A KNOWLEDGE BASE APPROACH TO THE

DESIGN OF TENSION LEG PLATFORMS

o'By Oriol R. Rijken and John M. Niedzwecki

,Cla
rji r
OFFSHORE
TECHNOLOGY
RESEARCH
CENTER

A National Science Foundation Engineering Research Center

12/91-A-24-100

For more informcdion contact:

Offshore Technology Research Center

Texas A&M University
1200 Mariner Drive
College Station, TX 77845-3400
(409) 845-6000

or,
Center for Offshore Technology
The University of Texas at Austin
WRW 217
Austin, TX 77812-1085
(512) 471-3753
 MINIM UNNERSITST
Laboratorlum vow
Scheepshydromethanica
Archief
ABSTRACT &Mohr3g 2, 2628 CD Delft
id: 018 - 786873 - Far. 015- 781338

This study examines the preliminary design of a tension leg platform

(TLP), using a knowledge base approach. A TLP is a floating deep water
compliant structure designed for offshore hydrocarbon production. Only a few
TLP's have actually been built, but these platforms have demonstrated the
feasibility of this new concept. Important field experience about their response
behavior due to the environment is not generally available, although many
technical articles on analysis methods have been published. The use of an expert
system approach can be used to compensate for this lack of information through
the introduction of expert knowledge and the use of simplified mathematical or
empirical models.
The preliminary design model developed in this study focuses on general
aspects of the TLP platform configuration and its tethers. The tether
characteristics provide three constraints. These constraints and the requirement
of minimum cost are combined in a strategy to identify the optimum TLP
configuration for a given offshore site. The final configuration can be plotted and
its engineering details shown in a spreadsheet format. This knowledge base was
used to study four different TLP configurations which are comparable with either
existing TLP's or TLP's which will be installed in the near future. These
simulations indicate that it is possible to approach practical design configurations
using this methodology.
 "C) "1.7 -"J..IZI:ji*i4

f'"7 TABLE OF CONTENTS

Page
ABSTRACT

TABLE OF CONTENTS rn

LIST OF FIGURES

LIST OF TABLES vi

1 INTRODUCTION
1.1 Tension leg platform 1

1.2 Design strategies 4

1.3 Computer oriented approach 4

2 KNOWLEDGE BASE FOR UP DESIGN 6

2.1 Basics of KB design 6
22 Expert system application 8
2.3 Knowledge, software and hardware requirements 9

3 TLP DESIGN THEORY

3.1 TLP constrained optimization 12
3.2 Expansion of model 21
3.2.1 Buoyancy 23
3.22 Mass 25
32.3 Pontoon diameter 26
3.2.4 Added mass 26
3.3 Assessment of expert system's knowledge 29
 Page
4 IMPLEMENTATION FOR EXPERT SYSTEMS 35
4.1 Knowledge inventory 36
4.1.1 Knowledge smoother 36
4.12 Implementation knowledge 39
42 Knowledge base description 41

ILLUSTRATIVE EXAMPLES 47
5.1 TLP in 150 m of water depth 47
5.2 TLP in 536 m of water depth 47
5.3 UP in 870 in of water depth 48
5.4 MP in 1500 m of water depth 48

6 CONCLUSIONS 53

REFERENCES 57

APPEND! C A NOMENCLATURE 59

APPENDIX B LIST OF OBJECTS 62

APPENDIX C EXPERT SYSTEMS SUGGESTIONS 63

APPENDIX D MP DESCRIPTION SHEET 64

APPENDIX E USING THE KNOWLEDGE BASE 69

iv
 LIST OF FIGURES

Figure Page
1 Tension Leg Platform Definitions 2
2 Tension Leg Platform Configuration 14
3 Constraints and Optimum 20
4 TLP Column Configuration 24
5 Added Mass of Heaving Column 28
6 Different Approach to Added Mass Calculation , 30
7 .NEXPERT Parameter Selection 38
8 Overall Flow Chart of Knowledge Base 40
9 Rule Description 42
10 Search Path for Modification 145
11 Nexpert Screen Boolean Input ,66
12 Nexpert Screen Numerical Input 66
13 Nexpert Screen Selection Input 68
14 Nexpert Screen Start Application 68
15 Nexpert Screen Main Screen of Knowledge Base 69
 LIST OF TABLES

Table Page
1 Description of TLP (Nordgren, 1989) 22
2 TLP in 150 m 49
3 TLP in 536 m 50
4 TLP in 870 m 51
5 TLP in 1500 m 52

vi
 1 INTRODUCTION

1.1 Tension leg platform

A Tension Leg Platform (TLP), as illustrated in Fig. 1 is a floating structure
which is designed to produce hydrocarbons from deep water offshore petroleum
fields. The first TLP was constructed for Conoco (Offshore, 1985) and was
eventually installed in the Hutton Field in the British sector of the North Sea in
the early eighties. The Hutton TLP demonstrated the feasibility of this type of
compliant offshore structure even though it was located in a water depth of 150
m. Conceptually a TLP is designed to be an excessively buoyant floating
platform, which is connected to the seafloor by vertical mooring lines called
tendons or tethers. These tethers are designed to adjust the draft of the TLP and
to limit the amount of sway and surge.
The objective of building an offshore structure is to create a cost efficient
work area away from shore. In this study only tension leg platforms are
considered for deep water offshore field development. The objective of this study
is to develop a computer tool for use in preliminary design using the latest
computer technology. Of particular interest is the general TLP configuration and
their associated cost estimate. One approach is to minimize the cost of building
the TLP with regards to key parameters, each offshore site has different
environmental parameters and requirements, an optimization procedure has to
be performed for each location. Ideally one single strategy can be used to handle
the optimization for different offshore sites. To be meaningful, the optimization
procedure must be subjected to several constraints. The basic constraints in this
study are sufficient buoyancy, tether tension, tether stress, heave frequency of the
platform and sufficient clearance between the highest wave crest and the bottom
of the deck. Several parameters in the optimization are site related and can only
be modified within a very small range. These parameters are, for instance, the
maximum wave height and the water depth. The type of petroleum reservoir, the

1
Teas ion Pi form

deck

column
brace 4diameter

column draft

hare column
height length

pontoon flare
'diaineter
tether
waterdepth
riser

sub sea base

-
. _

Fig. 1. Tension Leg Platform Definitions

2
maximum wave height and the water depth. The type of petroleum reservoir, the
type of petroleum product and the distance to further processing facilities, the
size of the living quarters and the drilling equipment dictate the amount of
processing equipment. These are site specific parameters, as is the top tension
in the riser which depends primarily on the water depth. The other type of
parameters can be called design parameters and these parameters will change
throughout the preliminary design process.
One of the problems which arises when trying to develop a 1'LP design
is the lack of coherence between the available knowledge in the open literature.
For example a lot is known about the behavior of floating structures and tethers
but, there is hardly any information available about the behavior of a floating
structure with tethers. This lack of information can not be compensated by field
data because there are only a few TLP's in operation and data on their response
in proprietary.
A simple yet robust methodology will be required in order to develop an
optimized TLP configuration based upon our sometimes incomplete knowledge
about various details of TLP design. It will be expected to estimate a variety of
parameters and guide the user to an optimum solution, it should be noted further
that the methodology should be tolerant of users with a minimum knowledge
about TLP's. The resulting solution should be regarded as a different approach
to obtain a starting point in the design spiral (American Petroleum Institute,
1987).

3
1.2 Design strategies
Designing any structure always requires an iterative design procedure. The
steps in this procedure are well defined when a similar structure has been
previously designed. Guidelines or design codes do exist when several similar
structures have been built before. It is rather easy to design a structure when
these guidelines exist. One studies the steps in the previous design and applies
them to their specific problem and it is generally not difficult to develop a similar
structure. The guidelines are designed to avoid known problems or tackle
problems efficiently and make the structure comply with existing norms and
regulations. The significant alteration of an existing structural design into a very
different structure makes the design procedure a lot more complex. The
complexity arises from the fact that one may not know how the various
parameters interact with each other.
Existing guidelines or building codes are completely inadequate for the
development of a new type of structure. A different approach is required to
develop a new design. One must acknowledge that for this kind of structure it is
impossible to give an exact description. One can only guess about the design or
describe it more quantitatively. An iterative procedure is again required to
develop a preliminary description of the structure. The typical approach to the
iteration procedure is to estimate the values of key design parameters, then
calculate the structure's characteristics, followed by a check and modification of
several of the parameters. The iteration procedure stops when the designer feels
that the system is adequate.

1.3 Computer oriented approach

There are many calculations involved in the preliminary design of a TLP
and they can be preformed in a variety of computer environments. A common
approach would be to write your own computer program using a higher level

4
language like FORTRAN, PASCAL or 'C'. Mother possibility would be to use
a spreadsheet which is becoming a common engineering approach. However, the
use of a spreadsheet for optimization requires a significant amount of numerical
computation and knowledge about the problem by the user. In order to find an
acceptable solution the user has to interpret the spreadsheet computations.
Another possibility is the use of a computer environment which captures not only
the content of the spreadsheet but also captures the knowledge of the
experienced engineer. This type of approach is called Expert System Approach
or in it's most general context a Knowledge Base System (ICBS) approach. The
combined knowledge used in the spreadsheet and the experienced user's
knowledge is called a Knowledge Base (KB). Nowadays there are softwre
environments available whicl3 allow for an expert system approach to efficiently
build a knowledge base.
 2 KNOWLEDGE BASE FOR TLP DESIGN
2.1 Basics of KB design
The strategy behind problem solving with an expert system approach is a
very powerful strategy which can be very complex and sometimes abstract. One
of the first facts one must realize when taking this approach is that the solution
strategy which is represented in the expert system is very subjective and reflects
the interpretation of the person who designed it. However, as one becomes more
adept at using the approach the subjective aspects can be minimal One of the
most powerful aspects of the expert system approach is the fact that loosely
organized information can be used (Coyne, 1990). Typically gaps exist in
engineering knowledge and the ICBS strategy can compensate for these gaps.
The first form of analysis in developing an expert system deals with the
interpretation of the rules within the problem domain. For instance in the case
of TLP modelling the rules describe how the draft and mass of a TLP are
calculated or that there are no braces when there are no pontoons. The result
of this analysis is called the rule based part of an expert system.
The second type of analysis deals with the analysis of the individual parts
which make up the information. One tries to organize these parts into groups
which have identical features. Each separate part is called an object. Each
feature is called a property. Object in the TLP design is for instance water with
the property depth, wave with the properties height, frequency and period. Another
objects is column with the properties shape, length, draft and diameter. When
objects have identical features they are grouped into a class. Classes can be
grouped too in other classes. An example of a class in the design of a TLP is the
class of the construction materials. A TLP can be made out of steel or concrete.
Hence there are two objects in the class Construction Materials : Steel and
Concrete. Each object has the properties density, stress, maximum stress and
Young's Modulus.

6
 The idea of using an expert system approach implies a good mapping of
the available knowledge into the ICBS approach. A key factor to developing an
efficient approach is the elimination redundancy in the information. This requires
the effective use of classes. Let's return to the example of the construction
materials. One knows that the stress in the utilized construction material should
not exceed a typical value. This is basic knowledge. It implies that one must
search in the class Constntction Materials for a material for which the value of
stress does not exceed the value of maximum stres.s. Now one operates on the
class Construction Materials instead of on all the individual objects within this
class. Concluding : One piece of information (i.e. search for material) is
translated into one operation. The advantage of using classes is that the
operation on the class is not influenced by the content of the class. It is
indifferent to the system whether there is only one object or whether there are
one hundred objects.
One must gather all the information or as much as possible information
before starting the process of analysis. This gathered information is known as
knowledge. This knowledge is the basis for developing a solution strategy. A poor
knowledge level translates directly into a poor strategy. Knowledge can lcie
gathered from books, journal articles, conference papers etc. and of course the
experienced practitioner. The interpretation of a problem by the expert himself
has a tremendous impact on the overall strategy.
The development of an expert system, the design of a knowledge base,
starts in this study with an initial description of the final result. Starting at this
description one reasons backwards to locate and determine which parameters
influence the design. This process is known as backward chaining (Coyne, 1990).
This approach moves just in the opposite direction of a normal design procedure
and initially might cause some difficulties for an inexperienced designer who tries
to utilize a knowledge base approach. One must define the nature of the final
result prior to the development of a knowledge base.

7
 The final and most important part of the development of a knowledge
base is the system validation. One has to be sure that the knowledge base acts
properly. This is not generally a simple matter because the knowledge base
generally reflects incomplete knowledge about the problem domain. Even with
these limitations one must validate the knowledge base. It is very important to
continuously validate the knowledge base during its development because it
allows the user to make corrections before the knowledge base becomes too
complex. A good method checking the validity of the knowledge base is to
discuss the results with a human expert. An indication of the accuracy of the
knowledge base is the when experts conform the results.

2.2 Expert System Application

Designing a TLP is a very complicated matter, especially because there
are so few 'TLP's in operation around the world. It is still very hard to come up
with some design criteria from operational experience. Field data has been
gathered about the behavior of the existing TLY's though most of it is propriety
to the oil companies. There is a lot of research activity focused upon the design
of TLP's. However, even with API-RP2T (American Petroleum Institute, 1987)
there remains a lot to learn regarding the design of a TLP.
Probably the biggest challenge in design is to find a good starting point
for the design spiral (see American Petroleum Institute, 1987). The better this
point is defined the faster one can obtain a solution and the more accurate
preliminary design will be. Incomplete information about the system being
designed does not facilitate the definition of this point.
Conventional computer programming methods are not really suitable for
this problem and hence the use of an expert system approach was selected. This
approach allows one to develop a preliminary description of the structure. The
expert system approach designs a TLP based only upon a few parameters like the
water depth and the number of columns. In these cases the expert system itself
contains enough knowledge so that it is capable of providing suggestions to the
user. The approach also allows the user to specify the structure more precise, as
additional knowledge is gained, hence overriding the expert systems suggestions.
The expert system will present the most cost effective configuration for the
specified constraints.
It is possible to solve the TLP problem in a spreadsheet format and
engineers have used this approach for other types of structures. Some of the
features of an expert system can be incorporated into a spreadsheet. The
advantage of the expert system is that it is extremely well equipped for guiding
the user towards an answer. The expert system interacts with the user by asking
well formulated questions and depending on the case, gives a list of options to
choose from. The presentation of selected questions and options are extremely
important factors. This control can deceptively lead the user to assume that the
knowledge base is quite simple. The use of an expert system approach allows
unexperienced users to develop a preliminary TLP design hence providing, a
learning tool.
As users get more experienced with the system they may wish to add and
modify the existing system. The text which follows provides the basis fOr
achieving this objective.

23 Knowledge, software and hardware requirements

The use of an expert system approach to problem solving is a relative new
approach to the filed of engineers. The expert system approach has seen limited
use in engineering because it required special work stations, it was too
complicated, too slow and very inefficient. More recently, the expert systems werle
moved to high end PC's but the software was slow developing adequate power.
Only recently with the aid of fast computers, like IBM-PC 486 with a minimum

9
 ?1,

of 6 Mbyte RAM, it is possible to use the expert system approach idea as a

method for problem solving. Surprisingly, some software development packages
are transportable to other computer systems such as a VAX or SUN work
stations. Thus substantial development efforts on small computers does not need
to be lost since it is transportable.
Building a knowledge base can be regarded as writing a computer
program thus, some of the skills learned using higher level languages can be used
to great advantages. The term environment is sometimes used because most
knowledge base software access spreadsheets' or data bases to draw conclusions
or to present results. It is also possible to communicate with other programs,
including finite element packages or independently developed software packages.
Several facts have to be taken into account before one maps the
knowledge into the expert system environment. First of all the knowledge must
be analyzed and well organized. All redundancies must be removed and one must
have a basic idea of bow to implement the knowledge. Furthermore the
-
knowledge must be prepared for a specific knowledge base environment since
they are not all identical. This preparation contributes extra knowledge to the
knowledge base and is environment specific.
The use of an IBM 486 in this study provided the bare minimum in speed,
a slower machine can be used for development but the communication with the
user is very slow. When running the knowledge base on a slower computer it may
seem like there is something wrong but the delay is only caused by the lack of
speed. The speed of the execution of a knowledge base has probably a
psychological effect. One may tend to doubt the reliability or the validity of the
knowledge base when one has to wait for a while. When the input is very
straightforward one expects the analysis to be executed quickly and one tends to
assume that the computer code is probably quite simple. However as we will see
this is not the case.

10
 The expert system environment used for the development of a Tp)
description in this study is called NEXPERT OBJECT commonly called
NEXPERT. It operates under Microsoft Windows environment and was designed
by Neuron Data. NEXPERT allows for both rule based knowledge and object
oriented knowledge, often these types of environments are called hybrid expert
systems. NEXPERT provides extra facilities for a more powerful knowledge base.
One of these facilities is called the META-SLOT. The META-SLOT can be
regarded as a rule or set of rules which will be evaluated only when an object,
related to this META SLOT, is evaluated. The knowledge base in this study v:rlas
developed with using a development version of NEXPERT. Most of the functions
of this version are graphical and provide the user with a very good idea about
what is going on in the complex knowledge _base. This graphical representatiOn
is probably the best way to represent the knowledge base and can be used very
efficiently to fix programming bugs.
The knowledge base developed during this study is transportable to other
versions of NEXPERT and to other machines There is run-time version Of
NEXPERT. This version allows the user only to analyze the knowledge base,
however the knowledge base can not be modified with this version. The run-time
version allows the developer of the knowledge base to distribute his product
without providing the capability of modifying the knowledge base. The knowledge
base can alto be encrypted preventing disclosure of the knowledge structure.

11
 3 TLP DESIGN THEORY
Before one maps the available knowledge into a knowledge base one has
to capture and analyze all of the relevant knowledge. The basis for the TLP
design methodology used in this research study is based the work of Nordgren
(1989). Following his approach it is assumed that the total cost of a TLP design
can be optimized based upon the column configuration and the tether
characteristics. This optimization procedure presently does not optimize for an
efficient deck layout, efficient location of processing equipment or other design
constraints which represent aspects of a refined design. A number of new ideas
were generated to make this methodology more applicable to the design of
TLP's.

3.1 TLP Constrained optimization

A TLP design constrained optimization method was recently developed
by R.P. Nordgren (1989). His goal was to reduce the weight-based cost of a TLP.
The optimization procedure takes into account the vertical equilibrium of the
platform. The equations presented in this research study differ slightly from the
ones presented in his journal article.
Optimization is obtained by imposing three constraints. The first constraint
states that the maximum stress in the tethers is not allowed to exceed a given
value. The second constraint requires that the tension in the tethers be above
specified value in order to eliminate the concern about the bucking of the
tethers. The third constrained requires that the response heave frequency of the
platform to the sea be significantly smaller than the natural heave frequency of
the platform. This to reduce the stretch of the tethers. Reducing the stretch
reduces the stress tethers hence reducing fatigue damage.
Fig. 2 presents a basic TLP configuration. This figure shows a four-legged
platform with flared columns, pontoons and braces. The platform is attached to

12
the subsea base by means of n, tethers and has n, risers. Though the platform
shown in Fig. 2 has four columns which is consistent with popular designs but for
increased versatility of the knowledge base, the platform used in the model for
this study has n, columns Thus it is possible to explore a TLP with three, four,
five, etc. legs. The riser system is are modelled as a single riser with a combined
top tension. In Fig. 2, h represents the incident wave height, 1 represents the
length of the tether, H is the water depth and a is the platform draft. In this
study we assume that the subsea base doesn't have any height therefore the
tether length plus the draft equals the water depth. The on-center spacing
between the columns is represented by the parameter s. The height of the deck
is represented by e while f represents the width of the deck. Further, the height
of the flue-out is represented by b and the column length is represented by c.
The basic column diameter is represented by di, d2 represents the diameter of
the flare-out while di represents the diameter of both the pontoon and the
braces. The fact that the diameter of the braces equals the diameter of tlle
pontoons reduces the number of parameters in the model. In the rest of this
paper the word 'diameter' will also be used for a square column. Actually this
word should be replaced by significant width but diameter is more obvious.
The first step in modelling the platform is to develop an equation for the
vertical equilibrium of the platform. The buoyancy force of the platform must be
equal to the combined riser tension plus the tether tensions plus the gravity
loading of the platform itself and the deck equipment. This can be expressed as

B=g + Fd+ 7; + 77,7; (1)

where B represents the buoyancy force, mp represents the mass of the whole
platform, Fd represents the load of equipment on the deck. The combined riser
top tension is represented by 7., the tension of each separate tether is
represented by Ti. The acceleration due to gravity is represented by g.

13
Fig. 2. Tension Leg Platform Configuration

14
 The buoyancy force of any partially submerged structure is controlled by
its displacement. The columns are either round or square. The displacement of
a column can be calculated by the following formula

V-=k-612a (2)

where V is the displacement, a is the draft of the column and d is either the
diameter of the round column or the beam of the square column. The parameter
k is a shape factor for the column, k = 1 for a square column and k = 14 for
a round column. The cobunns, pontoons and braces, if existing, contribute to the
displacement. The contribution to the displacement of the braces and pontains
is constant while the contribution by the columns depends on the area of the
column at the water line for one type of TLP configuration. Thus the buoyancy
force can be approximated by

B = zo + zi di2

The same idea about the columns, braces and pontoons is used to determine
the mass of the whole platform

mr = n, d j2 (4)

This equation describes the mass of the platform and not the weight. There is
close relatiottship between the factors nj and z1. They describe the contribution
of the column to the buoyancy force and mass of the TLP respectively. Utilizing
Equations (1), (3) and (4), the tether top tension can be expressed as

7; (z0 g no + ?lc (zi g ndc112 - Fd - Tr) / ni (5')

15

Ii
As the platform responds to the heave motion of the sea, this causes a cyclic
tensioning in the tethers. The maximum dynamic tension in the tethers can be
approximated by

D (I, + b1 nc. ) hA (6)

where D is the tension amplitude, bo and bi are constants relating the effect of
the submerged TLP configuration and the wave height to the dynamic tension
amplitude. Both of these constants are determined from hydrodynamic analysis.
The stress in the tethers originates from the axial tension in the tethers
and from bending of the tethers. The bending stress is an important factor but,
for simplicity, it is assumed that the axial stress is domirtant The maximum
allowable stress in the tethers is reduced to compensate for this assumption
(Nordgren, 1989). The maximum stress occurs at the top of the tethers where

max [a I = (T, + D) / A

where am represents the reduced maximum allowable stress, max (a] represents
the actual maximum stress and A represents the cross-sectional area of one
tether.
Combining Equations (5),(6),(7) and (8) one obtains the constraint
equation

(z1 - g n1 b1 h) d2 + n a. A > - g no + b0 h - Fd - T,. (9)

The smallest tension is expected to occur near the subsea structure. The
tethers are not allowed to experience compression since they would buckle or

16
cause damage to the tether mounts. The minimum tension in the tether can be
expressed as

mm. = 7; - g -(1 - p, -1 -A - D (10)

Here min [T] represents the minimum occurring tension, p, represents the mass
density of the tether. The term b, represents the buoyancy factor of the tethers;
it ranges from zero to one and takes into account the attachment of buoyancy
tanks or synthetic foam to the tethers. The tether must be kept under adequate
tension all the time hence

min [T] >7; (11)

where Tb represents the minimum acceptable tension at the bottom of the tether.
Combining Equations (6),(10) and (11) one obtains the second constrained
equation

(z3 -bih)d +n, g -b,) pi 1A < zo -g no-boh -Fd -Tr -n,Tb (12)

The final constraint used in this study is specified in order to minimiie

the fatigue damage of the tethers. An important contribution to fatigue of
tethers is caused by the heave of the platform. The heave of the platform induces
elongation of the tethers which in turn effects the stress levels . In order to
reduce the stress levels in the tethers both the heave of the platform and the
stretch of the tethers must be reduced. This leads to the specification of a
constrained on the heave frequency of the TLP. The longitudinal natural
frequency of the tethers must be far away from the response frequency of the
platform. In mathematical terms

17
 (13)
where
wh = (kh/mh) 1/2 (14)

kh = ni E + ne k, to. 412 (15)

= nip +. Fd/g + ms5 (16)

The parameter kc is the shape factor of the column and wh is the radial frequency
of TLP heave, ws is the dominant wave frequency. The virtual mass of the
heaving platform is represented by mh, m55 is the added mass in the heave mode,
pw represents the mass density of the surrounding fluid. The spring constant kh
includes the contribution of the axial stiffness if the tethers as well as the
influence of the columns Similar to the approximation of the buoyancy of the
TLP, an approximation can also be made about the form of the added mass
expression

mss = so + Sj nc (17)

Combining Equations (13),(14),(15),(16) and (17) one obtains the final

constrained equation

nc - (nj + Si) wh2Jd2 + (n, E / 1) > (no + .so + Fd/g) wh2 (18)

The cost of building a TLP can be regarded as the sum of the costs of the
TLP sub systems i.e. the tethers, deck, columns and subsea base. The cost of the
tethers is assumed to be proportional to its combined mass (mt).

mi=n,P,1A (19)

18
 1

The cost of the subsea base is considered to be proportional to the mass of the
sub sea base. The mass of the subsea base is considered to be proportional to the
maximum effective tension at the bottom of the tether (max 1 TbI)

max Tb ] Tb +D (20)

With these equations the combined cost of a TLP (Cc) may be written as

Cc Cp m + C, m, Cb nr .max 1 Tb] (21)

Combining (4),(19),(20) and (21) one obtains

= Co + CaA + Cd d 2 (22)
with
Co = Cp no + ct, (z0 - g no - + b, h) (273)

Ca n, 1 IC, p, - Cb p, (1-b,)]

1Cp g ni + Cb (z1 bi h)] (25)

The combined cost of the TLP has to be reduced as much as possible. In other
Words

c, <_ co + A + Cd d 2 (26)

Comparing (9) with (12) and (18) and (26) we see that the in this case the
minimum cost can be obtained by IninimiAng to the column diameter and the
tether cross section area within the given constraints. The number of columns has

19
 constant
cost
0 Lines
0 \\
decrOas\i g
feasab e I

0 feasable
cost V
0
r \\ Dr easab e I

0 or 3
= 3 or 1
optimum
conf I gur'at Ion

column area

Fig. 3. Constraints and Optimum

20
to be specified. As more columns are specified the cost will increase (Equation
(25)). The lines representing the constraints are shown in Fig. 3. The three
constraint equations can be represented by three lines. The equations for these
lines are
ci <al d2 + k A (27)
for i 1, 2, 3.
There is a difference in cost caused by different design philosophies for
the North Sea and the Gulf of Mexico (Amold,1989; Roobaert,1989). However,
this cost difference is not taken into account in this study.
Some caution is required with the parameter Fd which represents the load
of the equipment on the deck. It should include the living quarters, process
equipment and process fluids. This parameter does not represent the whole
topside load. The major part of the whole topside load is included in the mass
of the platform. The parameter Fd is a true site specific parameter which
depends on the water depth, field size and petroleum type (i.e oil, gas, amount
of water, amount of sand, chemical contents etc.).
A hydrodynamic analysis was performed on a TLP which had a square
deck and has four circular columns (Nordgren, 1989). The distance between the
columns (column spacing) was 61 m. The columns were flared at 15.2 m from the
bottom, where the flare-out diameter was 19.2 m. The column diameter was 146
m at the water line and the draft of the whole platform was 33.5 m. The total
mass of this platform estimated to be 31.6 106 kg. The platform had braces and
pontoons. Seven tubular members at each column (four in tethers total) had a
combined cross sectional area of 0.143 m2, the maximum allowed stress in the
tethers was specified as 155 Nimm2 Additional parameters which describe this
TLP are shown in Table 1. With these values Nordgren calculated the most wit
efficient l'LP configuration using the three constraint equations.

21
 Table 1. Description of TLP (Noragren, 1989)

=4
lip
=4
= Ir
Ac,k = /4
= 14.6 m
d2 = 19.2 m
d3 = 1.8m
a = 33.5 m
b = 15.2 m
c = Unknown
s = 61 m
pb, = 1025 kg/m3
= 9.81 m/s2

The following values are obtained from the hydrodynamic analysis

zo = 10.14 MN
z1 ?lc = 1430 kN/m2
no = 8.33.106 kg
n1 nc. = 23.8103 kg/m2
bo = 569 kN
b1 nc = 9.33 kNfm 2
so = -5.70.10° kg
si nc = 69.7.103 kg/m2

22
3.2 Expansion of model
The objective is to design a knowledge base which can optimize a variety
of TLP configurations. The methodology presented by Nordgren (1989) can be
expanded to include more types of TLP designs. The original model was not valid
for TLP configurations with three or five columns and did not allow for a TLP
without pontoons or braces. Further the number of tethers was fixed at four and
the column diameter and shape were fixed. These limitations have been removed
in the current study.

3.2.1 Buoyancy
Consider the TLP column as shown in Fig. 4. In this study it is assumed
that the TLP has n, columns evenly distributed along the perimeter. As a direct
result of this assumption it was necessary to reevaluate the parameters zo and.;I.
As presented earlier a new variable in the model is the shape of the columns
(k) The value of lc,. equals 1 if the columns are round, it equals ir/4 if the
columns are round. The parameter kp has the same function for the pontoons.
The parameter k, represents the shape factor of the braces. When kp equals 0, it
indicates that there are no pontoons. It is assumed that the shape and cross
sectional area of both the pontoons and the braces are the same and the,ir
diameters are a function of the column spacing. Now it is possible to calculate
the buoyancy force. The flare diameter is assumed to be a function of the column
diameter

=k

The contribution of one column

B. = di /a + b (k2 - 1)1 pp, (29)

23
V
d2

Fig. 4. TLP Column Configuration

24
The contribution of one pontoon

B2 = 432 (s - di) g (30)

A brace connects the top of the column with the center of a pontoon. The length
of a brace, when present, is 1/2 42 bigger than the length of the pontoOn.
There are tvice as much braces as pontoons. The contribution of one brace is

B3 = kb d32 - 1/2 v2 (s - di) g Py, (31)

There are 5, pontoons and columns The total buoyancy force is calculated
accordingly i.e.

B n B1 + np B2 2 71), B3 (32)
and f011ows that

B= fn k d12 [a + b (k2 1)] Is, d32 (1 + 2)-('s k dill g pp, (33)

Note that the variations of the free surface are neglected in these estimates

3.2.2 Mass
Related to the buoyancy of the MP is the mass of the TLP. The
contribution of the columns, braces and pontoons are included in the following
estimate of the platform mass as

ml Inc k + b (k 2 ' Pc. llp kp d32 (I +1r 2)- (S - d1) Pp I

The contribution of the deck to the total mass is

 = (35)

The terms PC, pp and P, are the average mass density of the columns, pontoons
and deck. These terms are the result of the division of the mass and the volume
of either the column or the pontoon or the deck. This enables the consideration
of columns, braces and pontoons as solid objects.
The total mass of the platform is then obtained as the sum

mp "2/ m2 (36)

3.2.3 Pontoon diameter

The pontoons provide extra buoyancy to the TLP and they also stiffen the
columns. A function for the size of the pontoons is required. This function
describes the diameter of the pontoons when they are round, or the width of the
pontoons when they are square. Assuming that the diameter or width of the
pontoon can be approximated by the following formula

d3 = s+q (37)

Where L,, takes into account the influence of the column spacing and q is the
minimum diameter.
Substituting Equation (37) in Equation (33) one obtains the following
expression for the TLP buoyancy

B = fn, lc di fa + b (k -1)] + /5, + q) (1-1-v2)(s -k dill g p,,, (38)

26
3.2.4 Added Mass
It is rather difficult to specify the theoretical added mass for a TLFis of
variable configuration since the added mass depends on the T'LP's heave
frequency of the TLP. Numerical analyses have been performed on a heaving
structure which crosses the water line (Wybro, 1980). A graphical presentation
of the results of Wybro is shown in Fig. 5. This figure clear shows that the added
mass coefficient depends on the heave frequency. It is possible to approximate
the normalized added mass as a function of the normalized frequency. But 4en
the normalized frequency is bigger than 0.25 one can estimate the normalized
added mass has a value of 1.3. The added mass can now be estimated as 1.3
times the mass of displaced fluid. The mass of the displaced fluid by a column
can be obtained from Equation (29).
The values of bo and b1 in Table 1 were probably obtained through curve
fitting and hence these values presented by Nordgren (1989) are only valid in a
very confined range of column diameter and platform configuration. Therefore
the values of the parameters bo and b., are calculated differently for this study.
A comparison of the present approach with that of Nordgren (1989) is shown in
Fig. 6. One can see that the values of the added mass don't differ significantly
when the column diameter ranges from 18 to 25 m. For diameters smaller than
20 m the present theoretical approximation approach overestimates the added
mass predicted by the Nordgren (1989) model. For bigger diameters the opposite
is true.
The approximation calculations of the added mass is as follows. First, the
contribution of the columns to the added mass is

mssj=1.3.n.k.dj2fa+b(k21)J.p,,, (0)
The contribution due to the pontoons is

27
 2d

2D

Normalized added mass for waterline crossing and heaving

structure

2.5

normalized 1.5
added mass
rn33/pV 1

0.5

0
0 0.5 1 1.5

frequency parameter (w^2*d/g)

Fig. 5. Added Mass of Heaving Column

 mss, np kp d32 (s - ) 1:) (40)

The contribution due to the braces is

mssj = 2- kb - d32. 1/2 ,r2 (s. - (41)

There are n3, pontoons and; columns The total added mass is then calculated
accordingly

mss = mss m2
55, + M55,3 (42)

or substituting the appropriate equations

mss = [ 1.3 n, di2 [a + b (k2 1)] + ni, lcp d32 (1 +,(2)-(s - k )1 A, (43)

Comparing Equations (43) with Equation (17) one obtains the following
approximation for the coefficients in Equation (17)

so = d32 (1 + v2)-(s - k ddi -

= [ 1.3lc [a + (k 2 - 1 )11 13,,

3.3 Assessment of expert .system's knowledge

The next objective is to find values for the different parameters used in
the previous equations. An existing TLP or model is required to lay a basis. The
TLP as described by Nordgren (1989) is the basis to calculate the parameters.
The values of these parameters are used to describe the TLP initially, when the
user doesn't know these values the knowledge base will use these values. Later

29
 different added mass calculations

50 T 10

5
40 -
0
30 - error
added mass -5
[106 kg] 20 -
-10
10 -15

0 -20
14 15.6 17.2 18.8 20.4 22 23.6 252 26.8
diameter [m]

-=. Nordgren.. Expert System error %

Fig. 6. Different Approach to Added Mass Calculation

30
on in the design phase these values can be modified.
It is easy to calculate k, see Equation (28)

k = d2/d1 03)

k = 19.2 / 14.6

k = 1.31
Calculating the parameters and q, see Equation (37)

s = 1.8
Or
61 + q = 1.8

q Represents the initial diameter of the pontoon. A reasonable value for q is 1.2.2
m or 48". More slender pontoons are prone to bend or buckle easily. This results
in a value for is. l equals 9.5.104.

Substituting these values into Equation (38)

B =Pre112[33.5 + 15.2(1.31 2-1)] + irl.8 (1+,12)(61-1.31 d1)J-9.81-1025 (44)

and
B = zo zrn, d12

hence
77, = w [33.5+15.2 (1.31 2-1)1 9.81 1025

Zo = IT -1.8 -(1 + 12) (s - 1.31 - 9.81 1025

31
Notice that there is a term in zo which depends on the column diameter. This
term is left in zo, its influence is rather small
Calculating the value of zo and z1

zo 10.31 MN
n, = 1402 IcN/m2

These values are within 2 percent of the ones provided by Nordgren

(1989). The fact that these parameters describe Nordgren's 'TLP model well
allows for continuation with his model.
The next step is to calculate the mass parameters. The displacement of the
platform is 14263 m3, the mass of the TLP of equals 32.2-106 kg. The length of
the column (c) is not yet defined. Assuming that the column length equals the
column spacing, then

C = 61 m

The contribution of the deck mass to the mass of the platform is a

function of the deck height (e) and the deck width (f). The deck height is
assumed to be a linear function of the deck width. The deck width is also
assumed to be a linear function of the column spacing. The deck mass can be
expressed as

m2= (45)

where pd is the mass density of the deck.

Combining Equations (34),(36) and (45) one obtains the following
expression for the platform mass

32
 =Ir (112161 +15.2(1.31 2-1)13c+g 1.82 (1 +v2)(61-1.31 di)Pp+61 Pd (46)

compare Equation (46) with form of Equation (4) which is

Mp = no + ncn, (4)

one can deduce the coefficients no and nj to be as follows

no = IT 1.82(1 + v2)(61 - 1.31 aid Pp + 6i3 Pd

n n=ir 161 + 15.2 (1.31 2-1). pc

The next step is to find some acceptable values for the densities. Based
upon the values presented in Table 1 one obtains the following values for no and
nc, for d1 = 14.7 m

no 8.33106 kg
n n=
c 23.8-103 kg/m 2

These values imply a value of the column mass density of

pc = 105 keth3

This value is smaller than 1025 kg/m 3, the mass density of the
surrounding fluid and therefore the structure will float. This value is about 10%
of the mass density of the surrounding fluid and implies that the columns provide
a vast amount of buoyancy.

33
 Assuming that the mass density of the pontoons equals the mass density
of the colnmns, i e.

= Pc (47)

This assumption is allowed because the pontoons and braces don't contribute a
huge enough factor to the overall design of the TLP. This allows us now to
calculate the deck mass density.

Pd = 36 kg/m3

This value is smaller than k and this can be explained on the basis that
there more equipment is located in the cohimns than in the deck structure.
Another reason is that the prime function of the deck is to provide a work floor
and not to house equipment.
At this stage most of the basic parameters in the TLP model are defined
and one can proceed with the application of the model in the expert system
environment.

34
 4 IMPLEMENTATION FOR EXPERT SYSTEM
The objective of the knowledge base is to allow the design of a TLP from
a minimum of available information. A theory has been developed which
optimizes a TLP design. The optimization procedure has been implemented
utilizing a spreadsheet accessed from the expert system environment. The first
step in the implementation process requires the available information to be
organized and analyzed by the knowledge base developer. This is probably the
most important part of the process because from this process one develops the
knowledge base. In the case of TLF' design the knowledge analysis, i.e. the
optimized description, has already been performed. This knowledge is now
mapped onto a knowledge base. Knowledge preparation is the next step. This
primarily implies the development of the object based knowledge base. Objects
and properties are defined. Several classes are defined which enable easy
manipulation of the objects.
The final product prompts the user for different values of water depth aid
column spacing, this is the Minimum required amount of information. Other
aspects can be user specified. From this point on the knowledge base will
calculate an optimized TLP configuration. The knowledge base keeps track of it's
own suggestions and provides a list of all parameters and suggestions to the user.
A graphical representation is available to provide a different view point to the
TLP configuration. It is possible to modify parameters. The knowledge base will
check for a logical combination of numbers of tethers and the amount of
columns It can also check whether the waves will hit the bottom of the deck.
The questions the knowledge base asks the user must be formulated well.
A nicely formulated question enhances the validity and accuracy of the final
design.

35
4.1 Knowledge inventory
Before describing the implementation of the available knowledge it is
worthwhile to list the present types of knowledge. There are five basic types of
knowledge which contribute to the TLP design. The core knowledge is the
information about the three equations which describe the effects of platform
configurations and constraints on the column diameter and the tether area. This
type of knowledge is covered by Nordgren (1989). A second type of knowledge
can be described as the hidden erpert's knowledge. This knowledge contains the
suggestions the system makes when the user's own knowledge is insufficient. A
third type of knowledge represents the compatibility of the model. This type of
knowledge is inherent to all systems. It imposes basic conditions to the design.
In the case of a TLP design it checks whether the waves hit the bottom of the
deck or whether there is a logical relation between the number of columns and
number of tethers. The fourth and fifth type of knowledge can be called the
knowledge smoother and the implementation knowledge. The last two types of
knowledge will be explained in more detail.

4.1.1 Knowledge smoother

One of the basic tasks of an expert design system is guiding the user to
a possible design configuration. This implies that when the user may not be
familiar with a problem or doesn't want to define a parameter, the expert system
itself will provide a value for the parameter or the system will reprompt the user
for a value when the knowledge base cannot provide this value. This happens
quite often in the knowledge base. The knowledge smoother invokes other types
Of knowledge, like the hidden expert's knowledge or the compatibility knowledge.
The knowledge base is capable of providing values for several parameters. These
parameters are for instance the number of columns, the number of tethers, the

36
shape of the deck and columns. For further information about all suggestions see
appendix C.
Providing a good user interface is an important part of the expert system
approach. One of the ways this is achieved is to keep track of all the parameters
which describe the TLP. This is done in a spreadsheet which runs concurrent
with the knowledge base. This spreadsheet can be made visible all the time.
is also possible to make a print out of the spreadsheet. The use of a spreadsheet
has no influence on the knowledge base itself though the idea of using a
spreadsheet is a part of the knowledge smoother.
The user of the knowledge base does not notice on a direct basis the
effect of the system's advice. Programming this is rather complicated though.
The system asks the user to make a decision or to provide a value. One of the
option on the selection menu is the NOTKVOWN function, see Fig. 7. The value
Of the current parameter is now NOTICNOWN. This value must be transformed
into a system's suggestion. At this point the META SLOT is the right tool to
tackle this problem. The META SLOT is specific to NEXPERT. It enables the
user to perform specific tasks when a parameter is evaluated. Examine the case
when the value of a parameter is NOTKNOWN. The META SLOT contains an
instruction which invokes the analysis of a rule. This rule contains the specific
information about the current parameter. When the value of the parameter is
NOTICNOWN, the rule will asses a value (expert system's advice) to the
parameter. The rule also writes to a spreadsheet the fact that this value of the
parameter was specified by the expert system. (The instructions in the META
SLOT are invoked again but the parameter has a value unequal to NOTKNOWN
hence the rule will not be executed).
Another location where the META SLOT is used is to define the tether
length. The expert system is designed to prompt the user early on for the water
depth and the platform draft. The META SLOT of both parameters refer to one
rule where the tether length is calculated. The tether length is calculated

37
 --si.ON1.F10.1::.

Fig. 7. Nexpert Parameter Selection

38
immediately when both parameters are known. This also allows the user ' to
modify the water depth; the expert system immediately modifies the tether length
accordingly
The META SLOT also allows the developer of the knowledge base to
define the questions which will be asked. For instance, when the expert system
prompts the user for the value of the water depth it asks : "WHAT IS THE DEPTH OF
WATER" because the water depth is represented by the object water with the
property depth. An experienced user knows that the system likes to obtain,
value for the water depth. The META SLOT for the parameter water.depth
contains an option for the prompt line. The prompt line contains "'WHAT is TIM
WATER DEPTH" which is quite straightforward.

4.1.2 Implementation knowledge

The theoretical analysis which has been presented describes an
optimization for a TLP design. However, it doesn't describe how this
optimization process should proceed in an expert system environment. The
optimization process is typical engineering knowledge, it can be described as a
tool which allows the designer to perform his task. The optimization is one part
Of the knowledge base, the knowledge base is also capable of modifying values
of specific parameters. How, when and where the modifications occur depends
on the designer, all these aspects are not included in the theoretical analysis.
They are the engine of the knowledge base.
One of the aspects of a rule based knowledge base system is that thie
system analyzes the rules. Initially there is no order in the rules, all are of equal
importance. The expert system environment links the rules together depending
on their hypotheses. The analysis of the rules is achieved by both forward and
backward chaining through these hypothesis. It is the task of the knowledge base

39
 '

User. Expert SyStem

specified specified
va ues values

Optimization

Spreadsheet MOdificatiOn
TLP Of values &
description descriptlon I/O actions

Desired
resuit

Fig. 8. Overall Flow Chart of Knowledge Base

40
developer to prepare these rules for the linking and to have an initial idea of
how the overall flow, see Fig. 8. The basic idea is to obtain all the parameters
which describe a TLP. Then the optimum of column diameter and tether area
is calculated. From this point on the knowledge base allows for alterations Of
values of specific parameters or provides the user with a description of the T'Lf,
either in the form of a table or schematic diagram.

4.2 Knowledge Base Description

The one of the main hypothesis in the knowledge base is platform defined.
When this hypothesis is true, the optimization process can start and later on
values can be modified. The hypothesis platform defined is related to a rule with
the following conditions: Mass jactors, Buoyancy _factors and Added jactors, see
Fig. 9. These conditions are hypotheses themselves. All these hypotheses have io
be true in order to define the TLP. These hypotheses require the user to describe
the TLP so that respectively the mass, the buoyancy and added mass 'TLP can be
calculated. The _factors hypotheses will force the user to define a TLp
configuration. When the hypothesis platform _defined is true several actions follow.
First of all the three constrained equations (Equation (27)) are defined by means
of their coefficients, i.e. al, a2, a3, b1, b2, b3, c1, c2 and c3. The next action is to
find the optimum, followed by writing the TLP description to a file which allows
the TLP to be plotted.
The optimization procedure can be done very efficiently in NEXPER'T.
It prevents using too many rules and is easily expandable. Let's analyze the
optimization procedure. There are three equations which have to be satisfied. As
can be seen from Fig. 3 (page 20) there are three points where lines intersect.
Only at one of these three points the three equations are valid. Therefore the
objective is to calculate these three points and check whether it complies with the
equations. This is achieved in the following manner. First the intersection of line

41
 ... 6.1

Ir`u

Yes cokrntshaps? \ II!'

.)Do pontoon_present pont ?.. .,.
.Do flare_exists flare exisb? ....,::,...,
ii
.>Do Won shape braces.? 1.. r.:?
=>Do shape cokortslcolur ? - ';,".;1 \
.>Do (IcsPlYwicolurnn.spac? .1,.: ' 4!!.
9?
e .>Do belcalunn.re/4 bel

Yes colurnn_shape?.,
,

\ ..

steel?

.Ii. .
s>Do pontecr_presert pail ?
Yes rears factors? ........ I"'
s>Do braces_shape braces.? ' r) ..."? II -platforrs_defined ? ...-..
. ..... Yes bussercy factors ?--.=4 --
.>Do Rare_esists flare watt?
/Y02 addeci_factors? "
1......
,.. Do shace_coeurnnslcolur ? 1 `,./
soDo POWUrcolumn.spaci? / ..'
...

,,i.

. Ir
Yes csissetshape? .,. r ,

-,Do portocri_shape panto ?, - ,\ :, ': 1.

.,

-)Do brazes share braces.? ..-i!tt.' ?i

ii
...)Do 1.7shape column:1c? -:;/' ;

i...,' =>Do POWarcolurnrimsaer

si
....:!..:.,....,,.:!..::::...::::.:.]:::::::..::::.:T.i.:jiLj....;;::.,;::.::::;::i.E..::::;::1..ti:::fr:.:2J'AIL ' . ..;fv:III,,....-E.i:12.:!:::::!::..,:

Fig. 9. Rule Description

42
1 and 2 is calculated, yielding a value for the column diameter (resultdiam) and
a value for the tether area (result area). There is a rule which checks the values
of resultarea and result.diam. When the values are correct, i.e. the values of
resultdiam and resultarea comply with the three constrained equations, the action
of the rule implies that the value of result area is copied onto tetherarea and the
value of resultdiam is copied onto column.diam. This procedure is repeated for
the intersections of lines 2 and 3 and lines 3 and 1.
A good knowledge base contains no redundancy. This optimization
procedure provides a good example of how to avoid redundancy. As stated before
one must check the results of the different lines. The most efficient method (LI!.
no redundancy) is the following. The initial value of i is 1 and the initial value
off is 2.
CONDITIONS=
<=

HYPOTHESIS= eeleet_optinniin

ACTIONS=
Reset (define __j)
Do (define jidefine _j)
Do (('sAj\m"cAi\m-'a'\i\.1,0"e\j\.*)/('sAj\.vsib'\i\.v-'sAiVir""bAj\.v))
.(reiult.area)
Do (SQR.T(CbAj \ m'c'\)\.v)/Ca'\i\ .sonbAj \ .v-'sAj\.ronbAi\.v) \ )
(result.diam)
Reset (vsilid_tlp_selection)).
Do (valid_tlp_selectio(ajalid_tlp__seleetion)
Do (i+I) (i)
Do (j+1) (j)
Reset (select optimum)

CONDITIONS=
3

HYPOTHESIS= define...)
ACTIONS=
Do 1.

43
The rule define j controls the value of j. It allows the value of j to cycle in the
following order : 2,3,1. The string 'alji.v defines the parameters al, a2 and a3.
There are some programming requirements like the single quotes and the
addition of a property but this does not influence the performance of this rule.
The hypothesis valid tlp selection checks whether the current values of diameter
and area are the optimum ones. The Reset function enables the user to re-
evaluate that hypothesis. The Do (hypothesis narne) (hypothesis name) function
allows the user to evaluate the hypothesis immediately. The Do (j+1) (j)
function assess the value of j+/ to value off. The function Reset select_optimum
enables a loop until the value of i is greater than three.
The second major loop in the knowledge base allows for modifications
of values. The user can choose for a parameter to change a value when the TIP
is defined and the optimization procedure is completed. The whole new
optimization process will be executed. The idea to make this work is as follows.
The user selects an item to be modified, e.g. the columns Then the knowledge
base reasons backwards and prompts the user to select which part (of the
column) will be modified. This can be the amount of columns or the shape of the
columns, etc. When a new value is submitted to the system, all the hypotheses
which govern the calculations are reset. The knowledge base gets the command
to re-evaluate the hypothesis platform defined. A good visual feed back enhances
the effectiveness of the knowledge base. Therefore, the old value of the
parameter to be modified is presented to the user before this value is modified.
This can be achieved by using the META SLOT. The META SLOT allows for
variables to be displayed on the screen. Fig. 10 shows how the modifications can
be made and where they can be found.
The knowledge base also allows for more information about TLP design.
A graphical representation of a TLP can be brought directly to the screen
showing the TLP definitions. There is the capability to plot the designed TIP on
paper. This is a scaled representation of the TLP with dimensions of column

44
 amoUnt
eilaPe
presence coiumns
flare exist
flare height
dfctffiete
flare
fixed ,..
iwarlabieLl.h.flare diameter
presence
/diameter> pontoons
emOvnt modification
shape
presence braces
amount
max stress
young tethers
PuoYancY
density
wateroepth
wave height ,\>eiTte parameters
Wave period

Fig. 10. Search Path for Modification

45
spacing and wave height. The knowledge base can execute an external program.
This program reads the significant information about the TLP form a file and
plots a TLP configuration. The plot program had to be developed, there was no
program available which could perform this task. This program is written in the
'C' language and utilizes a data file generated by NEXPERT. The plot program
is executed from within the NEXPERT environment.
One of the questions the knowledge base will ask the user is to make a
choice of tether material. This feature is a good example of how easy it the use
of an expert system is. This question is prompted through the META SLOT of
either the Young's modulus or maximum allowable stress of a tether. When one
of these two parameters become current the previous question will be shown.
When a choice is made about the material the knowledge base will immediately
assigns values to both the Young's modulus and maximum allowable stress. The
different values of Young's modulus and the maximum allowable stress for
different materials are captured in the knowledge base. When the answer to the
material question is NOTKVOWN, the knowledge base will prompt the user for
both values. Another feature from this example is the fact that it is rather easy
to add a new material type to the knowledge base. One rule is enough to add
one material type and one doesn't have to take into account where this rule will
be located, as long as the rule has the same hypothesis as the other two.
There is nothing really fancy about the development of a knowledge base.
The basics are very simple though the implementation is a very complicated
because of the backward reasoning which occurs in the expert system
environment. Most of the work was done to make the knowledge base more user
friendly and to provide good communication with the user. Debugging the
knowledge base is very time consuming because one is not really aware of what
is going on inside.

46
 5 ILLUSTRATIVE EXAMPLES
The knowledge base was used to develop several TLP configurations. It
should be noted that it is possible to adjust values of several parameters in order
to obtain the desired result, and that certain values were assumed because tlie
actual data was not available.

5.1 TLP for 150 m water depth

Table 2 shows an overview of a TLP in a water depth 150 m. The
description in the right column is obtained by using most of the available
knowledge form the knowledge base. The Hutton TLP in the British sector of thl e
North Sea is installed in the same water depth (Niedzwecki, 1989). A big
difference in configuration is the number of columns The Hutton TLP has six
while the suggested configuration has only four. The suggested TLP has flared
columns, braces and pontoons. The Hutton TLP has just pontoons. There is a big
difference in column diameter. One of the most significant reasons for this is the
difference in the number of columns Note the slender pontoon of the suggested
TLP.
Another 'TLP configuration was created. This configuration is an approach
to the Hutton configuration and can be used to check the model used fnr
calculations. The middle column of Table 2 lists the parameters of this
configuration. This configuration has thick pontoons, no braces and straight
columns. The calculated column diameter is 16.64 m. This value is bigger than
the smallest diameter (inner columns) but is smaller than the diameter of the
outer columns

5.2 TLP for 536 m water depth

Table 3 shows an overview of a 'TLP in a water depth 536 m. The right
column shows a TLP configuration created by using as much as possible the
build-in knowledge. The middle column represents a TIP configuration which

47
resembles the Jolliet TLP. The Jolliet TLP is located in the Gulf of Mexico in
a water depth of 536 m (Offshore, 1990). It is used as a well head platform,
therefore the deck load is rather small compared to other TLP configurations.
A lot of relevant data was not available about the Jolliet TIP. This example is
used to develop some numbers about a TlY in this water depth.

5.3 TLP for 870 m water depth

The third water depth for an example configuration was chosen to be 870
m. The knowledge base generated description of this configuration is shown in
Table 4. The TLP has four round and flared collimns It also has slender
pontoons and braces. The calculated column diameter is 1834 m. This
configuration resembles Shell's Auger TLP which starts producing in the Gulf of
Mexico in 1993 (Offshore, 1990). Like the Jolliet TLP there was no data readily
available about the column diameter though notice that the knowledge base
suggests a diameter of 17.64 m for a water depth of 536 m and a diameter of
18.54 m for a water depth of 870 m.

5.4 TLP for 1500 m water depth

The last example describes a TLP configuration in a water depth of
1.524 m. The description of such a TLP can be found in Table 5. The values for
the tether area and the column diameter seem to be too large. This optimum
is controlled by the minimum bottom tension and the maximum top stress in the
tether.

48
 Table 2. 112 IN 150 m

Item calculated suggested Unit

Hutton TLP in 150m TLP in 150m

COLUMNS
shape ? square round
amount 6 § 4
distance ? 65 65
draft 33.2 33.2 32.5
diameter 17.7& 14.5 16.64 19.27
length 69 69 65
flarediameter n/a 16.64 25.63
flareheight n/a 0 16.25

BRACES
amount n/a n/a 8
shape n/a n/a round
PONTOONS
amount 4 4 4
shape square square round
diameter ? 8 1.33

TETHERS
amount 16 16 4
area ? 0.09 0.26 m"2
length ? 116.8 117.5 m
material Steel pipe Solid Steel Solid Steel -
buoyancfactor ? 0 0
density ? 7900 7900 kg/m 3
max.stress ? 155.106 155.106 Pa
min.tension ? 9.2 -106 9.2 .106 N
young'smod. ? 2.1E+11 2.1E+11 Pa
MISCELLANEOUS
waterdepth 150 150 150 m
min waveperiod 13.9 13.9 13.9 S
max waveheight 16.6 16.6 16.6 m
topsideload ? 2.20E+08 2.20E+08 N
riser top tens. ? 1.43E+07 1.43E+07 N

FONTS bold calculatedby knowledgebase

normal suggestedby user
italic suggestedby knowledgebase

49
 Table 3. TLP IN 536 m

Item calculated suggested Unit

Jolliet TLPiñ536m TLP in 536 m

COLUMNS
shape square square round
amount 4 5 4
distance 42.7&54.8 48.8 42 m
draft 24.1 24.1 21 m
diameter ? 1723 13.1 m
length ? 42.1 44.4 m
flarediameter n/a 17.23 23.46 m
flare height n/a 0 10.52 m

BRACES
amount 8
shape round round

PONTOONS
amount 4 4 4
shape square square round
diameter ? 1.32 1.52

TETHERS
amount 4 4 4.
area 0.23 0:23 mA2
length ? 511.9 514.95 m
material Solid Steel Solid Steel Solid Steel -
buoyancfactor ? 0 0
density ? 7900 7900 kg/niA3
maximunstress ? 155.106 155i0 Pa
minim umte nsio n ? 9.97 405 9.97 405 N
young'smodulus ? 2.1 .1011 2.1 i0' Pa

MISCELLANEOUS
waterdepth 536 536 536 m
min waveperiod ? 13.9 13.9 S
max Waveheight ? 12 12 m
topsicleload 5.28 .10 5.28-106 5.28 .10 N
riser top tension ? 6.44407 6.44.107 N

FONTS: bold calculatecby knowledgebase

normal suggestedby user
italic suggestedby knowledgebase

50
 Table 4. TLP IN 870 m

Item calculated suggested Unit

Auger TLP in 871m TLP in 871 m

COLUMNS
shape round round round
amount 4 4 4
distance 65 65 65 m
draft 27 27 32.5 m
diameter ? 19.76 18.54 m
length ? 50 65 m
flarediameter n/a 19.76 24.66 m
flareheight n/a 0 16.25 m

BRACES
amount 8
shape round
PONTOONS
amount 4 4 4
shape square square round
diameter ? 10.4 1.62

TETHERS
amount 4 4
area 0.36 0.3 m^2
length 843 837.5 m
material Solid Steel Solid Steel -
buoyancfactor 0 0 -
density 7900 7900 kg/m^3
maximunstress 155 .106 155.106 Pa
minimuntension 1O.641 1O.641 N
young'smodulus 2.1E+11 2.1E+11 Pa
MISCELLANEOUS
water depth 871 870 870 m
min waveperiod ? 12.9 12.9 S
max waveheight ? 33.5 33.5 m
topsideload 1.00E+08 1.00E+08 N
riser top tension 4.06 407 4.06 407 N

FONTS: bold calculatedby knowledgebase

normal suggestedby user
italic suggestedby knowledgebase

51

li I
 Table 5. TLP IN 1500 in

Item Unit

COLUMNS
shape round
amount 4 -
distance 65 m
draft 32.5 m
diameter 44.26 m
length 65 m
flare diameter 58.86 m
flareheight 16.25 m

BRACES
amount 8
shape round

PONTOONS
amount 4
shape round
diameter 1.62

TETHERS
amount 4
area 3.74 mA2
length 1491.5 m
material Solid Steel
buoyancfactor 0.9 -
density 7900 kg/m"3
maximunstress 155.106 Pa
minimumtension 11.948.1d N
young'snodulus 2.1 -1011 Pa

MISCELLANEOUS
waterdepth 1524 m
min waveperiod 13.9 S
max waveheight 15 m
topsicleload 1.004 O. N
riser top tension 6.44 -107 N

FONTS: bold calculatecby knowledgebase

normal suggestedby user
italic suggestedby knowledgdmse

52
 6 CONCLUSIONS
Many engineering design problems can be solved by using spreadsheets
(Grecco, 1987). The user of the spreadsheet must have additional knowledge
about the spreadsheet and the specific problem. Unfortunately, additional
knowledge is not always available. The expert system approach allows the user
to be somewhat informed about all the details of the problem and still be
capable of developing a solution. Extra knowledge required such as that for
using a spreadsheet approach effectively is also captured. The available
information is transformed into a general category referred to as knowledge. Th
expert system approach consists of two parts. The first part involves the
identification of rules (rule base part) and the second part captures the relations
of facts and items (object oriented part). Combining these provides a powerf41
and efficient method to solve engineering problems. The objective of the expert
system approach is to integrate different types of symbolic and numeric
knowledge with that of engineering experts. Interestingly, it allows a variety of
people to get an idea of the impact of several parameters on the overall problem.
One must realize that this tool only provides a preliminary description and has
to be analyzed further prior to engineering development_ Although this study
focused upon preliminary design it is possible to expand the knowledge base to
control the entire design process.
Applying the expert system approach to problems calls for several special
actions. First of all, one must be sure that the approach is the best method for
the problem. Not all problems are suited for this approach. When one establishes
this fact the next step is to gather all related information. This information is
then analyzed and organized in accordance to expert system approach. To some
extend the software package being used can influence this process. This
followed by an idea of the developer about the performance of the final product.
There must be a plan regarding of the features and capabilities of the knowledge
base system. This type of knowledge allows the knowledge base to operate

53
smoothly. It also contains specific knowledge about typical features like
input/output, the implementation of mathematical formulae and how to present
the right questions and provide explications to the user during normal operation
of the software.
Developing a description for the preliminary design of a TIP can be done
by using the expert system approach. Information about the design and the cost
of design has been gathered and analyzed. The model developed during this
research study is based upon a constrained TLP optimization approach developed
by Nordgren (1989). His model for a square four-legged TLP was expanded to
allow for a variety of TIP configurations.
Mapping the available knowledge onto the expert system environment is
not very complicated though it requires some skill to create a presentable final
product. The smoothing of the knowledge base is a very complicated and time
consuming matter. The final product, i.e. knowledge base and utility programs
developed, are capable of minimizing the cost of building a TLP. For several
parameters the knowledge base provides values which are a guide to the user in
case of a lack of information. The knowledge base also provides the user with a
graphical representation of the MP definitions and it is also capable of plotting
a scaled drawing of the MP. A spreadsheet runs concurrent with the knowledge
base. This spreadsheet keeps track of the suggestions the knowledge base
provided. It also lists the parameters which describe the TLP. The user can make
a hard copy of this spreadsheet during a session. Furthermore, the knowledge
base allows for all parameters to be modified dining execution allowing the
engineer the opportunity to study the influence of various parameters. All of
these aspects contribute to a very dynamic and user friendly environment for the
preliminary design of development TLP.
Knowledge base validation, i.e. the comparison with existing designs,
provides some helpful suggestions. It shows that the current knowledge base does
correctly configure and describe the size of a 'TLP when parameters for existing

54
TLP's are specified. When a lot of the parameters are unknown the description
will change significantly. The suggestions the software system currently supplies
are not as accurate as they could be, especially in greater water depths (i.e. >
1000 m). This means that the user still must interpret the results the knowledge
base provides. Values of parameters must be modified in order to obtain a better
optimum.
The illustrative examples showed that the knowledge base is a reasonable
representation of current TLP design procedures discussed in the open literature.
Further the examples showed that with several iteration steps the user is capable
of converging on the design of an existing TLP. The suggestion on the pontoon
diameter does not work for greater water depths. The pontoon diameter is too
small, the pontoon must provide more buoyancy. Therefore the suggestion of the
pontoon diameter probably should be a function of the water depth and not of
the column spacing. The current knowledge base does not prevent the pontoons
from having too much buoyancy. Exercising an extreme case one can have, a
solution converging to a point where there is no need for vertical colmnns which
does not make sense. Thus the knowledge base should contain a rule which
checks the column diameter and modifies the pontoon diameter as necessary.
The current lcnowledge base allows the development of a preliminary TLP
design, but there are still several points which could be improved. First of all the
knowledge base does not calculate the influence of the configuration on the
dynamic tether tension (parameters 60 and 61). Furthermore the optimization
only optimizes for column diameter and tether area. The number of columns is
not explicitly included. An estimate of the cost can be obtained when the cost
parameters (Cb, c, and c, see Equation (21)) are known. The current knowledge
base only allows for the development of steel TLP's. Concrete 'TLP's have not
been included in this version. However one can expect that the major influence
the use of concrete will have will be on the mass and the displacement of the
TLP design (de Oliveira, 1988; Offshore, June 1990).

55
 -I.

The research study on this problem is continuing. First of all the

knowledge base will check the vertical equilibrium of the TLP, and modify the
TLP if necessary. The pontoon diameter will be made a function of the column
diameter to provide the preliminary TLP design with bigger pontoons.
Furthermore it is the intention to provide more information about the TLP such
as the natural heave frequency, top tension in the tethers and the added mass.
The study will also focus on how to further optimize the preliminary design
description.

56
 REFERENCES
AMERICAN PETROLEUM INSTITUTE,./?eCOMMEPIded Practice
for Planning Design
and Construction Tension Leg Platforms. API Recommended Practice 2T
(RP2T). First edition, April 1987, Washington D.C.

ARNOLD, K.E. and ROOBAERT, N.C. 1989, Comparison of North Sea and Gulf
of Mexico Design Philosophies. Proceedings of the 21st Annual Offshore
Technology Conference (no 6117), Houston, Tx.

Coy, R.D. [et al], 1990. Knowledge Based Design Systems. Addison-Wesley
Publishing Company, Reading Ma.

Grecco, M. G. 1987, Deep Water Development Economics. Proceedings of

the 19th Annual Offshore Technology Conference (no 5545), Houston, Tx.

MEDZWECKI, J.M. and FRENTZ, G.T. 1989. An Overview of Tension Leg

Platforms. Tension Leg Platform: A state of the art review. A.S.C.E,
New York, NY, pp 297-309

NORDGREN, R.P. 1989. The Design of Tension Leg Platforms by a Constrained

Optimization Method. Journal of Offshore Mechanics and Arctic Engineering,
Vol 111, pp 194-202

OLIVIEIRA, J. DE and FJELD, S. 1988. Concrete Hulls for Tension Leg Platforms.
Proceedings of the 20th Annual Offshore Technology Conference (no 5636),
Houston, Tx.

57
 WYBRO , P.G. 1980. On the Dynamics of Column-Stabilized Platforms
Including Three-Dimensional Interaction Effects. Marine Technology, Vol 17,
no 2, pp 174-198.

OFFsH0RE,1990. Shell's Auger TLP demonstrating cost insensitivity as depths

increase. vol 50, no 1, pp. 28-55

10. OFFsHoRE,1990. Technical Innovations highlight Haltenbanken Heidrun TLP.

vol 50, no 6, pp. 41-47

11. OFFsHoitz,1985. TLP Theory Fulfilling Expectations. vol 46, no 1, pp. 37-39

58
 APPENDIX A
NOMENCLATURE
a draft of platform
A = tether area
height of flare-out
bo heave parameter
bl heave parameter
b, buoyancy factor of tether
buoyancy force
buoyancy force contribution from column
B2 buOyancy force contribution from pontoon
B3 buoyancy force contribution from brace
column length
parameter relating column diameter to mass
cost factor, tether area related
cost factor, tether tension related
combined cost
cost factor, column diameter related
cost factor, mass platform related
cost factor, mass tethers related
cost parameter
column diameter
column diameter
flare-out diameter
braces and pontoon diameter
nominal tether tension
deck height
Young's modulus
deck width

59
Fd deck load
acceleration due to gravity
wave height
water depth
ratio flare-out diameter - column diameter
shape function for columns
spring constant for heave
shape function for pontoons and braces
tether length
constant for pontoon description
maximum tension the tethers it sub sea base structure
maximum stress in tethers
mass of column
virtual mass of heave
min [T] minimum tension in tethers
mp mass of platform
= combined mass of tethers
mass of columns, pontoons and braces
M2 mass of deck
MSS added mass of platform
M15,1 added mass of columns
M55,2 added mass of pontoons
"155,3 added mass of braces
n, number of columns
np number of pontoons
nr number of risers
number of tethers
no mass parameter
n, mass parameter

60
q constant for pontoon description
minimum allowable tethet bottom tension
riser top tension
tether top tension
V displacement of colunm
minimum acceptable response frequency of platform
Wh natural heave frequency of platform
zo buoyancy parameter
zi buoyancy parameter
maximum allowable stress in tethers
Pg mass density of tether material
Pw mass density of surrounding fluid

61
 APPENDIX B

LIST OF OBJECTS
Class Object Property
exist= false
mod= Unknown
braces no= 4
round= False
vorm= square
diam= 32.87
draft= 32.5
flare= true
flare_height46.25
length= 65
column mod= Unknown
no= 4
round= False
vorm= square
width flare43.72
-
diam= 1.42
exist= True
mod= Unknown
pontoon modl= Unknown
no= 4
round= False
vorm= square
output data
area= 0.72
buoy= 0.0
density= 7900.0
length= 621.5
material= solid steel
mod= Unknown
tether no= 4
spec= 7900
stre.ss= 155000000.0
tno= Unknown
young= 21000000000.0
depth= 654.0
water mod= Unknown
spec= 1025.0
freq= 0.41
wave height= 13.2
mod= Unknown
per= 15.0

62
 APPENDIX C
EXPERT SYSTEM SUGGESTIONS

item suggestedvalue why

Columns
amount 4
shape round
draft (columnspacing)/ 2 columnspacingis basic value
diameter (columnspacing)/ 4 columnspacingis basic value
length columnspacing columnspacingis basicvalue
flareout yes
flarediameter * columndiameter
flareheight (columnspacing)/ 4 columnspacingis basic value
1.31 fromTLP analysis

Braces
existence true
shape pontoonshape

Pontoons
amount columnamount fitsin betweenall columns
shape columnshape
diameter 1* (columnspacing)+ q
1 0.0065 fromTLP analysis
1.22m (48") fromTLP analysis

Tethers
amount columnamount one tether per column
length (water depth)- draft boundarwondition

Miscellaneous
riser top tension 36500*watedepth+ 818E6 fromNordgren'examPle
tetherbot. tension 2000*watedepth+ 8.9E6 fromNordgrenand Hutton

63
 APPENDIX D
TLP DESCRIPTION SHEET

Item Value Unit ExpertSystem'sadvice

COLUMNS
columnshape round yes
amountof columns 4 yes
distancebetweencolumns 68 m
draft 34 m yes
columndiameter 36.88 m yes
columnlength 68 m yes
flare diameter 49.05 m
flare height 17 m yes

BRACES
amountof braces 8 yes
shape of braces round yes

PONTOONS
amountof pontoons 4 yes
shape of pontoons round yes
diameterof pontoons 1.44 calculatedvalue

TETHERS
amount 4 yes
area 0.82 mA2
length 620 m calculatedvalue
material Solid Steel -
buoyancfactor 0
density 7900 kg/m^3
maximunstress 155000000 Pa
minimurntension 12000000 N
young'snodulus 2.1E+11 Pa

MISCELLANEOUS
waterdepth 654 m
max allowablewaveperiod 15 S
max expectedwaveheight 12 m
topsideload 1.16E+09 N "Guess,not vetyaccurate"
riser top tension 3.27E+07 N

64
 APPENDIX E
USING THE KNOWLEDGE BASE

The knowledge base is developed with NEXPERT on an IBM-PC

computer. NEXPERT requires a hardware key. The available knowledge base
runs in the NEXPERT & MS-WINDOWS environment. Before running a session
make sure that the following files are all in the same directory: TLP.TICB
(knowledge base), TLP.PIF (MS-WINDOWS plotter control), TLP.EXE (plot
program) and TLP.XLS (spreadsheet file, Excel format for UP configuraticon).
The user starts NEXPERT. The knowledge base is loaded in the expert system
environment. A session is started by typing the following commands : CTRL R
(reset the knowledge base), CTRL S (Suggest a starting point) and CTRL K (start
knowcess). The starting point of the knowledge base is the hypotbeis
Platform defined. This hypothesis defines the description of the TLP. The expert
system environment interprets the knowledge base and asks questions. Examples
of questions are shown in Fig. 11 through Fig. 15 The expert system asks the
question, The platform has pontoons. The slot pontoon.exist is the current slot in
Fig. 11. This slot represents a boolean expression hence the tiles False and True.
The NOTICNOWN tile can also be selected. A selection can be made by either
double-clicking a mouse button on the required tile or by selecting the tile
followed by selecting the OK tile. The expert system environment searches for the
value of the slot column.no. Hence it asks the question How many columns do4s
the platform have. The black bar with the text Type a value allows the user to
specify a number. The value is submitted to the expert system by hitting the
ENTER key or by choosing the OK tile. A different type of screen can be
presented, see Fig. 12. Another type of question the expert system's environment
can ask is shown in Fig. 13. The environment not only asks a question but it also
gives some options. Select the required option by either double-clicking on the
option or single-click and OK. The expert system's suggestions are recorded in

65
file.Edit Espe Engydoperlls Newark :11epott:.:Wladoors :Help

False True

NOTKNOWN

Fig. 11. NEXPERT Screen Boolean input

MONAMEEMIVM4
file-,Espert ::Earidopedlo
--''
Repot...Windows .
Network .. .. : :bele

lb

I ypt: valtir

NOTKNOWN

Fig. 12. NEXPERT Screen Numerical input

66
an EXCEL spreadsheet. When the required spreadsheet is not active (not even
in the background) the environment creates a screen as shown in Fig. 14. The
left hand side of the box in the foreground explains what the system is searching
for, i.e. EXCEL and the file TLP.XLS The right hand side gives the user some
options. Select Start Application with Topic. The screen shown in Fig. 15, tells' the
user that an optimum TLP is calculated. The user can now modify a parameter.
More options are present and can be made visible by using the scroll bars.,

67
 "fIrt""
flle UN Expert Eigyelopedls Network . &PM ii1Ondwas Help

lyric: a valtic OK
Solid Carbon Fiber obese : inca 255 MPG
Solid Steel 'dress : max 155 MN

NOTKNOWN

Fig. 13. NEXPERT Screen Selection Input

Fig. 14. NEXPERT Screen Start Application

68
 '""rmt'- ' 73".
e Edft Weft Eagidspedia: Ilebiork.:: &PM :iltilftddwd liftiP

I ype value
braces
column
modify sae parameters
Plot platform 4
NOTKNOWN

Fig. 15. NEXPERT Screen Main Screen of Knowledge Base

69