ro), is expressed by the equation
Sii(ro) = 2~
-
j <g(t)gT(t+'t» exp(-iro't) d't
From linear system theory, the response spectral density matrix is computed by the
(5.46)
following well known fonnula

(5.47)
where Sli<ro) is the spectral density matrix of the force vector £(t). The force spectral
density matrix is obtained by substituting equation (5.32) into the analogous fonn of
equation (5.46) for force. Specifically,
=
(1)* (\)T
Sn(ro) Bl (ro)lij (ro)Sl1l1(ro)
The force spectral density matrix consists of two parts. The first part is the spectral
density of the linear force, £(I)(t). The second part is the spectral density of the quadratic
force, £(2)(t), and involves a frequency domain convolution.
5.7 Axial Tendon Force

For the idealized TI..P shown in Figure 1.1, the fluctuation of tendon tension, ft(t),
over the pretension force depends on the heave and pitch responses, and is computed by
the following equation
(5.49)
106
where k t is the tendon axial stiffness. The plus or minus sign refers to the downstream or
upstream tendons, respectively.
The statistics for the axial tendon force are computed from the joint statistics of the
heave and pitch response. For example, the spectral density of the zero-mean axial tendon
force, tt(t), is found to be
'~f'
It t
(00) = "tJr2 " (00) + b12SX3X3
{ S.X2A2 .. (00) ± 2h'1 Re[S"A2A3
" (oo)]} (5.50)

For the implementation of the preceding solution scheme, an iterative procedure is
required as the quadratization coefficients are initially unknown. Zero-valued coefficients
can be used for initial guesses. In fact those guesses 1!ave been found to lead to convergent
solutions in 4 to 6 iterations. The integrations implicit in equation (5.11) and explicit in
equations (5.22), (5.34), and (5.35) are performed numerically using the trapezoidal rule.
To make the iteration process efficient, only the moments of the relative water
velocity, v(x,z,t), are computed to calculate the quadratization coefficients. The
displacement response and axial tendon force statistics can be computed after convergence.
5.9 Numerical Example

Numerical results are presented for a TLP model in a water depth of 600 meters.
The TLP parameters used are given in Table 4.1. For this idealized model, the surge and
pitch degrees of freedom are coupled in the mass and stiffness matrices. The values of the
natural frequencies, however, are not dramatically different from the values corresponding
to a system which is not coupled. That is, the coupling is small. Further, the natural
frequencies computed for surge and pitch are respectively .057 and 3.24 rad/sec. The
heave degree of freedom has a natural frequency of 3.32 rad/sec and is not coupled with the
other degrees of freedom.
The damping matrix, c, is computed by converting an assumed diagonal damping
matrix for the modal degrees of freedom into a coupled damping matrix for the physical
degrees of freedom. For the diagonal modal damping matrix, damping ratios of .05, .01,
and .01 respectively are used for the modal degrees of freedom which correspond to surge,
heave, and pitch. A higher value of damping in the surge mode is used to account for
107
potential hydrodynamic damping effects. The hydrodynamic added mass is computed by

assuming that the inertia coefficient is equal to 2.
The wave kinematics is described by the Pierson-Moskowitz spectrum with wind
speed of 30 rn/sec which yields a significant wave height of 19.2 meters and an average
period of 16.9 sec. The current velocity profile is assumed to be linear and in the direction
of wave propagation. In all cases, the ratio of the current velocity at the still water line to
the current velocity at the elevation z =-cis is equal to 2.
Figure 5.1 shows the one-sided form of the power spectral density of the zero-
mean part of the surge response, Xl' obtained by equivalent stochastic quadratization,
equivalent stochastic linearization, and simulation for a surface current velocity of 1.0
rn/sec. The simulation was performed by using the sum of harmonics method to produce
100 realizations of the wave kinematics. Details of the simulation procedure are given in
Appendix D. The high frequency peak is the wave frequency response and is due to the
linear force. The low frequency peak is a resonance response at the surge natural
frequency. This is the slowly-varying drift response and results from the quadratic force.
It can be seen that the spectra obtained by simulation and quadratization are in good
agreement for both peaks. The quadratization method, however, is computationally much
more efficient, to the extent that it reduces the requisite computation time by two orders of
magnitude. The spectrum obtained by the linearization method is nearly identical to the
spectrum obtained by quadratization in the wave frequency response region. However, the
linearization spectrum is negligible in the low frequency region, since quadratic effects are
not accounted for.
The probability distribution of the surge response is shown in Figure 5.2. The
distribution obtained by simulation and quadratization are in good agreement. The vertical
lines mark the computed mean responses and clearly indicate that the simulated and
quadratization responses are skewed. The linearization response is gaussian.
In Figure 5.4, the one-sided form of the power spectral density of the zero-mean
pitch response, x3' is shown. This response is due predominantly to linear forces. The
coupling with the surge response causes a small quadratic component at the surge natural
frequency, but it is almost negligible. Thus, the probability distribution of the pitch
response is nearly gaussian.
The heave response and axial tendon force spectra are shown in Figures 5.3 and
5.5, respectively. The axial tendon force is for one of the downstream tendons. Since
108
80
60
&l
CI)
- - Equiv. Quad.
N
E - - Equiv. lin.
........ Simulation
40
~
o{
(M
CIl
20
0.2 0.4 0 .6 0.8 1.0

(0) rad/sec
Figure 5.1 Surge Response Power Spectral Density

(Ha = 19.2 m, Tayg . 16.9 sec, U(z) = [1 + .5 z/dsl m/sec)
0.16
0.14
/1,\
0.12 - - Equiv. Quad. /
.J \
\
- - Equiv. lin.
\
1:: 0.10
........ Simulation \
\
.. .\
0.08
..~
a: 0.06
0.04
0.02
0.0
-10 ·5 o 10 15 20
Figure 5.2 Surge Response Probability Distribution

(H. = 19.2 m, Tayg = 16.9 sec, U(z) = [1 + .5 z/dsl mIseC)
109
&l
til 6
N
- - Equiv. Quad.
E
-- Equiv. Un.
.., 4
....••.. Simulation
51
x
]:
'''""
(I)'" 2
OC===~~____-L~==~C=====~====~
0.0 0.2 0.4 0 .6 0.8 1.0
CJ) fad/sec
Figure 5.3 Heave Response Power Spectral Density

( H" = 19.2 m, Tavo z 16.9 sec, U(z),. [1 + .5 zldsl mlsec )
6
~ - - Equiv. Quad.
N
"0 5
~ - - Equiv. Un.
•...••.. Simulation
4
CD
0
x 3
8"
-;~ 2
o'
(I)
0
0.0 0.2 0.4 0.6 0.8 1.0
CJ) rad/sec
Figure 5.4 Pitch Response Power Spectral Density

(H. = 19.2 m, Tavo. 16.9 sec, U(z). [1 + .5 zId.J mlsec)
110
10
8
g - - Equiv. Quad.
III
N
6 - - Equiv. Un.
Z
::E •.. •.•. . Simulation
4
~<--
<.:-
(/)
0
0.0 0.2 0.4 0 .6 0.8 1.0
(!) rad/sec
Figure 5.5 Axial Tendon Force Power Spectral Density

(H. = 19.2 m , Tayo = 16.9 sec, U(z) = [1 + .5 z/dJ mlsec)
E
4
- - Equlv. Quad.
2
- - Equiv. Un.
• Sm
i ulation
O~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ____ ~
0.0 0.2 0.4 0.6 0 .8 1.0 1.2

U(O) m/sec
Figure 5.6 Mean Surge Response vs. Current Velocity

(Ha • 19.2 m,Tayg = 16.9 sec, U(z). U(O) [1 + .5 zldJ )
111
there are no quadratic viscous forces in the heave direction and the heave displacement is
not coupled with the other displacements, only wave frequency responses are present in the
heave response spectral density. Therefore, linearization, quadratization, and simulation
give virtually identical results. It can be seen that the axial tendon force is primarily due to
the heave displacement. This indicates, at least for this sea state, that the tendon tension is
not significantly affected by quadratic forces. The probability distributions for the heave
and axial tendon forces are not shown since they are gaussian.
In Figures 5.6, 5.7, and 5.8, the mean, standard deviation, and skewness of the
surge response are plotted versus the surface current velocity. In Figure 5.9, the mean
pitch response is plotted versus the surface current velocity. The equivalent stochastic
quadratization method results is in good agreement with the simulation results for the mean
surge and pitch responses. The surge standard deviation obtained by quadratization
slightly underestimates the simulated results for higher current velocities, but still is quite
reliable. This is especially true in comparison with linearization, which significantly
underestimates the standard deviation, because no quadratic forces are accounted for. The
surge skewness computed by quadratization has the same trend but overpredicts the
skewness computed by simulation. Yet, this is better than linearization, which yields
skewness equal to zero.
In Figure 5.10, the 1-1 element of the matrix Ceq normalized by the factor .5Nkllmll
is plotted versus surface current velocity. For a system with no coupling, this represents a
surge viscous hydrodynamic damping ratio. For the present model, which has slight
coupling, it has a similar effect on the system. Compared with typical damping ratios for
structural damping, the hydrodynamic damping is much higher. This has a direct bearing
on the magnitude of the quadratic response because it is a resonance response. This is also
an important effect which should be included when analyzing the response due to potential
drift forces.
Figures 5.6 and 5.7 also illustrate the relation of the current velocity to the
symmetry of the nonlinear drag force and its effect on the quadratic force. For zero current
velocity, the drag force nonlinearity given by equation (5.3) is perfectly symmetric. Thus,
the equivalent quadratic system has only linear terms. Consequently, the response statistics
are identical to those determined by equivalent stochastic linearization. For nonzero current
velocity, the nonlinear drag force is non-symmetric, and induces mean and quadratic
responses. These responses become larger with increasing current velocity.
112
w
..
E
3
- - Equiv. Quad.
- - Equiv. Lin.
• Sm
i u lation
0.0 0.2 ~4 ~6 ~8 1.0 1.2

U(O) mlsec
Figure 5.7 Surge Response Standard Deviation vs. Current Velocity

(H" = 19.2m, Tao.og" 16.9 sec:, U(z) = UtJ)[1 + .5z/d,.)
12
10 - - Equiv. Quad .
- - Equiv. Lin.
8 • Simulation
'"E •
6
::J.M •
4
•
2
0
0.0 0 .2 0.4 0.6 0.8 1.0 1.2
U(O) mlsec
Fig. 5.8 Surge Response Skewness vs. Current Velocity

(H" = 19.2 m, Tao.og • 16.9 sec:, U(z). UtJ)[1 +.5z/d,.)
113
3.0
2.5
i 2.0
'"~ 1.5
x
=1:" 1.0
- - Equiv. Quad.
0.5 - - Equiv. lin.

• Simulation
0.0
0.0 0.2 0.4 0 .6 0 .8 1 .0 1.2
U(O) m/S9C
Figure 5.9 Mean Pitch Response vs. Current Velocity

(H,.z 19.2m, TII\'g. 16.9 sec, U(z) - U(l)[1 + .5z/d,,)
0.5
0.4
~
-j
0.3
S 0.2 - - Equiv. Quad.

III
- - Equiv. Un.
0.1
0.0
0.0 0.2 0.4 0.6 0 .8 1 .0 1 .2
U(O) m/sec
Figure 5.10 Surge Hydrodynamic Damping vs. Current Velocity

(H,. • 19.2m, T.vg. 16.9 sec , U(z) - U(l)[1 + 5z/d,,)
114
S.10 Summary and Conclusions

A stochastic analysis of a three-degree-of-freedom model of a Tension Leg Platfonn
(TLP) is perfonned. The drag force is modeled by the generalized nonlinear Morison
equation. Applying the equivalent stochastic quadratization method, the nonlinearity is
replaced by a polynomial nonlinearity up to quadratic order. The coefficients of the
polynomials are detennined by using a mean square minimization. The resulting
"equivalent" nonlinear system has a fonn Whose solution can be approximated by the
Volterra series method. A third order Gram-Charlier expansion is used to describe the non-
gaussian response probability distribution. An iterative method is used to compute the
unknown quadratization coefficients.
Numerical results are obtained for an idealized TLP in a water depth of 600 meters.
The results are compared to simulation to verify their reliability. The solution obtained by
the proposed method provides a notable improvement over the solution obtained by the
equivalent stochastic linearization method. The nonlinear forces primarily affect the surge
response. In addition to the wave frequency response and mean response, the
quadratization solution yields a significant slowly-varying drift response at the surge
natural frequency of the TLP. The probability distribution of the surge response is non-
gaussian. The solution is obtained quite efficiently and compares well with time domain
simulation data. The magnitudes of the mean and slowly-varying drift response increase
with increasing current velocity. The linearization solution does not account for the slowly-
varying response and is gaussian. .
CHAPTER 6
STOCHASTIC RESPONSE OF A TENSION LEG PLATFORM
TO VISCOUS AND POTENTIAL DRIFT FORCES
6.1 Introduction
In this chapter, the dynamic response of a TI.P subject to viscous and potential fIrst
and second order forces is obtained utilizing the equivalent stochastic quadratization
method. The analytical procedure is similar to the procedure adopted in the preceding
chapter except that the potential force here includes the quadratic potential forces which
have been derived in Chapter 4.
Numerical results are presented for an idealized TI.P in various wave states. In
addition to obtaining responses for the combined quadratic forces, responses due to
individual quadratic forces have been obtained to isolate the signifIcance of each effect.
The focus is mainly on the low frequency surge response. In this respect, the validity of
Newman's approximation is evaluated. The high frequency axial force in the tendons is
also investigated.

The equation of motion for the three-degree-of-freedom TI.P system is given by the
equation
(6.1)
where fp(t) is the potential force vector and fO(xl,t) is the drag force vector. The drag force
vector is modeled by the Morison equation drag force given by equation (5.3). The
potential force vector is the sum of linear and quadratic potential forces and can be written
as the following second order Volterra series
00 00
fp(t) =I hg>(tl) ll(t-tl) dtl + II h~(tl,t2) 1l(t-tl)1l(t-t2) dtldt2 (6.2)
The vectors h~>(t) and h~>(tl,t2) are the linear and quadratic Volterra kernels for the
potential force and have the associated vector transfer functions It>(co) and H~>(COI ,CO2)
116
which have been derived in Chapter 4 using linear diffraction theory. The quadratic
potential force includes the wave elevation drift force, the velocity head drift force, and the
body motion drift force.
The solution to equation (6.1) is again separated into mean and zero-mean
components in the form
x(t) = lIx + i(t) (6.3)
where lIx is the mean displacement vector, and i(t) is a vector of stationary, zero-mean
processes, which are not gaussian.
Substituting equation (6.3) into equation (6.1) and ensemble averaging yields
lIx = k-1<fo(i1,t) + fp(t» (6.4)
This equation can be rewritten by substituting equatio~ (6.2) into equation (6.4) and
evaluating the expectations in the form
00
lIx = k-1<fo(i1,t» + k- 1 I H~)(-rolorol)S1'\1'\(rol) drol (6.5)
Thus, the mean response vector consists of identifiable viscous and potential components.
The components, however, can not be computed for the individual quadratic order forces
and then superimposed. This is due to the fact that the expectation of the drag force is
computed from the probability distribution obtained for the response of the system to the
combined forces. The potential quadratic transfer function, due to the body motion effect,
also depends on the response of the system to the combined forces.
The zero-mean response, i(t), is obtained by determining the response of an
"equivalent" quadratic system. The form of the equivalent system is
m l. + c t + k i = fp(t) - <fp(t» + io(i1,t) (6.6)
where the drag force vector, io(i1,t), is defined by equation (5.13).

To solve for the unknown quadratization coefficients, ~(x,z)'s, the mean square of
the following error vector is minimized
£ ;: k J.I.x - fo(i1,t) + io(i1,t) - <fp(t» (6.7)
Substituting equation (6.4) into equation (6.7) and recalling the analytical expressions for
fo(il,t) and io(i1,t), the error vector can be rewritten into the following form
117
Nm
t =L [ J ~DIl ( <IU+vl(u+v» - IU+vl(u+v)
n=l Ln
It is noticed that this equation is the same as the equation for the error vector in Chapter 5
for which no quadratic potential forces are included in the excitation. This is a logical result
because the error which is minimized is between the exact and "equivalent" drag force.
Thus, the quadratization coefficients are again computed using equations (5.17) to (5.20).
Applying the Volterra series method to approximate the solution of the equivalent
quadratic system leads to
m ~ + (c + Ceq)! + kg = let) (6.9)
where l(t) is the following second order Volterra series
(6.10)
and the hydrodynamic damping matrix, Ceq, is given by equation (5.22). The linear force
transfer function, ~l)(ro), includes viscous and potential components and is again
computed by equation (5.33). The quadratic force transfer function also includes viscous
and potential components. The quadratic transfer function for the drag force is given by
equation (5.35). The quadratic transfer function for the various potential forces are in
Chapter 4.
The response of the Volterra system in equation (6.9) again can be written as the
following second order Volterra series
(6.11)
The response moments, spectral densities, and probability distributions are computed in a
manner identical to the one used in Chapter 5. Note that the use of the either the
quadratization or linearization method applies only to the modeling of the drag force. Thus,
even if the linearization method is used the TLP response is still of quadratic order and non-
118
gaussian due to the quadratic potential forces. The equivalent stochastic quadratization
method, however, must be used to account for the viscous quadratic forces.
6.3 Numerical Results

Numerical results are presented for the idealized TLP shown in Figure 1.1. The
model parameters are given in Table 4.1 and the natural frequencies of the system have
been given in Chapter 5.
6.3.1 Response to Quadratic Drag Force

The fIrst response case considered is the same as in Chapter 5 in which the only
quadratic force in the excitation is the viscous drag force. Results are presented in Chapter
5 for the TLP response to a sea state defmed by a P-M wave spectrum with H s=19.2 m and
Tavg= 16.9 sec and for different current velocities. Th~ equivalent stochastic quadratization
method yields reliable statistics and the magnitude of the quadratic response is dependent
on the current velocity. The quadratic drag force response to different wave states,
however, has not been addressed. Therefore, in this section the TLP response due to
various wave states is investigated. The wave states are described by Pierson-Moskowitz
spectra with wind velocities ranging from 20 to 36 rn/sec. These correspond to moderate to
severe wave states. The associated signifIcant wave heights and average periods are given
in Appendix C. A current profIle which varies linearly from a value of 1.0 rn/sec at the
mean surface to .5 rn/sec at the keel of the hull is assumed. The current velocity is in the
direction of the wave propagation.
The surge, heave, pitch, and axial tendon force response spectra obtained by
quadratization, linearization, and simulation for a wind speed of 30 rn/sec, H s=19.2 m, and
T avg= 16.9 sec are shown in the previous chapter.
In Figures 6.1 and 6.2, the mean surge and mean pitch response are plotted versus
the signifIcant wave height. Linearization, quadratization, and simulation results are all in
good agreement.
In Figures 6.3, 6.4, 6.5, and 6.6, the surge, heave, pitch, and tendon force
standard deviation are plotted versus the signifIcant wave height. Quadratic order effects
are only notable in the surge response. Linearization underestimates the surge response
standard deviation because the low frequency resonance response is not included. As the
wave states become less severe, the nonlinear low frequency response becomes larger
119
.f - - Equiv. Quad.
- - Equiv. Lin.
2
• Simu lation
o ~ ______ ~ ______ ~ ______ ~ ______ ~ ______ ~
5 10 20 25 30
m
Figure 6.1 Mean Surge Response vs. Significant Wave Height

Quadratic Forces: Viscous Drag
(U(z) ., [1 + .5 zldJ mlsec)
3.0
2.5
-"-'=~ --- ---
~ 2.0
'"0
~
- - Equiv. Lin.
1.0 • Simu lation
::£'"
0.5
0.0
5 10 15 20 25 30
H. m
Figure 6.2 Mean Pitch Response vs. Significant Wave Height

(U(z) • [1 + .5 zldJ mlsec)
120
5 - - Equiv. Quad.
,."
,.,,'
- - Equiv. Lin. ,."
,."
E
4 • Simulation
,."
,."
,."
,."
,."
,."
3 ,."
,."
,."
,."
,."
2 ,."
,."
,."
,."
,."
./
./
o
5 10 15 20 25 30
m
Figure 6.3 Surge ReslJOnse Standard Deviation vs. Significant Wave Height
(U(z) = [1 + .5 z/dsl m/sec)
5 - - Equiv. 9uad.
E - - Equiv. Lin.
4 • Simulation
...
N
0
3
X
l£d";N 2
o ~ _____ ~ ______ ~ ______-L______ ~ ______ ~
5 10 15 20 25 30
H. m
Figure 6.4 Heave Response Standard Deviation vs. Significant Wave Height
(U(z) a [1 + .5 z/dsl m/sec)
121
2.0
- - Equiv. Quad.
- - Equiv. Lin.
~ 1.5
• Simulation
. ...
0
x
1.0
li;~
d"
0.5
0.0 '"'-_ _ _........_ _ _........_ _ _-'-_ _ _---'-_ _ _---'

5 10 15 20 25 30
He m
Figure 6.5 Pitch Response Standard Deviation vs. Significant Wave Height
(U(z) = [1 + .5 zldal mlsec)
2.0
- - Equiv. Quad.
- - Equiv. Lin.
1.5 • Simulation
z
::::t
IJ"..:
J;
1.0
0.5
0.0 L -_ _ _........_ _ _........_ _ _- ' -_ _ _---'-_ _ _---'
5 10 15 20 25 30
He m
Figure 6.6 Tendon Force Standard Deviation vs. Significant Wave Height
(U(z) • [1 + .5 zldal mlsec)
122
relative to the linear response. This feature is attributed to the decreasing of the viscous
hydrodynamic damping ratio with decreasing severity of wave states as illustrated in Figure
6.7.
As a reference, the highest axial tendon force standard deviation in Figure 6.6 is
18% of the axial force pretension. For the same wave state, the mean axial tendon force is
.5% of the pretension.
In Figure 6.8, the surge response skewness versus the significant wave height is
plotted. As noted in Chapter 5, quadratization overestimates the simulation results, and no
skewness is obtained by linearization.
6.3.2 Response to Quadratic Wave Elevation/Velocity Head Force

The response to the wave elevation/velocity head quadratic wave force is
investigated in this section. To isolate the quadratic response due to this force, the current
velocity is assumed to be zero, which results in a negligible quadratic drag force response.
In Figure 6.9, the surge response spectral density corresponding to a P-M spectrum for a
wind speed of 30 rn/sec is shown. The corresponding surge response probability
distribution is shown in Figure 6.10. It should be mentioned that the simulation of the
quadratic wave elevation/velocity head force is done directly from the second order Volterra
series functional. This procedure is described in further detail in Appendix D.
The low frequency resonance response at the surge natural frequency is due to the
quadratic wave elevation/velocity head force. Excellent agreement between simulation and
quadratization is obtained. As with the response due to quadratic drag forces, this response
is also non-gaussian as evidenced by the surge probability distribution in Figure 6.10. In
addition, this distribution is skewed in the same direction as the response distribution from
quadratic drag forces.
The mean, standard deviation, and skewness of the surge response are plotted
versus the significant wave height in Figures 6.11 to 6.13. Again good agreement between
the simulation and the quadratization results is observed.
Although it is not shown, quadratization and linearization yield similar results for
this force excitation. This should be expected for a zero current velocity condition, since
the quadratic drag force is negligible compared the quadratic potential force. The linear
drag effects, however, are not negligible and are the source of the viscous hydrodynamic
123
0.5
0.4
~J
0.3

- - Equiv. lin.
It!
0.1
0.0
5 10 15 20 25 30
H. m
Figure 6.7 Surge Hydrodynamic Damping vs. Significant Wave Height

(U(z) • [1 + .5 zldal m/sec)
25
20 - - Equiv. Quad.
- - Equiv. lin.
I(
I( Simulation
<?
15
E
"- 10
;;."
0
5 10 15 20 25 30
H. m
Figure 6.8 Surge Response Skewness vs. Significant Wave Height

(U(z) .. [1 + .5 zldal m/sec)
124
~ ~----------------------------------I
60
f!l
(/) - - Equiv. Quad.
'"E ....••.. Simulation
40
]:
<><-
(f)<~
20
o
0.0 0.2 0.4 0.6 0.8 1.0
co radlsec
Fi~ure 6.9 Surge Response Power Spectral Density

Quadratic Forces: Viscous Drag. Wave ElevationNelocity Head
(Hs .. 19.2 m. Tavg = 16.9 sec. U(z). 0 mlsec)
0.16
0.14

........ Simulation
';'
E 0.10
0.08
~
a. 0.06
0.04
0.02
0.0
·15 · 10 ·5 o 5 10 15
X, m
Quadratic Forces: Viscous Drag. Wave ElevationNelocity Head
(Hs • 19.2 m. Tavg. 16.9 sec, U(z). 0 mlsec)
125
2.0
- - Equiv. Quad.
• Simulation
1.5
----- •
E
i
1.0
.:f
0.5
0.0 r::...._ _ _....L..._ _ _ ~ _ _ _ _. l -_ _ _--L.._ _ _~
5 10 15 20 25 30

Quadratic Forces: Viscous Drag, Wave ElevationNelocity Head
( U(z) = 0 m/sec )
4 - - Equiv. Quad.
E
~
d"
2
o ~ ___. l -_ _ _....L..._ _ _--L.._ _ _ ~ _ _ _~
5 10 15 20 25 30
Figure 6.12 Surge Response Standard Deviation vs. Significant Wave Height
( U(z) = 0 m/sec )
126
120
100 - - Equiv. Quad.

• Simulation
80
'"E
60
"'-
d" 40
20
0
5 10 15 20 25 30
H,. m

( U(z) = 0 m/sec )
127
damping which significantly reduces the magnitude of the quadratic resonance response.
For this excitation, the surge hydrodynamic damping ratio is .25.
6.3.3 Response to Quadratic Body Motion Force

The response due to the body motion force which results from computing the linear
forces at the displaced position of the TI...P is investigated in this section. The current
velocity is again assumed to be zero to isolate the quadratic response resulting from this
effect. The surge response power spectral density corresponding to a P-M spectrum for a
wind speed of 30 rn/sec is shown in Figure 6.14. Note the good agreement between the
simulation and the quadratization results. For the simulation, the body motion effect is
modeled by computing the linear force at the displaced position of the TI...P hull. Appendix
D gives the details of this procedure. The probability distribution of the surge response is
shown in Figure 6.15. The distributions obtained by simulation and quadratization are
both skewed, but the location of the mean responses are slightly different.
In Figures 6.16 to 6.18 the mean, standard deviation, and skewness of the surge
response are plotted versus the significant wave height. The agreement between the
quadratization and the simulation results for the surge standard deviation and skewness is
good. There is obviously poor agreement between the quadratization and simulation results
for the mean response. The magnitude of the error as related to the total response,
however, is quite small. For example, for the most severe sea state considered there is a
difference of .55 meters between the mean response obtained by quadratization and
simulation. Compared to a standard deviation of 5.44 meters this is small. Compared to a
mean response of 7.49 meters for the current velocity profile assumed in section 6.3.1, this
difference is also small. This error in the mean response prediction is due to a higher order
interaction of the drag force and the body motion effect which is captured by the simulation
method while it can not be accounted for by the analytical method. This is deduced by
obtaining simulation results for a case in which the body motion effect is modeled, but the
drag force is set equal to zero. The resulting mean responses for that case are close to the
quadratization results.
Spanos and Agarwal(l984) use the equivalent stochastic linearization method to
model the TI...P response including the displaced position effect. Simulated results are
obtained for comparison. The body motion force, which can be viewed as a time varying
stiffness, is replaced by an equivalent constant linear stiffness. The nonlinear drag force is
128
80 ~-----------------------------------------,
&l 60
N
'" - - Equiv. Quad .
E ........ Simulation
40
]:
C/)
''"'-
M
20
0
0.0 0.2 0.4 0.6 0 .8 1.0
CIl radlsec
Figure 6.14 Surge Re~nse Power Spectral Density

Quadratic Forces: iscous Drag, Body Motion
(H,. = 19.2 m, Tavg = 16.9 sec, U(z) = 0 m/sec)
0.16
0.14

........ Simulation
0.10
E
0.08
~
.!:!. 0.06
a.
0.04
0.02
0.0
·15 ·1 0 ·5 0 5 10 15
Xl m

Quadratic Forces: VISCOUS Drag, Body Motion
(H,. • 19.2 m, Tavg • 16.9 sec, U(z) • 0 m/sec)
129
0.5
0.4 - - Equlv. Quad.

• Simulation
0.3
E
0.2
0.1
.f
0.0 •
•
-0.1
•
-0.2
5 10 15 20 25 30

Quadratic Forces: Viscous Drag. Body Motion
( U(z) = 0 m/sec)
E 4 - - Equiv. Quad.
• Simulation
3
~ 2
o ~ ______ ______ ________ ______ ______

~ ~ ~ ~ ~
5 10 15 20 25 30
H. m
Figure 6.17 Surge Response Standard Deviation vs. Significant Wave Height
Quadratic Forces: Viscous Drag. Body Motion
( U(z) =0 m/sec )
130
80
60 - - Equiv. Quad.
..,
E
• Simulation
40
"'~
d"
20
O~ __~~~__L -_ _ _ _L -_ _ _ _~_ _~
5 to 15 20 25 30
H,. m

Quadratic Forces: Viscous Drag, Body Motion
( U(z) = 0 m/sec )
131
not modeled explicitly in either the linearization or simulation procedures. Instead, the
damping ratio is increased to values ranging from .2 to .3 based on results of drag force
linearization studies by Spanos and Chen(1981). The linearized mean response agrees well
with the simulated results, since the drag force is not modeled explicitly. However, the
linearized standard deviations underestimate the simulated results by up to 10%. This is
due to the fact that the linearized solution can not yield a quadratic response at the surge
resonance. Finally, the non-gaussian behavior of the surge response is not accounted for.
6.3.4 Response to Combined Viscous and Potential Quadratic Forces

The response of a TLP to combined viscous and potential drift forces has been
obtained assuming that the current velocity is 1.0 m/sec at the surface and .5 m/sec at the
keel of the hull.
In Figure 6.19, the surge response spectral density due to individual quadratic drift
forces and to the combined drift forces are shown. A P-M spectrum for a wind speed of 30
m/sec is used for the wave spectrum. The simulated spectral density is shown only for the
combined force case and is in excellent agreement with the spectral density obtained by the
quadratization method. The low frequency response due to the combined forces is
substantially larger that any of the low frequency responses due to individual quadratic
forces. In addition, the low frequency response contribution to the surge response
standard deviation is about the same as for the wave frequency response. It is clear that all
of the quadratic forces contribute to the resonance response. In Figure 6.20, the pitch
response spectral density due to combined quadratic forces is shown. Quadratization and
simulation results agree well, but again the quadratic order effect is quite small.
In Figure 6.21, the surge response probability distributions for the combined force
case are shown. In general, the simulation and quadratization results are in good
agreement. The differences that do exist are in the tails, and are attributable to the
limitations of the truncated Gram-Charlier expansion. Also there is a small difference in the
predicted mean values. This latter difference is attributed to the interaction between the
drag force and body motion effect and is small compared to the total mean response.
In Figures 6.22 to 6.24, the mean, standard deviation, and skewness of the surge
response are plotted versus the significant wave height. The combined response as well as
the response to individual quadratic order forces are shown. The response in which no
quadratic order forces are included is given by the linearized case. For the mean response,
132
160
140 - - Combined Forces. Equiv. Quad .• U(O) _ 1 mJsec

••••• •• • Combined Forces. Simulation. U(O) _ 1 mlsec
120 - - - Drag Force. Equlv. Quad. . U(O) - 1 mlaec

al
fJ) • • Drag Force. Equiv. Un .• U(O) - 1 mlsec
'"E 100 - - Wave EINel. Head Forces, Equiv. Quad.,
U(Ol _ 0 mlSec
- - - Bodv Motion Forca, Equlv. Quad.,
U(O) _ 0 mlSec
80
:g
<><
60
(1)<><
40
20
0
0.0 0 .2 0.4 0 .6 0.8 1.0
(() rad/sec

Quadratic Forces: Viscous Drag, Wave EI.Nel. Head, Body Motion
(H• • 19.2m, Tavg = 16.9 sec, U(z) = U(O)[l+.Sz/d.J)
al
fJ)
6 - - Combined Forces, Equlv. Quad.

'"
~ ........ Combined Forces, Simulation
.0 4
3
-:~ 2
'"
<I)
0
0.0 0.2 0.4 0.6 0.8 1.0
CIl rad/sec
Figure 6.20 Pitch Response Power Spectral Density

(Hs = 19.2 m , T.o;g . 16.9 sec, U(z) = [1 + .5 zIdaI mlsec)
133
0.12
- - Combined Forces. Equiv. Quad .
•• ••• ••• Combined Forces. SImulation

0.10
0.08
E
0.06
,(
a:
0.04
0.02
__ _ ___
0.0
~ o*O .
·10 0 20 30

(H. • 19.2 m, Ta"ll. 16.9 sec, U(z). (1 + .5 zldJ m/sec )
16
- - Combined Forces, Equiv. Quad., U(O) • 1 rnIsec
14 - - - D<ag Force. Equiv. Quad.• U(O) • 1 rnIsee

• - D<ag Force. Equiv. Un., U(O) • 1 rnIsee
- - Wave EINel. Head Forces. Equiv. Quad..
12 U(~.O mfsec
- - - ~(O) ~tgn: ' Equiv. Quad .•
10 • Comlilned Forces. SImulation.
U(O) • 1 rnJsec
E
8
-----_ .. - -----
~ _ ... _ __ oJ _ ..;.. -
~
f 6
--
4
2
- - - - - - - - - -- - - -
0
5 10 15 20 25 30
Figure 6.22 Mean Surge Response vs. Sig. Wave Height

(U(z) ~ U(O) (1 + .5 zldJ )
134
8
- - Combined Fo<ces. Equlv. Quad .• U(O) -1 mlsec
- - - Drag Force. Equiv. Quad .• U(O) - 1 mlsec
• Drag Force. Equiv. Un .• U(O) - 1 m/sec
6 - - waU(orl!'oe~ad Forces. Equiv. Quad .•

- - - ~ Motion Force. Equiv. Quad..
E
• Com/'~~;dOF=. Simulation.
U(O) _ 1 mtsec
L;...- 4
<
2
o ~ __ ~ ___ ~ _ _ _-L_ _ _ ~ __ ~
5 10 15 20 25 30
H. m
Figure 6.23 Surge Response Standard Deviation vs. Sig. Wave Height
Quadratic Forces: Viscous Drag, Wave EINel. Head, Body Motion
(U(z) = U(O) [1 + .5 Z/dsl )
300
- - Combined Forces. Equiv. Quad .• U(O) _ 1 m/sec
- - - Drag Force. Equlv. Quad .. U(O) - 1 mlsec
250
• Drag Force. Equlv. Lin .. U(O) - 1 mlsec
- -
Wave EINel. Head Forces. Equiv. Quad.,
U(O) - 0 mtsec
200 - - - Bodv Motion Force, Equlv. Quad ..
U(O) .0 mtsec
"E • Combined Forces, Simulation,
U(O) - 1 mtsec
150
~-
:1" 100
50
0
5 10 15 20 25 30
H,. m
Figure 6.24 Surge Response Skewness vs, Sig. Wave Height

(U(z) = U(O) [1 + .5 Z/dsl )
135
the viscous force is the major contributor, followed by the wave elevation/velocity head
potential forces. As noted before, the body motion force has a small effect on the mean
response, but is the source of the difference between the simulation and quadratization
results. Overall, quadratization gives reliable results for the mean surge response due to
combined quadratic forces.
For the surge standard deviation, good agreement between the simulation and
quadratization results is seen. The linearization solution is shown to indicate the standard
deviation response due only to the linear, or wave frequency response. The individual
quadratic forces contribute significantly to the total low frequency response.
The wave elevation!velocity head drift force has the largest effect in tenns of the
surge skewness response, followed by the body motion force and then the drag force.
Again, all of these forces are important in inducing skewness in the response probability
distribution.
Note that the relative contribution of an individual quadratic force to the surge
response depends on the order of the surge response statistic. For example, while the body
motion force induces only a small mean response, it has a relatively large effect on the
skewness. Vice versa, the drag forces can cause large mean responses, but have a
relatively small effect on the skewness. Thus, for stochastic TLP analyses it is
inappropriate to neglect any quadratic force.
6.3.5 Evaluation of Newman's Approximation

The reliability of the Newman method to approximate the quadratic transfer function
is investigated in this section. It is recalled that this approximation has been proposed to
obtain the low frequency surge response of floating systems. The reason for using the
approximation is that substantial computational savings are achieved since only a small part
of the quadratic transfer function is needed.
The effect of using the approximation for the different quadratic order forces has
been evaluated. A P-M wave spectrum corresponding to a wind velocity of 30 m/sec is
used. In Figure 6.25, the exact and approximate surge response spectral densities are
shown for a case in which the only quadratic force in the excitation is the viscous drag
force and the current velocity is 1.0 m/sec at the surface and .5 m/sec at the keel. This is
the same load condition considered in section 6.3.1. The response obtained using the full
quadratic transfer function is what is meant by "exact". The simulated spectral density is
136
80
60
- - Equiv. Quad., Exact
'"E
- - - Equiv. Quad., Newman Approx .
........ Simulation
40
20
o~__~~____~__~~======~====~
0.0 0.2 0.4 0.6 0.8 1.0
ffi rad/sec

(H. • 19.2 m, Tavg. 16.9 sec, U(z) a [1 + .5 zldsl m/sec )
4 - - Equiv. Quad., Exact

~ - - - Equiv. Quad., Newman Approx.
N
Z
::i 3
]: 2
H ~::
(/)
----... .. - ..........
'-- .. ""- ....................
o~__~__~~====~
...................... -..
---__ -=~====~
0.0 0.2 0.4 0.6 0.8 1.0
ffi rad/sec
Figure 6.26 Low Frequency Quadratic Surge Force PSD

(H,. = 19.2 m , Tavg - 16.9 sec, U(z) = [1 + .5 zldsl m/sec)
137
also shown for comparison. For the quadratic drag force, the Newman approximation
leads to results which are in good agreement with the exact solution. The reason for the
good agreement is understood by examining Figure 6.26 which shows the low frequency
quadratic surge force spectral densities obtained by the two methods. Although the two
spectra are not similar over the entire range of frequencies, there is very good agreement at
the surge natural frequency which is marked by the vertical arrow. This figure highlights
the premise of the Newman approximation: at low frequencies near zero, the approximate
and exact quadratic force spectra should be nearly equal since they are equal at the zero
frequency. This assumption works well for the quadratic drag force. However, for the
potential quadratic forces this does not prove to be the case. In Figure 6.27, the exact and
approximate surge response spectral densities for the quadratic wave elevation/velocity
head forces are given. The current velocity is equal to zero and the drag force is linearized
instead of "quadratized" so that only quadratic order responses due to the potential forces
are obtained. The Newman approximation significantly underestimates the low frequency
resonance response. The reason for this is evidenced by the quadratic surge force spectral
densities shown in Figure 6.28 where there is considerable difference between the exact
and the approximate spectra at the surge natural frequency. The same result holds true for
the body motion drift force as shown in Figures 6.29 and 6.30.
In Figures 6.31 and 6.32, the exact and approximate response and force spectral
densities are shown when all quadratic order forces are included. The Newman
approximation is unconservative and underestimates the surge standard deviation by 10%.
In Table 6.1, the statistics for the exact, approximate, and simulated solution have
been tabulated. The errors in the means response due to the approximation are small as
should be expected. The approximation, however, is unconservative in predicting the
standard deviation. The error is even more pronounced in underestimating the surge
response skewness.
6.3.6 High Frequency Axial Tendon Force

The wave states which have been considered thus far are considered to be moderate
to severe. Under these conditions, the resulting axial tendon force response is almost
completely dominated by the heave response, which for the present model exhibits no
quadratic order effects. In normal or operational wave states, however, the axial tendon
force becomes dominated by the pitch response. Also, the wave spectra corresponding to
138
80
~
U) 60
0/ - - Equiv. Lin., Exact
E
- - - Equiv. lin., Newman Approx .
...•.... Simulation
40
]:
,".-
,,.
(J)
20
0.0 0.2 0.4 0.6 0.8 1.0

(j) radlsec

Quadratic Forces: Wave ElevationNelocity Head
(H. = 19.2 m, Tavg" 16.9 sec, U(z) = 0 m/sec)
5 ~------------------------.
- - - Equlv. Un., Exact
4 - - - Equlv. Un., Newman Approx.
0/
Z
~ 3
a
0.0 0 .2 0.4 0.6 0.8 1.0
(j) radlsec

Quadratic Forces: Wave ElevationNelocity Head
(H... 19.2 m, Tavg. 16.9 sec, U(z) .. 0 m/sec )
139
80
60
, - - Equiv. lin., Exact
OJ
E - - - Equiv. lin., Newman Approx .
......•. Simulation
40
"'-
C/) '"
20
0.0 0.2 0.4 0.6 0.8 1.0

w rad/sec

Quadratic Forces: Body Motion
(H. = 19.2 m, Tavg Z 16.9 sec, U(z) = 0 m/sec)
10
- - Equlv. Un., Exact
OJ
z 6
::::!:
o '--_-_-_--=
-::..:-:=.-..;:-::..:-:.::-::..:-:.::-::..:-'-=-:.::-;;..--=
- _-_-
_-~_--
_ --
_ --
_ ....L.---
- _ --_--...;:.:.--::..
0.0 0.2 0.4 0.6 0.8 1.0
w fad/sec
Quadratic Forces: Body Motion
(H. = 19.2 m, Tavg. 16.9 sec, U(z) = 0 m/sec)
140
160
140
- - Equlv. Quad., Exact
120 _ . - Equlv. Quad., Newman Approx.
~ ••• ••••. SlmulaUon
N 100
E
80
~ 60
<~
CI) '" 40
20
0
0.0 0 .2 0.4 0.6 0.8 1.0
(J) rad/sec

(H• • 19.2 m, TallO" 16.9 sec, U(z) = [1 + .5 zldJ m/sec)
40
- - Equlv. Quad .. Exact
- . - Equlv. Quad .. Newm an Approx.
30
~,
N
z
:::E
20
~
g ~ ::
CI)
10
o ~~_~ _ _ _-L_ _ _- L_ _ _ ~_ _~
0.0 0.2 0.4 0.6 0.8 1.0

(J) rad/sec

(H. • 1.36 m, TallO" 4.51 sec, U(z) .. [1 + .5 zldJ m/sec)
Table 6.1
Comparison of Quadratic Order Response Solutions Obtained by

the Exact Method and Newman's Approximation
(U(z) = U(0)[1 + .5 z/d s ])
Quadratic Force ~ meters p:;;x, meters 11'3

x, meters3
Loading Condition
Exact Approx. Sim. Exact I Approx. Sim. Exact I Approx. Sim.
Drag Force, Equiv. Quad.,
U(O) = 1 m/sec 6.86 6.86 6.84 3.50 3.51 3.57 11.8 11.9 7.1 ...
I ~
Wave EI.Nel. Head,
Equiv. Lin., U(O) =0 m/sec 1.13 1.32 1.14 3.44 3.17 3.44 21.4 1.0 19.6
Body Motion, Equiv. Lin.,

U(O) =0 m/sec .21 .30 -.05 3.23 3.09 3.24 16.0 2.3 13.1
Combined Quad. Forces,

Equiv. Quad., U(O) = 1 m/sec 8.68 8.74 8.26 4.16 3.88 4.27 61.2 23.1 54.8
142
operational wave states, have less energy than the previously considered spectra, and are
distributed more uniformly over a wider range of frequencies. Thus, it may be expected
that quadratic order forces will cause significant resonance responses in the pitch mode and
consequently in the axial tendon force response.
In Figure 6.33, axial tendon force spectral densities obtained by simulation and
quadratization for the combined quadratic forces are shown. A P-M spectrum
corresponding to a wind speed of 8.0 rn/sec(Hs=1.37 m, Tavg= 4.5 sec), and a current
velocity profile with a velocity of 1.0 rn/sec at the surface and .5 rn/sec at the keel define the
environmental conditions. The spectra obtained by simulation and quadratization are in
reasonable agreement. The two dominant response peaks are at the wave frequencies and
at high frequencies near the pitch natural frequency which is equal to 3.24 rad/sec. In
addition, there is a low frequency peak at the surge natural frequency which for practical
purposes is negligible. For the two methods, the high frequency peaks occur at slightly
different frequencies. The peak from simulation occurs at a frequency which is slightly
less than the pitch natural frequency. This trend should be attributed to the well known
numerical integration phenomenon of period elongation which distorts the true frequency.
This distortion can be reduced by taking smaller time steps. However, this increases the
computational effort in the simulation. The present simulated spectrum is obtained for a
time step increment of.2 seconds and gives the high frequency peak occurring at 3.20
rad/sec. To understand the sensitivity of this distortion to the time step size, another
simulation has been conducted in which a time step size of .4 seconds is used. This
resulted in the peak occurring at 3.05 rad/sec. This frequency distortion deserves attention
in using simulation methods.
The corresponding axial tendon force probability distributions have been plotted in
Figure 6.34. For the most part, there is very little skewness and the axial force is nearly
gaussian.
In Figures 6.35 and 6.36, the source of the contributions to the high frequency
peak in the tendon force spectral density are presented. Figure 6.35 shows the high
frequency range of the axial tendon force spectral density for cases in which only quadratic
forces due to viscous effects are considered. For comparison, the spectrum for the case in
which there is no drag force is also shown. This latter spectrum shows that there is a
notable high frequency response due to linear potential forces. Including viscous forces
using the quadratization method increases the high frequency response. Also the response
143
600
- - Combined Forces, Equlv. Quad
500 -- --- --- Combined Forces, Simulation
~ 400
N
Z
~
300
~<..:-
<- -
200
n
II)
J\~-- -
100
) \~
1\;. ..
o t1.
0 2 3 4
(j) radlsec
Figure 6.33 Axial Tendon Force Power Spectral Density

(H. - 1.36 m, Tavg a 4.51 sec, U(z) = [1 +.5 zldsJ m/sec)
40
- - Combined Forces. Equlv. Quad. ~.

:$ ....
....... Combined Forces. Simulatlon ' "
30
20
10
o C===~===C=--L __~~__L-__~~-L__~
-0.06 -0.04 -0.02 0.0 0.0 2 0.04 0.06 0.08 0.10
I, MN
Figure 6.34 Axial Tendon Force Probability Distribution

(H. • 19.2 m, Tavg = 16.9 sec, U(z) = [1 + .5 zldsl m/sec)
144
500
400 - - No Drag Force

Drag Force. Equiv. Quad.. U(O) - 1 mlsec
l!l
'" - - Drag Force. Equlv. Qualt. U(O) _ 0 mlsec
'"Z 300
)oC
,I1
200 ,1
§; , I
<--
<-- I I
'·\
, I
C/)
, I
fU
100 , I
, /11
, II,
~'I "\
0 ~
2.4 2.6 2.8 3 .0 3.2 3.4 3.6
(Il rad/sec
Figure 6.35 High Frequency Axial Tendon Force PSD

Quadratlc Forces: Viscous Drag
(H. = 1 3. 6 m, Tavg = 4.51 sec, U(z),. U(O) [1 + .5 zlcI.J )
500
- - Combined Forces. Equlv. Quad .. U(O) - 1 m/sec
........ Combined Forces. Simulation. U(O) - 1 mlsec
- - - Drag Force, Equlv. Quad .. U(O) _ 1mlsec
400
- - Wave EINel. Head Foroes. Equlv. Quad.•
&l U (0) - 0 mlSec
N
'" - - - 80dv Motion Force. Equiv. Quad ..
U(0) - 0 mlSec
/
Z
1:AI/
)oC 300
: I
§; 200 ,/'
<-'"'
<-'"'
C/)
100
./~
o
2.4 2.6 2.8 3.0 3.2 3 .4 3.6
(0 rad/sec
Figure 6.36 High Frequency Axial Tendon Force PSD

(H. = 1.36 m, Tavo ,. 4.51 sec, U(z) = U(O) [1 + .5 zldsJ )
145
increases with increasing current velocity. Spectra have also been obtained using
linearization but have not been shown for sake of clarity. These spectra are nearly identical
to the spectra obtained by quadratization for both current velocity cases. This indicates that
for this model the linear viscous force has a more significant contribution to the high
frequency response than the quadratic viscous force. This may not be a general rule,
however, since the linear and quadratic viscous force spectral densities are very uneven in
the high frequency range due to the wave force cancellation effects. Thus, for models
which have slightly different natural frequencies or different column spacings the quadratic
force spectrum may have a larger value than the linear force spectrum at the pitch natural
frequency. Neglecting the wave force cancellation effect for the high frequency forces is
conservative and may be appropriate to do in light of the uncertainties in predicting the
natural frequencies.
In Figure 6.36, the contributions to the high frequency peak in the tendon force
spectral density by the various quadratic forces are plotted. The body motion quadratic
force has only a very small effect on the high frequency response. The wave
elevation/velocity head force, however, adds substantially to this response.
The occurrence of this high frequency response will have a significant impact on the
fatigue design of the tendons. It is also another argument for using full quadratic transfer
functions since the Newman approximation yields no high frequency response.

The response statistics of a TI..P subject to viscous and potential first and second
order forces are obtained. The drag force nonlinearity is treated by the equivalent stochastic
quadratization method. The potential forces are described by the second order Volterra
series equations derived in Chapter 4. A third order Gram-Charlier expansion is used to
describe the non-gaussian response probability distribution.
The quadratization results agree well with simulations results for a wide range of
sea states. For moderate to severe wave states, only the surge response of a TI..P is
significantly affected by higher order wave forces. Of the quadratic forces considered, all
have a significant influence on this response. The relative contribution of an individual
quadratic force to the surge response depends on the order of the surge response statistic.
The Newman approximation is found to be unconservative for predicting the magnitude of
the low frequency surge response.
146
For operational wave states, both linear and quadratic forces induce significant
responses in the pitch mode, and consequently in the axial tendon force. In particular, the
high frequency response at the pitch natural frequency will have a significant impact on the
fatigue design of the tendons.
CHAPTER 7
SUMMARY AND CONCLUSIONS
A method called equivalent stochastic quadratization has been proposed for

stationary response analyses of nonlinear systems. The procedure can be viewed as an
extension of the equivalent stochastic linearization method. Equations have been developed
for general single and multi-degree-of-freedom systems subject to fIltered gaussian white-
noise excitation. An "equivalent" quadratic system is constructed by replacing the.
nonlinearity with polynomials up to quadratic order. The equivalent system has a form
whose solution can be approximated by using the Volterra series method. The coefficients
of the polynomials are determined from a mean square minimization. An iterative method
is required. A third order Gram-Charlier expansion is used to describe the non-gaussian
response probability distribution. Numerical results are presented for one and two-degree-
of-freedom models with quadratic type nonlinearities. The results are compared to
simulation to verify their reliability. The proposed method offers a notable improvement
over linearization when the nonlinearity is non-symmetric and the excitation frequencies are
banded in a range outside the natural frequency of the system. In this case, significant
resonance responses can occur which are not accounted for by linearization. When the
nonlinearity is symmetric the quadratization solution reduces to the linearization solution.
The equivalent stochastic quadratization method is used to analyze the response of a
compliant offshore platform to wave and current induced forces. A three-degree-of-
freedom model of a Tension Leg Platform (TLP) is used to exemplify the method. To
minimize the TLP's dynamic response, the natural frequencies are designed to be outside
the frequencies of the dominant part of the wave spectrum. The surge natural frequency is
lower than the average wave frequencies and the pitch and heave natural frequencies are
higher than the average wave frequencies.
Wave force nonlinearities due to potential and viscous effects are considered. The
potential force nonlinearities are derived in the form of second order Volterra series. The
full quadratic transfer functions are derived. Therefore, reliance on the Newman
approximation is not needed. The viscous force is modeled by the generalized nonlinear
Morison equation. The quadratization method approximates this nonlinearity with a
Volterra series expansion. Thus, the drag force is easily combined with the potential force.
148
Numerical results have been obtained for the various quadratic forces acting alone
and in combination. For the case with only viscous quadratic forces, the quadratization
method yields, in addition to the wave frequency response, a significant mean response and
low frequency resonance response at the surge natural frequency of the TLP. The
magnitude of the mean and low frequency response increases with the current velocity.
These results have good agreement with simulation. Linearization yields the wave
frequency response and the mean response but no low frequency response. Consequently,
the second and third order statistics are significantly underestimated. For moderate to
severe wave states, only the surge response is significantly affected by the nonlinear drag
forces.
In the combined quadratic force case, it is found that all of the quadratic forces have
a significant contribution to the surge response. For the most part, the results have very
good agreement with simulation. The relative contribution of an individual quadratic force
on the surge response depends on the order of the surge response statistic. The Newman
method for approximating the low frequency surge response is found to be unconservative
in the second and third order moments.
In operational wave states, the pitch response exhibits a wave frequency response
induced by the linear forces and a high frequency response at the pitch natural frequency
induced by the linear and quadratic forces. Consequently, the axial tendon force response
exhibits the same features. This is an important consideration for the fatigue life calculation
of the tendons.
In terms of treatment of the quadratic order forces, Table 7.1 gives a comparison of
previously published TLP frequency domain studies. Three basic approaches have been
used to treat quadratic forces. Linearization of the quadratic force is the basis of one
approach which has frequently been used for the drag force nonlinearity. A second
approach involves approximating the quadratic force transfer function using the Newman
approximation. This method has been used for quadratic forces due to both drag and
potential effects. A third approach, which is analytically the most precise approach, relies
on deriving the full quadratic force transfer functions. The present study is the frrst one in
which this has been done for the drag force. This is accomplished by relying on the
equivalent stochastic quadratization method. Others have obtained quadratic transfer
functions for the potential forces. Most, however, rely on numerical methods such as finite
element methods or sink-source methods which are more accurate than analytical methods
Table 7.1 Treatment 01 Quadratic Forces in Previous TLP Studies
DRAG FORCES POTENTIAL FORCES

Morison Equation Body Motion ~ave ElevationNelocity Heae
Published Study Quad. Quad. Quad. Quad. Quad. Quad.
Lin. Approx. Full Lin. Lin.
Approx. Full Approx. Full
Analytical
Paulling and Horton(1970) x
Natvig and Pendered(1977) x
Pijlers and Brink(1977) x
Albrecht et al(1978) x
Kirk and Etok(1979) x
Mercier et al( 1982) x ~
Bums(1983) x
Spanos and Agarwal(1984) x
Petrauskas and Liu(1987) x x
Present Study(1989) x x x
-------------- ---- ---- ---- ---- ---- ----- t----- >----- ----
Numerical
Tan and deBoom(1981) x x x
Botelho et al(1984) x x x
Kobayashi et al(1987) x x x
Marthinsen(1989) x x x
150
but are computationally costly. An analytical method for obtaining the quadratic transfer
function for the potential forces has been reported in one study, although no details are
given. In addition, the quadratic transfer function for the body motion potential forces are
not included. An important contribution of the present study is that the full quadratic
transfer functions are derived for all of the quadratic forces considered. In this respect, no
other study is as comprehensive as the present one.
There are several possible improvements which can be made to the proposed
analytical method. One improvement is to include higher order terms in the polynomial
expansion of the nonlinearity. Another improvement is to include higher order terms in the
Gram-Charlier expansion of the probability distribution. Both of these improvements
could yield more reliable statistics. However, they would substantially reduce the
efficiency of the procedure. Therefore, the implementation of these improvements should
be balanced against level of uncertainty in the model and the desired accuracy of the
solution.
Clearly, as a method for analyzing general nonlinear systems in which it is
presumed that the analytical model is an accurate representation of the physical system, the
improvements are desirable. Efficient numerical methods may be found which could give
reasonable computer tum around times. The multiple integrals and convolutions in this
study are determined using direct digital methods. Some of these operations may be
amenable to Fast Fourier Transform (FFT) techniques.
For compliant platform applications, it is questionable whether the increased
expense of these improvements is worth the effort. This is especially true with regards to
including higher order terms in the viscous force expansion since there are uncertainties in
parameters such as the drag coefficient as well as the validity of Morison's equation itself.
Also it would be unreasonable to include third order viscous forces when the present state
of knowledge of potential forces is incomplete and only goes up to second order..
For reliability analysis purposes, however, it is desirable to obtain higher order
response moments. It is recommended that analytical improvements should be directed to
this end first. Analytical expressions for the probability distribution which do not have
negative regions should be found. However, it should be kept in mind that it is always
difficult to describe the tails of distributions with only a few moments.
APPENDIX A
GRAM-CHARLIER COEFFICIENTS
A.1 Introduction
In this section, the relations between the Gram-Charlier coefficients and the
moments and cumulants of joint random variables are given. These relations have been
obtained from Beaman and Hedrick(1981) and Nigam(1983).
A.2 Gram-Charlier Coefficients

A multivariate Gram-Charlier expansion of a probability distribution p(x) for the n-
dimensional vector X is written as
().iT cp(x)
(A.I)
where
(A.2)
The symbol cP(x) represents a multivariate gaussian distribution which can be written as
(A.3)
where the matrix S is the covariance matrix for the vector x.

The Gram-Charlier coefficients, rjlj2'''jn' are related to the cumulants, Kjd2'''jn ' of
the joint random variables by the equation
1 jT= 0
0
rjd2'''jn = 0< jT< 3 (A.4)
(_l)h . ,Kjlh'''jn
. i .,
h· J2' ... In' jT~ 3
The joint moments and cumulants are related to each other through the joint
characteristic function, MOO, which is defmed as
152
M(e) !E < exp(i e . x.) > = J .... J exp(i e . 3) p(3) d.x (A.5)
In this equation, the symbol (.) represents the scalar product of two vectors and i ={:l.
The joint moments of the vector ~ are related to the derivatives of the characteristic function
by the equation
(A.6)
The above moments are algebraically related to the joint central moments, J.l.hh'''jn ' which ~
are defmed as
= <n
n .
lLIJ'Z"'J'n
r-J i=1 (x·_<X·>)Ji
1 1 > (A.7)
The joint cumulants of the vector ~ are related to derivatives of the natural logarithm
of the characteristic function by the equation
(A.8)
APPENDIX B
EVALUATION OF EXPECTATIONS
B.1 Introduction
In. this appendix, details of the evaluation of two types of expectations are given.
The first kind involves the quadratic nonlinearity used throughout this study. Fonns of this
expectation are given by equations (2.66), (2.71), and (5.20). The second kind of
expectation involves central moments of random variables whose probability distribution is
approximated by a Gram-Charlier expansion. Recall that due to computational limitations,
only moments up to third order are included in the expansion. Thus, fourth and higher
order cumulants of the approximate probability distribution are zero. The fourth and higher
order central moments, however, are not zero and may need to be computed. In. particular,
the fourth order central moment is needed to compute the quadratization coefficients.
Higher order moments may also be needed depending on the nonlinearity. A simple
method for approximating higher order central moments based on the truncated probability
distribution is described.
B.2 Expectations Involving Quadratic Nonlinearity

Consider the following expectation involving a general nonlinearity g(x)
k = 0,1,2 (B.1)
The non-gaussian random variable x is a represented by its mean, J.I.x, and a zero-mean
x
component by the equation
(B.2)
The expectation in equation (B.1) needs to be evaluated to detennine the mean response and
the quadfatization coefficients as described in Chapter 2. Using the defmition of
expectation, Ek is written as the following integral
k=O, 1,2 (B.3)

154
where p(i) is the probability distribution of i. This non-gaussian distribution is

approximated by the truncated Gram-Charlier expansion
(B.4)
where cp(i) is a gaussian distribution defmed by the equation
(B.5)
and ~im is the m th central moment ofx. Substituting equation (B.4) into equation (B.3)
leads to
(B.6)
(B.7)
(B.8)
where
n=O, ... ,5 (B.9)
For the quadratic nonlinearity used throughout this study, equation (B.9) can be
written in the following general form
In = J-yla + ~x + il(a + ~x + i) in cp(i) di n=O, ... ,5 (B.IO)
This integral is evaluated by the following procedure. A new variable I; =i/{iiii is

introduced and substituted into equation (B.IO). This leads to
n+2 00
In = )'(~2)T JI it + I; I( it + I; )I;n cp!;(I;) dl; (B.ll)

"""
where
155
_ a+J.l.x
a ::;:-- (B.12)
..fIii2
The integration in equation (B.ll) is split over several regions as
n+2 -li
In ::;: y(~2r2 J-( a + ~ )2 ~n cp~(~) d~
"""
n+2 0
+ y(J.LX2)"'2 J( a + ~ )2 ~n cp~(~) d~
-a
+ y(J.Lx2f2
n+2
J(a +
co
~ )2 ~n cp~(~) d~ (B.13)
Making several changes of variables, equation (B.13) is rewritten as
(B.14)
These integrals can be evaluated using standard integral tables and integration by parts.
Evaluating In for n up to 5 yields
10 ::;: 2y (J.Lx2){ (a2 + l)ql + it q2) (B.15)
11 ::;: 4Y(J.Lx2)3/2{a ql + q2} (B.16)
12 ::;: 2y (~2)2{ (a2 + 3)ql + a q2) (B.17)
13 ::;: 4y (J.LX2)5/2{ 3a ql + 4 q2) (B.18)
14 ::;: 2Y(J.L x2)3{3(a2 + 5)ql + aq2} (B.19)
Is ::;: 4y (J.Lx2)7/2{ 15a ql + (a2 + 24)q2) (B.20)
where
(B.2l)
q2 ::;: _1_ exp( - ! a2) (B.22)

{2rt 2
156
B.3 High Order Central Moments

By defmition, the m th central moment, ~im , of a random variable, x, is computed
by the equation
~im = J (x - ~x)m p(x) dx (B.23)
where p(x) is the probability distribution of x. It is assumed that the probability

distribution is approximated by a Gram-Charlier expansion of the form
aj",(x)
= j=O
1 r·
00
p(x) ~ (B.24)
J axJ
where C\l(x) is a gaussian distribution and rj is related to the jth moment of the random
variable x as described in Appendix A. If the distribution is truncated at order j=jv the jt+ 1
and higher order cumulants of the approximate probability distribution are zero. The jt+ 1
and higher order central moments, however, are not zero and may need to be computed. In
this study, the expansion is truncated at the third order. However, the fourth order central
moment is needed to compute the quadratization coefficients. A simple method for
approximating these higher order central moments is described.
If p(x) is given by the truncated Gram-Charlier expansion, evaluation of the integral
in equation (B.23) yields integrals of the following form
(B.25)
To evaluate this integral the relations
(B.26)
(B.27)
(B.28)
are used. The first equation reflects a property of gaussian distributions. The second
equation defmes the Hermite polynomial, Hi~). The third relation states simply that the
157
quantity ~m Hj(~) can be written as an expansion of Hennite polynomials where ~'s are the
coefficients. To initiate the evaluation of I.nj. equation (B.26) is substituted into equation
(B.25) and the variable change ~ = (x - 11x)/~ is made. Then. applying equation (B.27)
yields
.!!!i
= (~2) 2 J~m (-l)l H/~) q,t;(~) d~
00 •
Imj (B.29)
"""
where q,t;(~) is defined in equation (B.12). Substituting equation (B.28) into (B.29) and
recalling the variable change gives
.!!!i ~akq,E.(~)
= (~2)2 J (-I)l~ek(~X2)2 a k
00 • l+j
I.nj d~ (B.30)
""" k=O ~x
Interchanging the order of summation and integration yields
(B.31)
As an example. the fourth order and fIfth order central moments of a random
variable whose distribution is approximated by a third order Gram-Charlier expansion can
be computed. Specifically. it is found that
(B.32)
(B.33)
APPENDIX C
PIERSON-MOSKOWITZ WAVE SPECTRUM
In offshore engineering, the wave elevation spectrum is frequently defmed by the

Pierson-Moskowitz spectrum. The analytical expression for the two-sided fonn of this
spectrum, which is used throughout this investigation, is given by the equation
(C.I)
where U w is the wind speed, g is the acceleration of gravity, and Ol is the frequency. The
constant parameters (X and p are commonly assumed to have the following values
(X = .0081 (C.2)
P = .74 (C.3)
For given (X and p, the wave spectrum given by equation (C.l) is completely
defmed by the wind speed. In offshore practice, however, it is more common to specify
the wave spectrum in tenns of the significant wave height, H s' and the mean period, Tavg.
The significant wave height is the average height of the highest one third of the waves in a
particular sea state. The mean period is related to the spectral moments. These two
parameters are related to the wind speed by the following equations
(C.4)
Olavg = 21t = .920 (1tP) 1/4 LU (C.5)

Tavg w
More details on these relationships can be found in references such as Chakrabarti(1987).

Using these equations, an alternative fonn of the Pierson-Moskowitz spectrum can be
159
produced in tenns of the significant wave height and the mean frequency, coavg •
Specifically,
4
Sllll(CO) = .111H; ~:I~ eXP[_.444(CO!V g j] (C.6)
As an example, Figure C.1 shows a one-sided fonn of the P-M spectrum for
Uw = 30 m/sec. Further, the significant wave height, mean period, and mean frequency
for this and other wind speeds based on the values of a and ~ given in equations (C.2) and
(C.3) are given in Table C.l.
Table C.1
Jlw mlsec Hli: m Tru sec rom radlsec

20 8.5 11.3 .557
22 10.3 12.4 .506
24 12.3 13.5 .464
26 14.4 14.7 .429
28 16.7 15.8 .398
30 19.2 16.9 .371
32 21.8 18.0 .348
34 24.6 19.2 .328
36 27.6 20.3 .310
160
120
100
N
.
~
U)
E
80
60
:§:
cl 40
20
0
0.0 0.2 0.4 0.6 0.8 1.0
CD radlsec
Figure C.l Pierson-Moskowitz Wave Spectrum

(Uw = 3OmJsec. H. = 19.2m. T""II = 16.9sec)
APPENDIX D
SIMULATION METHODS
D.1 Introduction
The simulation methods used to generate random wave force time histories are
reviewed in this section. The force time histories are used in a Newmark numerical
integration procedure to obtain simulated responses of an idealized TLP.
D.2 Linear Wave Simulation

In offshore engineering, random wave force time histories are typically computed
by transforming random wave elevation time histories based on suitable physical relations.
The sum of harmonics method described by Borgman( 1969) is a common method used to
simulate linear waves. Wave elevation time histories are generated from a two-sided target
spectrum, Sl1l1(OO), by the following summation
N
11 (x,t) = ~
)=1
11j COS(lCjX - OOjt + £j) (D.I)
where
(D.2)
(D.3)
The harmonic component with frequency OOj has an amplitude, 11j' which is related to the
area of a slice of the target wave spectrum at OOj. In this study, the frequency width of each
slice, L\oo, is the same, although this is not necessary. In addition, the phase angle, £j' is a
random variable with a uniform probability distribution from 0 to 2n. For a constant x
value, it follows that the random process, l1(X,t), is zero mean. Also by virtue of the
central limit theorem it approaches a gaussian probability distribution as the number of
harmonic components tends to infmity.
D.3 Linear Wave Force Simulation

A linear force, rl)(t), is obtained from the wave elevation by the following linear
transform written in the frequency domain as
162
00
fl>(t) = I H~>(ro) 'i1(X,ro) exp(irot) dro (0.4)
where H~I>(ro) is the linear force transfer function and 'i1(x,ro) is the Fourier transfonn of the
wave elevation. Force time histories are generated by taking the Fourier transfonn of the
hannonic summation in equation (D. 1) and substituting into equation (0.4) which yields
the following hannonic summation for the linear force
It is noted that the linear force depends on the horizontal position x. If x is time dependent,
as is the case in a compliant platfonn, the linear force actually has higher order behavior.
This is the so called body motion effect. By accounting for the time dependent position,
the force is no longer zero mean and is no longer gaussian, even though it is obtained by a
linear transfonnation. This equation is used to simulate the linear force in the numerical
integration computer program. As an option, the displaced position effect can be neglected
to isolate its contribution to the total response.
D.4 Drag Force Simulation

The nonlinear Morison equation with relative velocity effects is used to model the
drag force. The total drag force on a vertical cylinder is obtained by the following
integration over the submerged length of the cylinder
ll(X,t)
fD(t) = tpocCD I I U(z) + U(X,z,t) - xl(t)1 (U(z) + U(X,z,t) - Xl(t») dz (0.6)
-ds
The variables have been defmed previously in Chapter 5. The wave field water velocity,
u(x,z,t), is a random process which is simulated by an equation similar to equation (D.5).
Accounting for the displaced position of the TI..P and the integration to the free surface as
opposed to the mean free surface can readily be accomplished in the simulation. These
effects, however, are not readily modeled in the analytical procedure presented in this
study. They are neglected in the simul~tion to prevent obscuring the validity of the
163
analytical procedure. Thus, the simulated drag force is computed for a TLP in the
undisplaced position, and the integration is carried out only to the mean free surface. The
integration in equation (D.6) is performed by the trapezoidal rule.
D.S Quadratic Potential Force Simulation

The wave elevation and velocity head drift forces are simulated directly from the
second order Volterra series functional. The force is written in the frequency domain as
00
£<2)(t) = If H~)(COl,C02) fi(x,col) fi(x,co2) exp(ico1t) exp(ico2t) dcoldco2 (D.7)
where Hi2)(COl,C02) is the quadratic force transfer function.
A straightforward procedure to simulate the quadratic order force is to take the

Fourier transform of the harmonic summation in equation (D.I) and substitute it into
equation (D.7). This yields the following double summation for the quadratic force
N N
£<2)(t) = ~1 ~ t11m 11n {Re[Hi)(com,con)] COS[(lCm+lCn)X - (com+con)t + (Em+en)]
At a particular time, t, 4N2 harmonic computations are performed to evaluate the

force. IfNg is the number of steps in the time history, the total number of harmonic
computations is 4Ns N2. Since N g is in general several times greater than N, the number of
harmonic computations for one time history is of order N3. This proves to be too costly
for practical use. A much more efficient method makes use of digital FFf algorithms. In
order to use this more efficient method, a tradeoff must be made in that the displaced
position cannot be included in the quadratic force calculation. However, as with the drag
force, the displaced position effect in the quadratic force causes higher order effects which
are neglected in the analytical method anyway.
164
The digital procedure makes use of the Fourier series transform which along with
its inverse is defmed by the following equations
t J f(t) exp(-ikL\rot) dt
T
fk = (D.9)
f(t) =Lk~
- fk exp(ikL\rot) (0.10)
where T is the period over which f(t) repeats itself and fk is the Fourier series transform of
f(t).
The Fourier series transform of the quadratic force is obtained by substituting the
time domain form of the second order Volterra series which is written as
00
£<2}(t) = II h7}('tl,'t2) 11(t-'tl) 11(t-'t2) d'tld't2 (0.11)
into equation (0.9). Changing the order of integration and summation yields
~} = i
m=-oo
H7}(m.(\ro,(k-m).(\ro) 11m 11k-m (0.12)
where 11m is the Fourier series transform of 11(t). Using the sum of harmonics
representation for the wave elevation and using the same frequency increment for the
harmonic summation and the Fourier series it is readily shown that
r t11m exp( -im) m~O

11m = ~ m 0
(0.13)
*
l
~
11_m
Since the wave elevation is band limited at the frequency NL\CI) the summation in equation
(0.12) need only be carried out over a fmite number of points. The summation can be
rewritten as
k (2) N·k (2) __
L H f (m,k-m) 11m 11k-m + 2 m=l
m=O
L H f (k+m,-m) 11-m 11k+m (0.14)
165
where the symmetry properties of the quadratic transfer function have been utilized. Also
the presence of the frequency increment, Aco, in the arguments of the transfer function is
implied. The first tenn on the right hand side of equation (0.14) is the Fourier series of the
high frequency part of the quadratic force, the other tenn is the low frequency part.
The quadratic force time history is obtained from Ji> by applying the inverse Fourier
series transfonn given by equation (0.10). An FFr algorithm can be used to do this
efficiently. The time savings of this method compared to the straightforward method is
obvious. The number of operations in equation (0.13) is of order N, in equation (D.14) of
order N2, and for the FFr algorithm of order Ns 10g(Ns). Since the total number of
operations for one time history is obtained by addition of all these operations, this
procedure requires an order N2 number of operations. This is a substantial savings over
the direct method.
This digital procedure has been verified by averaging the quadratic force power
spectral density from 100 realizations and comparing it to the analytical power spectral
density. This has been done for quadratic surge force and pitch moment on the idealized
TLP for a P-M spectrum with Hs=1.36 m and Tavg= 4.51 sec. The results are shown in
Figures D.1 and D.2 and show good agreement in both the low frequency and high
frequency regions.
166
0.016
0.014
0.012 - - Analytical
&!
~ ••• •••• . Simu lation
OJ
Z 0.010
~
0.008
g 0.006
g (/)=:
0.004
0.002
0.0
0 2 3 4
ro rad/sec
Figure 0.1 Spectral Density of Quadratic

Wave ElevationNelocity Head Force on a TLP
(llw = 8 mlsec. Ii. = 1.36 m. Ts..g" 4.51 sec)
0.20
~
"'.-. 0.15 - - Analytical
~ •....... Simulation
Z
~
0.10
~
N'''';:
-rn- 0.05
0.0 E::::::~:::"""":==========--~..I...- ___ --1._ _ -==::J

o 2 3 4
ro rad/sec
Figure 0.2 Spectral Density of Quadratic

Wave ElevationNelocity Head Moment on a TLP
(llw .. 8 mlsec . Ii.. 1.36 m. TB"9" 4.51 sec)
REFERENCES
Abramowitz, M., and Stegun, I.A., 1972, Handbook of Mathematical Functions, Dover
Publications
Albrecht, H.G., Koenig, D., Kokkinowrachos, K., 1978, "Non-linear dynamic analysis
of tension-leg platforms for medium and greater depths", Proceedings of the 10th
Annual Offshore Technology Conference, Houston, Texas, OTC 3044, pp. 7-15
Angelides, D.C., Chen, C.-Y., Will, S.A., 1982, "Dynamics response of tension leg
platform", Proceedings of the 3rd Int. Conf on Behavior of Off-shore Structures,
MIT, Boston, MA, August, Vol. 2, pp. 100-120 .
Atalik, T.S., and Utku, T., 1976, "Stochastic linearization of multi-degree-of-freedom

non-linear systems", Earthquake Engng. Struct. Dynamics, Vol. 4, pp. 411-420
Beaman, 1.1., and Hedrick, I.K., 1981, "Improved statistical linearization for analysis and
control of nonlinear stochastic systems: Part I: An extended statistical linearization
technique", Journal of Dynamic Systems, Measurement, and Control, Vol. 102,
March, pp. 14-21
Beynet, P.A., Berman, M.Y., von Aschwege, I.T., 1978, "Motion, fatigue, and the
reliability characteristics of a vertically moored platform", Proceedings of the 10th
Annual Offshore Technology Conference, Houston, Texas, OTC 3304, pp. 2203-
2212
Booton, R.C., 1954, "The analysis of nonlinear control systems with random inputs", IRE
Trans. Circuit Theory, Vol. 1, pp. 9-18 .
Borgman, L.E., 1969,"Ocean wave simulation for engineering design", ASCE Journal of
the Waterways & Harbors Division, Nov., pp. 557-583
Botelho, D.L.R., Finnigan, T.D., Petrauskas, C., and Lui, S.V., 1984, "Model test
evaluation of a frequency-domain procedure for extreme surge response prediction of
tension leg platforms", Proceedings of the 16th Annual Offshore Technology
Conference, Houston, Texas, OTC 4658, pp. 105-112
Burns, G.E., 1983, "Calculating viscous drift of a tension leg platform", ASME
Proceedings of the 2nd International Offshore Mechanics and Arctic Engineering
Symposium, Houston, Texas, February, pp. 22-30
Caughey, T.K., 1963, "Equivalent linearization techniques", Journal of the Acoustical

Society of America, Vol 35, No. 11, Nov., pp. 1706-1711
Chakrabarti, S.K., 1975, "Second-order wave forces on large vertical cylinder", ASCE J.
Waterways, Harbours, and Coastal Engng. Div., Vol. 101, Proc. Paper 11476, pp.
311-317
Chakrabarti, S.K., 1984, "Steady drift force on vertical cylinder - viscous vs. potential",
Applied Ocean Research, Vol. 6, No.2, pp. 73-82
168
Chakrabarti, S.K., 1987, Hydrodynamics of Offshore Structures, Springer-Verlag, New

York, N.Y.
Chakrabarti, S.K., and Cotter, D.C., 1983, "First and second order interaction of waves
with large offshore structures", ASME Proceedings of the 2nd International Offshore
Mechanics and Arctic Engineering Symposium, Houston, Texas, February, pp. 171-
187
Crandall, S.H., 1985, "Non-gaussian closure techniques for stationary random vibration",
Internationallournal ofNon-Linear Mechanics, Vol. 20, No.1, pp. 1-8
Datta, T.K., and Jain, A.K., 1988, "Nonlinear surge response of a tension leg platfonn to
random wave forces", Engineering Structures, Vol. 10, July, pp. 204-210
Dean, R.G., and Dalrymple, R.A., 1984, Water Wave Mechanicsfor Engineers and
Scientists, Prentice-Hall, Englewood Cliffs, N.J.
deBoom, W.C., Pinkster, J.A., and Tan, S.G., 1983, "Motion and tether force prediction
for a deepwater tension leg platfonn", Proceedings of the 15th Annual Offshore
Technology Conference, Houston, Texas, OTC 4487, pp. 377-388
Denise, J-P.F., and Heaf, N.J., 1979, "A comparison between linear and non-linear
response of a proposed tension leg production platfonn", Proceedings of the 11 th
Annual Offshore Technology Conference, Houston, Texas, OTC 3555, pp. 1743-
1754
Gidwani, J.M., 1988, "Nonlinear dynamic analysis of deepwater compliant structures and
tension leg platfonns", ASME Proceedings of the 7th International Conference on
Offshore Mechanics and Arctic Engineering, Houston, Texas, February, pp. 295-
303
Guclmestad, O.T., and Conner, J.J., 1983, "Linearization methods and the influence of
current on the nonlinear hydrodynamic drag force", Applied Ocean Research, Vol. 5,
No.4, pp. 184-194
Faltinsen, 0.1., Van Hooff, R.W., Fylling, LJ., and Teigen, P.S., 1982, "Theoretical and
experimental investigations of tension leg platfonn behaviour", Proceedings of the
3rd Int. Con! on Behavior of Off-shore Structures, MIT, Boston, MA, August, Vol.
2, pp. 411-443
Ferretti, C., and Berta, M., 1980, "Viscous effect contribution to the drift forces on
floating structures", Proceedings of the International Symposium on Ocean
Engineering - Ship Handling, Gothenburg, Sweden, No.9, pp. 9:1-10
Finnigan, T.D., Petrauskas, C., and Botelho, D.L.R., 1984, "Time-domain model for
TLP surge responSe in extreme sea states", Proceedings of the 16th Annual Offshore
169
Herfjord, K., and Nielsen, F.G., 1986, "Non-linear wave forces on a fixed vertical
cylinder due to the sum frequency of waves in irregular seas", Applied Ocean
Research, Vol. 8, No.1, pp. 8-21
Hooft, J.P., 1982, Advanced Dynamics of Marine Structures, John Wiley, New York
Hu, S.-L.J., and Dixit, S., 1988, "Non-gaussian dynamic response to drag forces",
ASME Proceedings of the 7th International Conference on Offshore Mechanics and
Arctic Engineering, Houston, Texas, February, pp. 109-116
Isaacson, M. de St. Q., 1979, "Nonlinear inertia forces on bodies", ASCE J. Waterways,
Harbours, and Coastal Engng. Div., Vol. 105, Proc. Paper 14743, pp. 213~227
Iwan, W.D., and Yang, I.M., 1972, "Application of statistical linearization techniques to
nonlinear multidegree-of-freedom systems", Journal of Applied Mechanics, Vol. 39,
June, pp. 545-550
Jefferys, E.R, and Patel, M.H., 1981, "Dynamic analysis models of the tension leg
platform", Proceedings of the 13th Annual Offshore Technology Conference,
Houston, Texas, OTC 4075, pp. 99-107
Johnson, N.L., and Kotz, S., 1972, Distributions in Statistics: Continuous Multivariate
Distributions, John Wiley, New York, pp. 10-12
Kirk, C.L., and Etok, E.U., 1979, "Dynamic response of tethered production platform in a
random sea state", Proceedings of the 2nd Int. Con! on Behavior of Off-shore
Structures, Imperial College, London, England, pp. 139-163
Kitami, E., Ninomiya, K., Katayama, M., and Unoki, K., 1982, "Response
characteristics of tension leg platform with mechanical damping system in waves",
Proceedings of the 14th Annual Offshore Technology Conference, Houston, Texas,
OTC 4393, pp. 181-198
Kokkinowrachos, K., and Thanos, I., 1989, "Second-order forces on arbitrary vertical
axisymmetric bodies", ASME Proceedings of the 8th International Offshore
Mechanics and Arctic Engineering Symposium, The Hague, March, pp. 79-87
Kobayashi, M., Shimada, K., Fujihara, T., 1986, "Study on dynamic responses of a TLP
in waves", ASMEJournal of Offshore Mechanics and Arctic Engineering, Vol. 109,
pp.61-66
Kryloff, N., and Bogoliubov, N., 1947, Introduction to Nonlinear Mechanics, Princeton
University Press, Princeton, N.J.
Langley, RS., 1987a, "Second order frequency domain analysis of moored vessels",
Applied Ocean Research, Vol. 9, No.1, pp. 7-18
Langley, RS., 1987b, "A statistical analysis of low frequency second order forces and
motions", Applied Ocean Research, Vol. 9, No.3, pp. 163-170
170
Lundgren, H., Sand, S.E., and Kirkegaard, J., 1982, "Drift forces and damping in natural
sea states - a critical review of the hydrodynamics of floating structures", Proceedings
of the 3rd Int. Conf. on Behavior of Off-shore Structures, MIT, Boston, MA,
August, Vol. 2, pp. 592-607
Lyons, G.J., Patel, M.H., Sarohia, S., and Hartnup, G.C., 1983, "Theory and model test
data for tether forces on tensioned buoyant platforms", Proceedings of the 15th
Annual Offshore Technology Conference, Houston, Texas, OTC 4643, pp. 533-544
MacCamy, RC., and Fuchs, RA., 1954, "Wave forces on piles: a diffraction theory",
Technical MenwrandumNo. 69, Beach Erosion Board
Madsen, O.S., 1986, "Hydrodynamic force on a circular cylinder", Applied Ocean

Research, Vol. 8, No.3, pp. 151-165
Madsen, A.H., and Jensen, I.I., 1989, "On a non-linear stochastic wave theory and
Morison's formula", ASME Proceedings of the 8th International Conference on
Offshore Mechanics and Arctic Engineering, The Hague, March, pp. 45-51
Malhotra, A.K., and Penzien, J., 1970, "Response of offshore structures to random wave
forces", ASCE Journal of the Structural Division, Vol. 96, No. ST1O, pp. 2155-
2173
Marthinsen, T., 1989, "Hydrodynamics in TLP design", ASME Proceedings of the 8th
International Conference on Offshore Mechanics and Arctic Engineering, The Hague,
March, pp. 127-133
Mercier, J.A, Leverette, S.J., and Bliault, A.L., 1982, "Evaluation of Hutton TLP
response to environmental loads", Proceedings of the 14th Annual Offshore
Moe, G., and Verley, RL.P., 1980, "Hydrodynamic damping of offshore structures in
waves and currents", Proceedings of the 12th Annual Offshore Technology
Molin, B., 1979, "Second-order diffraction loads upon three-dimensional bodies",

Applied Ocean Research, Vol. 1, pp. 197-202
Naess, A.,1986, "The statistical distribution of second-order slowly-varying forces and

motions", Applied Ocean Research, Vol. 8, No.2, pp. 110-118
Natvig, B.J., and Pendered, I.W., 1977, "Nonlinear motion response of floating
structures to wave excitation", Proceedings of the 9th Annual Offshore Technology
Newman, J.N., 1974, "Second order slowly varying forces in irregular waves", Proc. Int.
Symp. on Dynamics of Marine Vehicles and Offshore Structures in Waves,
University College London, London, pp. 182-186
Nigam, N.C., 1983, Introduction to Random Vibrations, MIT Press, Boston, Mass.
171
Nielsen, F.G., and Herfjord K., 1985, "The importance of non-linear wave forces to
fatigue of deepwater structures", Proceedings o/the 17th Annual Offshore
Olagnon, M., Prevosto, M., and Joubert, P., 1988, "Nonlinear spectral computation of the
dynamic response of a single cylinder", ASME Journal 0/ Offshore Mechanics and
Arctic Engineering, Vol. 110, August, pp. 278-281
Paulling, J.R., and Horton, B.B., 1970, "Analysis of the tension leg stable platform",
Proceedings o/the 2nd Annual Offshore Technology Conference, Houston, Texas,
OTC 1263, pp. n.380-390
Petrauskas, C., and Liu, S.V., 1987, "Springing force response of a tension leg platform",
Proceedings o/the 19th Annual Offshore Technology Conference, Houston, Texas,
OTC 5458, pp. 333-342
Pijfers, J.G.L., and Brink, A.W., 1977, "Calculated drift forces of two semisubmersible
platform types in regular and irregular waves", Proceedings 0/ the 9th Annual
Offshore Technology Conference, Houston, Texas, OTC 2977, pp. 155-164
Pinkster, J.A., 1979, "Mean and low frequency wave drifting forces on floating
structures", Ocean Engineering, Vol. 6, pp. 593-615
Pinkster, J.A., 1980, "Low frequency second order wave exciting forces on floating
structures", Publication No. 650 of the Netherlands Ship Model Basin
Rahman, M., 1984, "Wave diffraction by large offshore structures: an exact second order
theory", Applied Ocean Research, Vol. 6, No.2, pp. 91-98
Rahman, M., and Chakravartty, I.C., 1981, "Hydrodynamic loading calculations for
offshore structures", SIAM J. Appl. Math., Vol. 41, No.3, pp. 445-458
Rahman, M., and Heaps, H.S., 1983, "Wave forces on offshore structures: nonlinear
wave diffraction by large cylinders", J. Physical Oceanography, Dec., Vol. 13,
pp. 2225-2235
Roberts, J.B., and Spanos, P.D., 1989, Random Vibrations and Statistical Linearization,
John Wiley, New York
Rugh, W J., 1981, Nonlinear System Theory - The Volterra/Weiner Approach, The Johns
Hopkins University Press
Sarpkaya, T., and Isaacson, M., 1981, Mechanics o/Wave Forces on Offshore Structures,
van Nostrand Reinhold, New York
Salvesen, N., von Kerczek, C.H., Yue, D.K., Stem, F., 1982, "Computations of
nonlinear surge motions of tension leg platforms", Proceedings 0/ the 14th Annual
Offshore Technology Conference, Houston, Texas, OTC 4394, pp. 199-215
Schetzen, M., 1980, The Volterra and Wiener Theories o/Nonlinear Systems, John Wiley,
New York
172
Spanos, P-T.D., 1980, "Fonnulation of stochastic linearization for symmetric or

asymmetric m.d.o.f. nonlinear systems", Journal of Applied Mechanics, Vol. 47, pp.
209-211
Spanos, P.D., and Chen, T.W., 1981, "Random response to flow-induced forces", ASCE
Journal of the Engineering Mechanics Division, Vol. 107, No. EM6, December, pp.
1173-1190
Spanos, P-T.D., 1981a, "Monte Carlo simulations of responses of a non-symmetric
dynamic system to random excitations", Computers and Structures, Vol. 13, pp.
371-376
Spanos, P-T.D., 1981b, "Stochastic linearization in structural dynamics", Applied

Mechanics Reviews, Vol. 34, pp. 1-8
Spanos, P.D., and Agarwal, V.K., 1984, "Response of a simple tension leg platfonn
model to wave forces calculated at displaced position", ASME J. Energy Resources
Technology, Vol. 106, December, pp. 437-443
Stansberg, C.T., 1983, "Statistical analysis of slow-drift responses", Journal of Energy

Resources Technology, Vol. 105, pp. 188-197
Tan, S.G., and deBoom, W.C., 1981, "The wave induced motions of a tension leg
platfonn in deep water", Proceedings of the 13th Annual Offshore Technology
Taylor, R.E., and Hung, S.M., 1987, "Second order diffraction forces on a vertical
cylinder in regular waves", Applied Ocean Research, Vol. 9, No.1, pp. 19-29
Thiagarajan, K., and Baddour, R.E., 1989, "Higher order wave loading on fixed, slender,
surface piercing, rigid cylinders", ASME Proceedings of the 8th International
Offshore Mechanics and Arctic Engineering Symposium, The Hague, March, pp.
213-220
Vassilopoulos, L.A., 1967, "The application of statistical theory of nonlinear systems to

ship motion perfonnance in random seas", International Ship Building Progress,
Vol. 14, No. 150, pp. 54-65
Yamanouchi, Y., 1974, "Ship's behaviour on ocean waves as a stochastic process", Proc.
Int. Symp. on Dynamics of Marine Vehicles and Offshore Structures in Waves,
University College London, London, pp. 167-181
Yoshida, K., Yoneya, T., Oka, N., Ozaki, M., 1981, "Motions and leg tensions of tension
leg platfonns", Proceedings of the 13th Annual Offshore Technology Conference,
Houston, Texas, OTC 4073, pp. 75-87
Edited by C.A. Brebbia and S.A. Orszag
Vol. 1: J. C. F. Telles, Vol. 11: M. B. Beck

The Boundary Element Method Water Quality Management:
Applied to Inelastic Problems A Review of the Development and
IX, 243 pages. 1983. Application of Mathematical Models
VIII, 108 pages. 1985.
Vol. 2: Bernard Amadei,
Rock Anisotropy and Vol. 12: G. Walker, J. R. Senft
the Theory of Stress Measurements Free Piston Stirling Engines
XVIII, 479 pages. 1983. XIV, 286 pages. 1985.
Vol. 3: Computational Aspects of Vol. 13: Nonlinear Dynamics

Penetration Mechanics of Transcritical Flows
Proceedings of the Army Research Proceedings of a DFVLR International
Office Workshop on Computational Colloquium, Bonn, Germany, March 26, 1984
Aspects of Penetration Mechanics VI, 203 pages. 1985.
held at the Ballistic Research Laboratory
at Aberdeen Proving Ground, Maryland, Vol. 14: A. A. Bakr
27-29 April, 1982 The Boundary Integral
Edited by J. Chandra and J.E. Flaherty Equation Method in Axisymmetric
VII, 221 pages. 1983. Stress Analysis Problems
XI, 213 pages. 1986.
Vol.4: W.S. Venturini
Boundary Element Method in Geomechanics Vol. 15: I. Kinnmark
VIII, 246 pages. 1983. The Shallow Water Wave
Equation: Formulation,
Vol. 5: Madassar Manzoor Analysis and Application
Heat Flow Through Extended XXIII, 187 pages, 1986.
Surface Heat Exchangers
VII, 286 pages. 1984. Vol. 16: G. J. Creus
Viscoelasticity - Basic
Vol. 6: Myron B. Allen III Theory and Applications
Collocation Techniques for Modeling to Concrete Structures
Compositional Flows in Oil Reservoirs VII, 161 pages. 1986.
VI, 210 pages. 1984.
Vol. 17: S. M. Baxter
Vol. 7: Derek B. Ingham, C. L. Mortey
Mark A. Kelmanson Angular Distribution
Boundary Integral Equation Analysis in Acoustics
Analyses of Singular, Potential, VII, 202 pages. 1986.
and Biharmonic Problems
IV, 173 pages. 1984. Vol. 18: N. C. Markatos,
D. G. Tatchell, M. Cross, N. Rhodes
Vol. 8: Linda M. Abriola Numerical Simulation of Fluid Flow
Multiphase Migration of Organic .and Heat/Mass Tranfer Processes
Compounds in a Porous Medium VIII, 482 pages. 1986.
A Mathematical Model
VIII, 232 pages. 1984. Vol. 19: Finite Rotations
in Structural Mechanics
Vol. 9: Theodore V. Hromadka II Proceedings of the Euromech
The Complex Variable Boundary Colloquium 197, Jablonna 1985
Element Method VII, 385 pages. 1986.
XI, 243 pages. 1984.
. Vol. 20: S. M. Niku
Vol. 10: C. A. Brebbia, H. Tottenham, Finite Element Analysis
G. B. Warburton, J. M. Wilson, R. R. Wilson of Hyperbolic Cooling Towers
Vibrations of Engineering Structures VIII, 216 pages. 1986.
VI, 300 pages. 1985.
Edited by C.A. Brebbia and S.A. Orszag Vol. 30: R. Dolezal
Simulation of Large State Variations
Vol. 21: B. F. Spencer, Jr. in Steam Power Plants
Reliability of Randomly Dynamics of Large Scale Systems
Excited Hysteretic Structures X, 110 pages. 1987.
XIII, 138 pages. 1986.
Vol. 31: Y. K. Lin, G.I. Schueller (Eds.)
Vol. 22: A. Gupta, R. P. Singh Stochastic Structural Mechanics
Fatigue Behaviour U.S.-Austria Joint Seminar, May 4-5,1987
of Offshore Structures Boca Raton, Florida, USA
XXI, 299 pages. 1986. XI, 507 pages. 1987.
Vol. 23: P. Hagedorn, K. Kelkel, Vol. 32: Y. K. Lin, R. Minai (Eds.)

J. Wallaschek Stochastic Approaches
Vibrations and Impedances in Earthquake Engineering
of Rectangular Plates U.S.-Japan Joint Seminar, May 6-7, 1987
with Free Boundaries Boca Raton, Florida, USA
V, 152 pages. 1986. XI, 457 pages. 1987.
Vol. 24: Supercomputers Vol. 33: P. Thoft-Christensen (Editor)

and Fluid Dynamics Reliability and Optimization
Proceedings of the First of Structural Systems
Nobeyama Workshop Proceedings of the FirsllFIP WG 7.5
September 3-6, 1985 Working Conference
VIII, 200 pages. 1986. Aalborg, Denmark, May 6-8, 1987
VIII, 458 pages. 1987.
Vol. 25: B. Hederson-Sellers
Modeling of Plume Rise Vol. 34: M. B. Allen III, G. A. Behie,
and Dispersion - J. A. Trangenslein
The University of Salford Multiphase Flow in Porous Media
Model: U. S. P. R. Mechanics, Mathematics, and Numerics
VIII, 113 pages. 1987. IV, 312 pages. 1988.
Vol. 26: Shell and Spatial Structures: Vol. 35: W. Tang

Computational Aspects A New Transformation
Proceeding of the International Symposium Approach in BEM
July 1986, Leuven, Belgium A Generalized Approach for
Edited by G. De Roeck, A. Samartin Quiroga, Transforming Domain Integrals
M. Van Laethem and E. Backx VI, 210 pages. 1988.
VII, 486 pages. 1987.
Vol. 36: R. H. Mendez, S. A. Orszag
Vol. 27: Th. V. Hromadka, Ch.-Ch. Yen Japanese Supercomputing
G. F. Pinder Architecture, Algorithms, and Applications
The Best Approximation Method IV, 160 pages. 1988.
An Introduction
XIII, 168 pages. 1987. Vol. 37: J. N. Reddy, C. S. Krishnamoorthy,
K. N. Seetharamu (Eds.)
Vol. 28: Refined Dynamical Theories Finite Element Analysis
of Beams, Plates and Shells and for Engineering Design
Their Applications XIV, 869 pages. 1988.
Proceedings of the Euromech-Colloquim 219
Edited by I. Elishakoff and H.lrretier Vol. 38: S. J. Dunnett, D. B. Ingham
IX, 436 pages. 1987. The Mathematics of Blunt Body Sampling
VIII, 213 pages. 1988
Vol. 29: G. Menges, N. Hovelmanns,
E. Baur (Eds.) Vol. 39: S.L. Koh, CG. Speziale (Eds.)
Expert Systems in Production Engineering Recent Advances in Engineering Science
Proceedings of the International Workshop A Symposium dedicated to A. Cemal Eringen
Spa, Belgium, August 18-22, 1986 June 20-22, 1988, Berkeley, California
IV, 245 pages. 1987. XVIII, 268 pages. 1989
Edited by C.A. Brebbia and S.A. Orszag
Vol. 40: R Borghi, S. N. B. M

urthy (Eds.) Vol. 49 : J. P.Boyd
Turbulent Reactive Flows Chebyshev & Fourier Spectral Methods
VIII, 950 pages. 1989 XVI, 798 pages. 1989
Vol. 41: w.J. Lick Vol. 50: L. Chi bani

Difference Equations Optimum Design of Structures
from Differential Equations VIII, 154 pages. 1989
X, 282 pages. 1989
Vol. 51: G. Karami
Vol. 42: HA Eschenauer, G. Thierauf (Eds.) A Boundary Element Method for
Discretization Methods Two-Dimensional Contact Problems
and Structural Optimization - VII, 243 pages. 1989
Procedures and Applications
Proceedings of a GAMM-Seminar Vol. 52 : Y. S.Jiang
October 5- 7, 1988, Siegen. FRG Slope Analysis Using
XV, 360 pages. 1989 Boundary Elements
IV, 176 pages. 1989
Vol. 43: C. C. Chao, S.A. Orszag, W. Shyy (Eds.)
Recent Advances in Computational Vol. 53: A S Jovanovic,
.
Fluid Dynamics K. F. Kussmaul, A C .Lucia.
Proceedings of the US/ROC (Taiwan) Joint P. P.Bonissone (Eds.)
Workshop in Recent Advances in Expert Systems in Structural
Computational Fluid Dynamics Safety Assessment
V, 529 pages. 1989 X, 493 pages. 1989
Vol. 44 R. S.Edgar Vol. 54: 1. J. Mueller (Ed.)

Field Analysis and Low Reynolds Number
Potential Theory Aerodynamics
XII. 696 pages. 1989 V, 446 pages. 1989
Vol. 45: M. Gad-el-Hak (Ed.) Vol. 55: K. Kitagawa

Advances in Fluid Mechanics Boundary Element Analysis
Measurements of Viscous Flow
VI I, 606 pages. 1989 VII, 136 pages. 1990
Vol. 46 : M.Gad-el- Hak (Ed.) Vol. 56: A. A. Aldama

Frontiers in Experimental Filtering Techniques for
Fluid Mechanics Turbulent Flow Simulation
VI . 532 pages. 1989 VIII , 397 pages. 1990
Vo l. 47 : H W Bergmann (Ed.) Vol. 57: M. G.Donley, P. D, Spanos

Optimization: Methods and Applications, Dynamic Analysis of Non- Linear
Possibilities and Limitations Structures by the Method of
Proceedings of an International Seminar Statistical Quadratization
Organized by Deutsche Forschungsanstalt fur VII, 186 pages. 1990
Luft - und Raumfahrt (DLR). Bonn, June 1989
IV, 155 pages. 1989
Vol. 48: P Thoft-Christensen (Ed.)

Reliability and Optimization
of Structural Systems '88
Proceedings of the 2nd IFIPWG 7.5 Conference
London, UK, September 26- 28. 1988
VII . 434 pages. 1989

(Lecture Notes in Engineering 57) M. G. Donley, P. D. Spanos (Auth.) - Dynamic Analysis of Non-Linear Structures by The Method of Statistical Quadratization-Springer-Verlag Berlin Heidelberg (1990)

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

(Lecture Notes in Engineering 57) M. G. Donley, P. D. Spanos (Auth.) - Dynamic Analysis of Non-Linear Structures by The Method of Statistical Quadratization-Springer-Verlag Berlin Heidelberg (1990)

Uploaded by

Copyright:

Available Formats

Lecture Notes in Engineering

The Editorial Board

Consulting Editors Materials Science and Computer Simulation:

Earthquake Engineering: Fluid Mechanics:

Geotechnical Engineering and Geomechanics: Structural Engineering:

ISBN-13: 978-3-540-52743-5 e-ISBN-13: 978-3-642-46715-8

2161/3020-543210 Printed on acid-free paper.

CHAPTER 1: INTRODUCTION ...................................................... 1

CHAPTER 2: EQUIVALENT STOCHASTIC QUADRATIZATION FOR

CHAPTER 3: EQUIVALENT STOCHASTIC QUADRATIZATION FOR

4.4 Linear Velocity Potential.... ....... .... ........... ..... ..... 70

CHAPTER 5: EQUIVALENT STOCHASTIC QUADRATIZATION FOR

CHAPTER 6: STOCHASTIC RESPONSE OF A TENSION LEG PLATFORM

CHAPTER 7: SUMMARY AND CONCLUSIONS................................. 147

APPENDIX A: GRAM-CHARLIER COEFFICIENTS .............................. 151

APPENDIXB: EVALUATION OF EXPECTATIONS .............................. 153

APPENDIX C: PIERSON-MOSKOWITZ WAVE SPECTRUM ................... 158

APPENDIX D: SIMULATION METHODS ........................................... 161

REFERENCES: ........................................................................... 167

Another source of nonlinearity is in the wave induced potential forces. These

1.2 Aim of Study

1.3 TLP Model

1.4 Environmental Loads

The following review of the methods to model hydrodynamic forces is separated

1.4.1 Methods to Compute Viscous Forces

polynomial expansion. Quadratic order forces are obtained by including polynomials up to

1.4.2 Methods to Compute Potential Forces

I.S Literature Review of TLP Analyses

1.6 Scope of Study

The significance of quadratic order terms is found to depend on the frequency

2.2 Analytical Method Formulation

rnX + cx + kx + g(x,x) = f(t) (2.1)

f(t) = /If + f hc('t) w(t-'t) d't (2.2)

where wet) is a zero-mean, gaussian, white-noise process. That is <w(t)w(t+'t» = 2n/)(r),

g(x) = 'Y 11 +xl(1 +x)

Figure 2.1 Symmetry of Nonlinearity

Substituting equation (2.2) into equation (2.3) yields

HIm) = f hlt) exp( -iOl"c) d't (2.5)

g(x,x) :# - g(-x,-x) (2.6)

x(t) = Ilx + x(t) (2.7)

The response, x(t), will be obtained by solving the response of an "equivalent"

The quadratization coefficients are computed such that

where the error, E, is defmed by the equation

E == kJlx+ g(~x+xJ) - ~f - (XIx - (X2k - (X3(x2-<x2» - (Xi k2 -<k2» - (Xsxk (2.12)

and the superscript, T, denotes transposition.

~X2 0 ~x3 ~xk2 0 (Xl <gb

x(t) = L AjX(j)(t) (2.21)

Hr)(CO I, C0 2) = If hT('tI,'t2) exp(-icoI'tI) exp(-ico2't2) d'tld't2 (2.29)

2.3 Derivation of Linear and Quadratic Transfer Functions

transfer functions by the relations

H (~\oo) = iooH~)(oo) (2.33)

i(t) = Jh( t) l(t-t) dt (2.37)

2.4 R~sponse Probability Distribution

The function <I>(xJ) is given by the equation

in which the state space vector 5 is defmed by the equation

1lx4= 3(l!x2)2 (2.55)

1!i4= 3(l!i2)2 (2.56)

1lx2i2 = J.l,x2 l!i2 (2.57)

2.5 Response Spectral Density

Sxx(ro) = 1 H(ro) 12 SU(ro) (2.59)

2.6 Solution Procedure

2.7 Example of Application

x + ~i + x + -yla + il(a + i) = f(t) (2.61)

o~~~=-~ ~====~

o ~ ~ ~ ~ ~