You are on page 1of 135

THESIS

EVALUATION OF BLINDED TEE MATERIALS FOR GAS PRODUCTION


FLOWLINE BASED ON QUANTITATIVE RISK ASSESSMENT AND
COST BENEFIT ANALYSIS
PICHAI TANAMAITREEJITT
A Thesis Submitted in Partial Fulfillment of
the Requirements for the Degree of
Master of Engineering (Safety Engineering)
Graduate School, Kasetsart University
2005
ISBN 974-9835-51-4
ACKNOWLEDGEMENTS
I would like to express my gratitude to my thesis advisor, Asst. Prof. Dr.
Thumrongrut Mungcharoen, for his guidance and advice throughout the process,
and especially for his help with this thesis research. I also thank all thesis
committee members, Asst. Prof. Dr. Prakob Surawattanawan and Archan Anucha
Borwornthammarat, for their useful and valuable suggestions during my thesis
defense.
I must express my gratitude to Mr. David Payne, Subsurface Operations
Director and Unocal Thailand Limited for continued support and sponsorship of my
master degree study program. A special thanks to Mr. Graham Dryden, Engineering
& Construction Director at Unocal Thailand Limited and his team who provided
historical data and technical advice necessary to carry me through this process and
achievement.
I also would like to thank my classmates and colleagues for their continued
help and valuable feedback during the process.
Finally, I would like to thank all my family members and my friends for
their tremendous support and encouragement which empowered me to accomplish
the work.
Pichai Tanamaitreejitt
April 2005
i
TABLE OF CONTENTS
Page
TABLE OF CONTENTS i
LIST OF TABLES iii
LIST OF FIGURES iv
LIST OF SYMBOLS vii
LIST OF ABBREVIATIONS viii
INTRODUCTION 1
Objectives 2
Scopes 3
LITERATURE REVIEW AND RELATED THEORIES 4
Literature Review 4
1. Specialist Erosion-Resistant Materials 8
2. Sand and Particulate Erosion 8
3. Velocity, Viscosity and Density of the Fluid 9
4. Erosion-Corrosion 10
5. Erosion Management Techniques 15
6. Erosion Prediction Methods 17
7. Statistical Theories 18
8. Risk Management 24
Related Theories 6
METHODOLOGY 31
1. Overview of Prediction Model 31
2. Qualitative and Quantitative Prediction Method 31
3. Inspection Program 32
4. Data Collection 33
5. Model Diagnostics 34
6. Risk Management and Decision-Making 43
ii
TABLE OF CONTENTS (continue)
Page
RESULTS AND DISCUSSION 52
Results 52
1. Results of Multiple Regression Analysis 52
2. Results of ETA and CBA 65
Discussion 71
1. Discussion of mixture and CO2 Regressors 71
2. Rules for using Multiple Regression Analysis 73
3. Application of Multiple Regression Equation 75
4. QRA ETA Discussion 75
CONCLUSIONS AND RECOMMENDATIONS 81
Conclusions 81
Recommendations 84
LITERATURE CITED 86
APPENDICES 90
APPENDIX A Overview of Oil and Gas Development Process 91
APPENDIX B Data Sets, Data Tables, and Specifications
of Ultrasonic Thickness Gauging 105
APPENDIX C Risk Framework, ETA, and Risk Calculation
and Cost Comparison 115
iii
LIST OF TABLES
Table Page
1 Example of Data Table for Multiple Linear Regression 23
2 Results of the three-regressor multiple regression analysis 53
3 Example of the Mean Value Calculation 54
4 Results of the two-regressor multiple regression analysis 58
5 Risk Level Calculation in $ / Year 69
6 Cost-Benefit Comparison Summary 71
Appendix Table
B1 Erosion-corrosion Data Sets of Blinded Flowline Tees 106
B2 Table of CI and PI of MINITAB Output 108
B3 Table of Values of Predictors for New Observations 110
B4 Table of Percentage Points of the F-Distribution 112
B5 Specifications of Ultrasonic Thickness Gauging 113
C1 ETA and Risk Calculation for FN Curve 118
C2 Failure Rate Data in Process Components 119
C3 Cost Comparison between the CS and the DSS 121
iv
LIST OF FIGURES
Figure Page
1 Pitting behavior observed in water, CO2 and sand flows 11
2 The paths of different sized particles through an elbow 12
3 Elbow versus Blinded / Plugged Tee 14
4 Blinded or Plugged Tee with Sand Probe 15
5 Radiographic Testing of a blinded tee from gas production
flowline 16
6 Simple Linear Regression Model 18
7 Deviation of Data from Estimated Regression Model 19
8 Multiple Regression Model or Equation 21
9 Cross Section of Blinded Flowline Tee 32
10 Ultrasonic Thickness Gauging 33
11 Plots of Residuals 35
12 Normal Distribution 41
13 How MINITAB? Works Multiple Regression Analysis 41
14 Risk Analysis, Risk Assessment, and Risk Management
Processes 44
15 Risk-Related Decision Support Framework 45
16 Flowchart of Formal Safety Assessment 46
17 Event Tree Analysis of Platform Collision Probability 48
18 Regression Plot of mm/year Response Variable versus density
Regressor 56
19 Plot of Residuals versus Fitted Values of density Regressor 56
20 Scatter Plot of mm/year versus mixture Regressor 61
21 Scatter Plot of mm/year versus CO
2
Regressor 61
22 Scatter Plot of mixture Regressor versus CO
2
Regressor 62
23 Residuals Versus the Fitted Values Plot - response is mm/year 63
24 Normal Probability Plot of the Residuals - response is mm/year 63
25 Plot of CI and PI mm/year versus mixture Regressor 64
v
LIST OF FIGURES (continue)
Figure Page
26 Plot of CI and PI mm/year versus CO
2
Regressor 64
27 ETA Diagram of the CS Blinded Flowline Tee Failure 68
28 Design and Operate Plant with Risk - ALARP 78
29 Sensitivity Analysis - DSS versus CS Cost-benefit Comparison 83
Appendix Figure
A1 D-Cycle Gas Engine - Gas Development Overview 92
A2 A Survey Boat at Seismic Survey Location 93
A3 J ackup Drilling Rig 94
A4 Construction of Unocal Offshore Remote Platform 95
A5 A Tender-assisted Rig while drilling at Remote Platform
Location 96
A6 Typical Well Path West-East Cross Section 97
A7 Offshore Platforms in the Gulf Of Thailand 98
A8 Offshore Remote or Wellhead Platform 99
A9 Simplified Diagram of Process Overview of Remote Platform 100
A10 Wellheads Orientation after completion of Drilling Operation 101
A11 Wellhead components, choke and Blinded Tee 102
A12 Flowline & Sand Probe Shutdown Diagram 103
A13 Production Flowline & Sand Probe Shutdown System 104
C1 Tolerability of Risk Framework 116
C2 Frequency and Number of Fatalities FN Curve 117
C3 ETA and Risk Calculation for Evacuation Event 118
vi
LIST OF SYMBOLS
b
0
Estimate of intercept
b
1
Estimate of the partial regression coefficient
b
2
Estimate of the partial regression coefficient
C Constant, typically 100-125
C
0
Initial cost
C
k
Cost in year k; for k=1 to L
F Test statistic or F-statistic
L Project life in years
m Mean
P p-Value
r Discount rate per year
r
2
Coefficient of correlation between x
1
, x
2
R
2
Coefficient of determination
s Standard deviation of residuals
v Velocity in ft/s
x Independent or regressor variable
x
1
Represents fluid flow rate in barrels per day
x
2
Represents % of Carbon Dioxide content
Y Response or dependent variable
Predicted value of Y for given x
m Fluid density in lb/ft
3



Significance level
vii
LIST OF SYMBOLS (continue)

0

Y intercept or regression coefficient

1

Partial regression coefficient

2

Partial regression coefficient
Estimated error or random term
Standard deviation
viii
LIST OF ABBREVIATIONS
ALARP As Low As Reasonably Practicable
ANOVA Analysis Of Variance
API American Petroleum Institute
CBA Cost Benefit Analysis
CFD Computational Fluid Dynamics
CI Confidence Intervals
CS Carbon Steel
DF Degree of Freedom
DNV Det Norske Veritas
DSS Duplex Stainless Steel
EPA Environmental Protection Agency
ETA Event Tree Analysis
FAR Fatality Accident Rate
FN Frequency and Number of Fatalities
FSA Formal Safety Assessment
FTA Fault Tree Analysis
HAZID Hazard Identification
HSE Health & Safety Executive
ICAF Implied Cost of Averting a Fatality
IMO International Maritime Organization
MAD Mean Absolute Deviation
MAPE Mean Absolute Percentage Error
MSE Mean Squared Error
MSR Mean Squared Regression
NDT Non-Destructive Test
NPV Net Present Value
PI Prediction Intervals
PHA Preliminary Hazards Analysis
POB Personnel On Board
QRA Quantitative Risk Assessment
ix
LIST OF ABBREVIATIONS (continue)
RAM Reliability, Availability and Maintainability
RBI Risk Based Inspection
RCM Reliability Centered Maintenance
RIA Regulatory Impact Assessment
RMSE Root Mean Squared Error
ROI Return On Investment
RP Recommended Practice
SE Standard Error
SSE Sum of Squared Error
SSR Sum of Squares due to regression
SST Total Sum of Squares
TOR Tolerability Of Risk
UKOOA United Kingdom Offshore Operators Association
VIF Variance Inflationary Factor
VPF Value of Preventing a statistical Fatality
1
EVALUATION OF BLINDED TEE MATERIALS FOR GAS PRODUCTION
FLOWLINE BASED ON QUANTITATIVE RISK ASSESSMENT
AND COST BENEFIT ANALYSIS
INTRODUCTION
Natural gas in the gulf of Thailand has been produced since 1981 and is
undersea piped to onshore power plant, gas separation plant and downstream
petrochemical complex in the eastern seaboard industrial estate.
For gas delivery systems, offshore pipelines and flowlines are critical parts
of the piping network. In an undesirable event of a flowline failure, a volume of gas
and condensate would be discharged to the atmosphere resulting in severe
consequences e.g. harmful to people, production loss, environmental and property
damages, etc.
Referring to the flowline failure incidents in the past, the Duplex Stainless
Steel [DSS] material has been used as a replacement for the Carbon Steel [CS]
material to prevent recurrence of the incident. The DSS material has proved to be a
material of choice due to its durability and maintenance free. The only constraint of
the DSS material is the cost which is eight times higher than the CS material, at the
pr esent .
This research focuses on the evaluation of the blinded tee materials in gas
production flowlines on wellhead or remote platforms operated by Unocal
Thailand, Ltd. The evaluation process covers safety engineering aspects, a multiple
regression analysis, a quantitative risk assessment, a cost benefit analysis, and a
sensitivity analysis.
2
This research provides a brief overview of erosion-corrosion mechanisms of
blinded tees in gas production flowlines. Erosion-corrosion is a complex process
that is affected by numerous factors and subtle changes in operational conditions
that can significantly affect the damage it causes. This can lead to the scenario in
which high erosion-corrosion rate occur in one production system, but very little
erosion-corrosion occurs in other very similar systems.
This research also provides a practical solution that predicts erosion-
corrosion rate in the blinded flowline tees. A primarily cause of flowline failure
incidents and factors identified in the American Petroleum Institute [API]
Recommended Practice [RP] 14E on erosion modeling are analyzed to correlate to
historical field data. A multiple regression analysis technique is utilized to
determine the best fitted model. Then a quantitative risk assessment and a cost
benefit analysis are applied to compare the two materials being used namely; the
DSS and the CS.
The gas development overview diagram and figures of a survey boat,
drilling rigs, offshore platforms, construction of wellhead platform, simplified
process equipment and shutdown loop/system diagrams are shown in Appendix A
Appendix Figures A1 to A13.
Objectives
1. To determine correlation of significant parameters affecting the erosion-
corrosion in the CS blinded tees.
2. To conduct a quantitative risk assessment and a cost benefit analysis to
determine suitable materials to be used as blinded tees in the gas production
flowlines.
3
Scopes
1. This research focuses only on materials used as the three inch nominal
line size of the first blinded tees of natural gas production flowlines on the remote
platforms in the Gulf of Thailand.
2. Stress and deformation analyses are not covered in this research.
4
LITERATURE REVIEW AND RELATED THEORIES
Literature Review
There are many studies, related to erosion and/ or corrosion prediction
models, which can be classified into 3 categories as listed below:
1. Using Computational Fluid Dynamics [CFD] models
2. Conducting experiments in Laboratories
3. Using empirical or historical data in comparison to models
In the oil and gas industry, there are many prediction models available but
most of the models are designed and built to deal with specific oil and gas
production parameters. The following are some studies and theories related to the
erosion and/ or corrosion mechanisms, statistical theories, and risk management
process:
Sritham (1999) studied the pressure and flow distribution of the ERAWAN
pipeline system. The research was carried out on the objective of determining the
suitable tee type for two-phase flow splitting and observing the effect of inlet flow
properties of F tee flow splitting. PIPESIM was used to formulate the two-phase flow
model in pipeline. Then, CFD technique with PHOENICS program was utilized to
establish the pipe tee model. The result shows the flow and pressure profile of fluid in
pipe, which indicates that for splitting of two-phase flow, the impact tee is better than
the F tee.
Thampanitchawong (1999) studied the corrosion phenomena in pipeline
system with multiphase flow simulation model. The research covered corrosive
factors and the effect of flow pattern on the corrosion of pipeline. Corrosion model
was constructed and PHOENICS software program was used to simulate and
5
display the flow pattern within the pipe. The major effects of corrosion include flow
velocity, Temperature, pH and chemical composition, and flow regime.
Igland and Moan (2000) investigated the reliability of the pipelines during
pipelines laying and consider the ultimate strength under combined loads. They
applied the reliability theory to achieve a more uniform and consistent safety level
for semi-probabilistic design criteria for pipelines with different geometry and load
conditions. Ultimate collapse of thick tubes under combined external pressure,
tension, and bending loads are studied with the finite element method.
Mockizuki et al. (2000) analyzed the residual stress in the pipe welded joints
using finite element method. The pipe joint used had X-shaped groove and the
sequences of the welding passes were changed. The optimum welding sequence
was determined from the residual stress distribution.
Hattakitjumreon (2000) studied stress and deformation of pipe tees in the
UNOCAL pipeline system. The research covered pressure and velocity profile for
fluid as well as the stress profile of F tee and impact tee at various fluid-flow
parameters. The objective also covered the sensitivity by varying inlet flow rate and
inlet gas liquid ratio. A recommendation is made to reduce stress concentration at
the tee junction.
Barton (2003) studied different mechanisms and discussed the factors that
influence erosion especially erosion in elbows of hydrocarbon production systems
in the North Sea. A number of prediction methods are then described in details and
their results are compared with highlight difficulties inherent in erosion estimation.
Conclusions & recommendations are made on limiting erosion elbows in oil and
gas production systems.
6
Nilkitsaranont (2004) studied a systematic approach to apply diagnostic and
prognostic techniques to determine the remaining life of a gas turbine engine. The
study demonstrates the method of employing scatter results and diagnostics in terms
of engine performance which are percentage of compressor efficiency and flow
capacity degradation to predict the remaining life of a gas turbine by using Simple
Regression Techniques, including both Linear and Quadratic models. The study
also includes a methodology to evaluate the accuracy of fitting a prediction line
with the scattering data or Compatibility Check. The results demonstrate that the
most suitable model - Linear or Quadratic of the prediction line can be selected in
order to determine accurately the remaining life of the engine.
Related Theories
Oil and gas wells produce a complex multiphase mixture of components
including liquids, solids, other gases, water, and sand particles. There is not a large
amount of published data on erosion problems in the field because operating
companies are reluctant to publicize their problems and that erosion may be more
common than published data implies. Potential mechanisms that could cause
significant erosion damage are particle erosion, liquid erosion, erosion-corrosion
and cavitation. It is generally accepted that particulates, sand and proppant, are the
most common source of erosion problems in hydrocarbon systems. However, all of
the other mechanisms are equally aggressive under the right conditions.
Venkatesh (1986) provides a good overview of erosion damage in oil wells.
Regardless of the erosion mechanism, the most vulnerable parts of production
systems tend to be components in which:
1. The flow direction changes suddenly
2. High flow velocities occur caused by high volumetric flow rates
3. High flow velocities occur caused by flow restrictions
7
Components and pipework upstream of the primary separators carry
multiphase mixtures of gas, liquid and particulates and are consequently more likely
to suffer from particulate erosion, erosion-corrosion and droplet erosion. The
vulnerability of particular components to erosion heavily depends on their design
and operational conditions. However, the following list is suggested as a rough
guide to identify which components are most vulnerable to erosion the first on the
list being most likely to erode:
1. Chokes
2. Sudden constrictions
3. Partially closed valves, check valves and valves that are not full bore
4. Standard radius elbows
5. Weld intrusions and pipe bore mismatches at flanges
6. Reducers
7. Long radius elbows
8. Blinded tees
9. Straight pipes
8
1. Specialist Erosion-Resistant Materials
Specialist materials such as tungsten carbides, coatings and ceramics are
often used in chokes and highly vulnerable components. These materials are
generally hard and brittle. Brittle materials erode in a different manner. Impacts on
brittle materials fracture the surface and erosion increases linearly with impact
angle, being a maximum for perpendicular impacts. This will affect the shape of the
erosion scar and the position of maximum wear.
Most of these materials have a superior erosion resistance to steel often
orders of magnitude better. However, some coated materials are vulnerable to
erosion. Initially they may show a high resistance, but once the coating, or its
substrate fails, their resistance may rapidly reduce.
2. Sand and Particulate Erosion
Particulate erosion mechanisms have been extensively studied and there has
been some success in predicting particulate erosion rates. Important factors
determining the rate of particle erosion are:
1) The flow rate of sand and how it is transported through the flowline
2) The velocity, viscosity and density of the fluid through the component
3) The size, shape and hardness of the particles
2.1 Sand Production and Transport
The nature of the sand and the way in which it is produced and
transported also determines the rate of erosion within a production system. The
sand transport mechanism is an important aspect controlling erosion within
productions systems. Gas systems generally run at high velocities >10 m/s making
them more prone to erosion than liquid systems. However, in wet gas systems sand
particles can be trapped and carried in the liquid phase. Slugging in particular can
9
generate periodically high velocities that may significantly enhance the erosion rate.
If the flow is unsteady or operational conditions change, sand may accumulate at
times of low flow, only to be flushed through the system when high flows occur.
This and other flow mechanisms may act to concentrate sand, increasing erosion
rates in particular parts of the production system pipework.
2.2 Sand shape, size and hardness
Sand sizes seen at the surface depend on the reservoir geology, the size of
sand screens in the well and the break-up of particles as they travel from the reservoir
to the surface. Without sand exclusion measures, such as down-hole sand screens,
particle sizes typically range between 50 to 500 microns. With sand exclusion in place
particles larger than 100 microns are usually excluded. A sand particle density of about
2600 kg/m
3
is generally accepted as being representative. Particle size mostly
influences erosion by determining how many particles impact on a surface. Very small
particles ~10 microns are carried with the fluid and rarely hit walls. Larger particles
tend to travel in straight lines and bounce off surfaces. Very large particles, ~ 1mm+,
tend to move slowly or settle out of the carrying fluid and therefore they are unlikely to
do much harm. It is well established that hard particles cause more erosion than soft
particles. There is also evidence to show that sharp particles do more damage than
rounded particles. However, it is not clear whether the variability of sand hardness and
sharpness causes a significant difference between the erosion rate in production
systems associated with different wells or fields.
3. Velocity, Viscosity and Density of the Fluid
The particle erosion rate is highly dependent on the particle impact velocity.
In cases where erosion is an issue the particle impact velocity will be close to the
velocity of the fluid carrying the particle. Therefore erosion is likely to be worst
where the fluid flow velocity is the highest. Small increases in fluid velocity can
cause substantial increases in the erosion rate when these conditions prevail. In
dense viscous fluids particles tend to be carried around obstructions by the flow
10
rather than impacting on them. In contrast, in low viscosity, low density fluids
particles tend to travel in straight lines, impacting with the walls when the flow
direction changes. Particulate erosion is therefore more likely to occur in gas flows,
partly because gas has a low viscosity and density and partly because gas systems
operate at higher velocities.
4. Erosion-Corrosion
Erosion damage and corrosion damage can usually be distinguished by
inspection of the damaged flowline and by consideration of the operating
conditions. Erosion often causes localized grooves, pits or other distinctive patterns
in locations of elevated velocity. Corrosion is usually more dispersed and
identifiable by the scale or rust it generates. Erosion-corrosion is the combined
effect of particulate erosion and corrosion. The progression of the erosion-corrosion
process depends on the balance between the erosion and corrosion processes as
demonstrated by Shadley et al. (1996) amongst others. In a purely corrosive flow,
without particulates in it, new flowline components typically corrode very rapidly
until a brittle scale develops on the surfaces exposed to the fluid. After this scale
has developed it forms a barrier between the metal and the fluid that substantially
reduces the penetration rate. This is also the case when very low-level erosion is
also taking place simultaneously with corrosion. In highly erosive flows, in which
corrosion is also occurring, the erosion process predominates and scale is scoured
from exposed surfaces before it can influence the penetration rate. Corrosion
therefore contributes little to material penetration. At intermediate conditions
erosion and corrosion mechanisms can interact. In this case scale can form and then
be periodically removed by the erosive particles. This produces a pitted surface as
shown in figure 1 and can result in penetration rates of orders of magnitude greater
than those caused by pure erosion or corrosion. Erosion-corrosion mechanisms are
very complex, combining as they do two mechanisms that can be quite case
specific. This makes prediction of erosion-corrosion penetration rates for a
particular field situation very difficult. Erosion-corrosion can be avoided by
ensuring that operating conditions do not allow either erosion or corrosion.
11
Figure 1 Pitting behavior observed in water, CO
2
and sand flows
Source: Shadley et al. (1996)
4.1 Droplet Erosion
The droplet erosion mechanism is less well understood than particulate
erosion. Droplet erosion is obviously confined to wet gas and multiphase flows in
which droplets can form. The erosion rate is dependent on a number of factors
including the droplet size, impact velocity, impact frequency, and liquid and gas
density and viscosity. As many of these values are unknown for field situations, it is
very difficult to predict the rate of droplet erosion. It should also be borne in mind
that control of many of these factors in laboratory-based tests is problematical.
Therefore a great deal of care is required when extrapolating lab test results to field
conditions.
The most practical approach is to identify whether droplet erosion
could be in progress and then act to alleviate the problem. Salama & Venkatesh
(1983) state that solids-free erosion only occurs at very high velocities. High
velocities cause unacceptably high pressure losses, therefore the conditions required
12
for droplet erosion are unlikely to occur in correctly designed production flowline
systems.
4.2 Cavitation
Cavitation can be very damaging to flowline and piping components.
When liquid passes through a restriction low pressure areas can be generated, for
example downstream of a sudden step. If the pressure is reduced below the vapor
pressure of the liquid, bubbles are formed. These bubbles then collapse generating
shock waves. These shock waves can be of sufficient amplitude to damage flowline.
Cavitation is rare in oil and gas production systems as the operating pressure is
generally much higher than liquid vaporization pressures.
4.3 Sand Erosion in Elbows
a) Small, light particles b) Medium particles
c) Large, heavy particles
Figure 2 The paths of different sized particles through an elbow
Source: Barton (2003)
13
From figure 2, the paths of particles as they are carried through an
elbow. The paths depend on the particle weight and the amount of drag imparted on
the particles by the fluid as they pass through the elbow. Small light particles
require very little drag to change direction. Therefore they tend to follow the flow
in figure 2a. Large heavy particles will have a relatively high momentum and they
will hardly be deflected by the fluid flow at all. Large particles therefore tend to
travel in straight lines bouncing off the elbow walls as they go as shown in figure
2c. Figure 2 can also be viewed by considering particles of a fixed size in fluids
with different properties. Figure 2a could therefore represent particles in a highly
viscous, dense fluid and figure 2c could represent particles in a low density, low
viscosity fluid for example a gas at low pressure.
Figure 2a shows particle paths typically seen for small sand grains, of
the order of 10 microns, in a liquid flow. Figure 2b is representative of typically
sized sand grains, of the order of 200 microns, in liquid flows and figure 2c is
representative of typically sized sand in gas flows.
In general, the wear scar is located on the outside of the elbow,
however, in liquid flows the scar may be swept round to the inside, downstream
surface. Occasionally, weld intrusions immediately upstream of an elbow cause a
recirculation zone on the inside radius of an elbow. For a given flow velocity and
with all other factors equal, the erosion rate in gas is likely to be considerably
higher than that for liquid as more particles will impact on the outside of the elbow.
The maximum wear location and the penetration rate with multiphase flows are
often intermediate, but this depends heavily on the multiphase flow regime.
14
4.4 Sand Erosion in Blinded Tees
Blinded tees are generally perceived as being less prone to erosion than
standard elbows and consequently some operators routinely replace elbows with
heavy weight blinded tees when erosion problems are suspected. If a blinded tee is
orientated correctly and the flow conditions allow it, a sand plug can build up in the
dead leg of the tee. Particles passing through the tee tend to impact into this plug
instead of on the walls and consequently erosion is reduced. However, this plug
may also prevent corrosion inhibitor reaching the wall leading to corrosion
problems. If the plug does not form when, for example, the blinded tee is vertical,
or when fluid drag is high enough to keep the particles suspended. Under certain
circumstances particles concentrate and recirculate in the blinded leg, scouring its
internal surface and generating significant erosion. Alternatively, specially designed
target tees are used in which the dead leg of the tee includes a layer of soft material
usually lead that absorbs the energy of particle impacts.
Figure 3 illustrates a typical elbow and a blinded tee.
Figure 3 Elbow versus Blinded / Plugged Tee
Source: Unocal Thailand, Ltd. (n.d.)
15
Figure 4 show sand probe and blinded tee.
Figure 4 Blinded or Plugged Tee with Sand Probe
Source: Unocal Thailand, Ltd. (n.d.)
5. Erosion Management Techniques
A number of measures can be taken to monitor and avoid erosion. These
include:
1. Reducing the production rate reduces both the sand production rate and
the flow velocity through the flowline.
2. Pipework should be designed to minimize flow velocities and avoid
sudden changes in flow direction. The use of full bore valves and blinded tees in
place of elbows can also reduce erosion problems.
3. Thick-walled pipes are often used to increase the wear life of flowline.
16
5.1 Measurement & Estimation of Sand Production
Sand collection in separators is often used as an indication of sand
production. Sand production is usually associated with geological factors and hence
if one well in a field is known to produce large amounts of sand then other wells are
suspect.
5.2 Wall thickness measurements
Radiographic Testing & Ultrasonic Thickness Gauging are often used
to gauge and evaluate material loss in eroding and/ or corroding pipeline or
flowline. The photo of radiographic testing is shown in figure 5.
Figure 5 Radiographic Testing of a blinded tee from gas production flowline
Source: Unocal Thailand, Ltd. (n.d.)
Simplified diagrams of wellhead components, blast joint, choke,
blinded flowline tee and sand probe shutdown system are illustrated in Appendix A
Appendix Figures A10 - A13.
17
6. Erosion Prediction Methods
Elbows or blinded tees are relatively simple shapes and therefore a number
of alternative, empirically or theoretically derived, particle trajectory models have
been used. The simplest of these assumes that particles travel through the elbow in
a straight line in a similar manner to that shown in figure 2c and that the first wall
impact is the most damaging. This is a reasonable first order approximation in gas
flows. However in liquid flows particles are dragged round the elbow by the liquid
as shown in figures 2a and 2b and the straight particle path approximation becomes
wor s e .
Referring to American Petroleum Institute Recommended Practice, API RP
14E (2000), which is probably the simplest and most commonly used erosion model
for elbows or blinded tees in production systems, erosion model defines an
acceptable mean pipeline velocity as being:
v = m / C (1)
Where:
v is maximum allowable velocity in ft/s, above which erosion would be
expected to occur for a clean, solids-free fluid

m
is density of fluid at flowing pressure and temperature in lb/ft
3

C is a constant, typically 100-125
From the above API RP 14 E erosion model, the two factors that reflect the
erosion rate are velocity and fluid density. Campbell (2004) clarifies that this
equation was originally developed by the United States Navy during World War II.
Based upon tests with solids-free fluids, the result of the C value was 160.
However, the API recommends C values of 125 for intermittent flow and 100 for
continuous flow. This equation is used for gas/liquid mixtures as well as liquids.
18
The above equation has no demonstrated relationship with corrosion.
However, if the fluid is corrosive, the onset erosion has a catastrophic effect on the
rate at which steel disappear. Corrosion rates usually increase with velocity. This is
due to the fact that increased fluid velocity speeds up the transport of diffusion-
limited species to and from the metal surface. In this research, historical data is
statistically evaluated to predict erosion-corrosion rate. The product of this process
is a risk based inspection/maintenance program.
7. Statistical Theories
Figure 6 Simple Linear Regression Model
Source: Montgomery (2001)
7.1 Simple Linear Regression Analysis
Simple linear regression analysis is a statistical tool that can produce
predictions and explanations of data. The basic principle behind simple regression
is to use one variable to predict another variable of interest.
From the above simple regression equation:
Y = response/dependent variable

0

= Y intercept or regression coefficient
19

1

= slope or regression coefficient
= Estimated error or random term with mean zero and
unknown variance
2
.
The random error corresponding to the different observation are also
assumed to be uncorrelated random variables.
Figure 7 Deviation of Data from Estimated Regression Model
Source: Montgomery and Runger (2002)
Figure 7 shows a typical scatter plot of observed data and a candidate
for the estimated regression line. The estimated of
0
and
1
should result in a line
that is considered a best fit to the data. The German scientist Karl Gauss, 1777-
1855, proposed estimating the parameters
0
and
1
in following equation, Y =
0
+
1
x
1
+ , to minimize the sum of the squares of the vertical deviation in figure 7.
20
This criterion for estimating the regression coefficients is called method
of least square, explained by Montgomery and Runger (2002).
7.2 Models
Models are a mathematical representation of some real process.
- Mechanistic models are models which relate variables in an exact
known way such as Ohms law and Orbit of planets etc.
- Mechanistic models are rare because it is very difficult to get the
exact relationship between 2 variables.
Therefore the empirical model or multiple regression analysis is used
for this thesis.
7.3 Multiple Linear Regression Analysis
Below is extract from the book - Design and Analysis of Experiments,
Montgomery (2001):
Multiple regression analysis is the simplest of the large family of
multivariate statistical techniques. That means it deals with numerous variables at
the same time. Multiple regression is a manifest variables techniques i.e. it says
things about the variables that are actually measured. Mathematically, multiple
regression is a straightforward generalization of simple regression, the process of
fitting the best straight line through the dots on an x-y plot.
Simple and Multiple Regression techniques are closely related to the
Analysis Of Variance [ANOVA].
21
Figure 8 Multiple Regression Model or Equation
Source: Montgomery (2001)
From figure 8,
1
and
2
are called partial regression coefficients
because
1
measures the expected change in Y per unit change in x
1
when x
2
is held
constant, and
2
measures the expected change in Y per unit change in x
2
when x
1
is
held constant.

This research focuses on fitting linear regression model. An empirical
model is developed relating the erosion-corrosion rate of the blinded tees to %
Carbon Dioxide content in the flow and the fluid flow rate which consists of natural
gas, condensate and produced water. A model that describes this relationship is:
Y =
0
+
1
x
1
+
2
x
2
+ (2)
Where
Y represents the erosion/corrosion rate in millimeters per year
x
1
represents the fluid flow rate of gas, condensate and produced water
in barrels per day
x
2
represents the % of Carbon Dioxide content.
22
This is a multiple linear regression model with two independent
variables which are often called predictor variables or regressors. The term linear is
used because the multiple regression equation is a linear function of the unknown
parameters
0
,
1
, and
2
. The model describes a plane in two-dimensional x
1
, x
2
space. The parameter
0
defines the intercept of the plane.
In multiple regression model, the response variable Y may be related to
k regressor variables.
Y =
0
+
1
x
1
+
2
x
2
+ ... +
k
x
k
+ (3)
This model is called a multiple linear regression model with k regressor
variables. The parameters
j
j=0,1,,k, are called regression coefficients. This
model describes a hyper-plane in the k-dimensional space of regressor variables
{x
j
}. The parameter
j
represented the expected change in response Y per unit
change in x
j
when all the remaining independent variables x
i
(i j) are held
constant.
From the above multiple regression model, the fitted model is:
= b
0
+ b
1
(x
1
)+ b
2
(x
2
) + ... + b
k
x
k
+ . (4)
Where
b
0
is intercept,
b
1
and b
2
are coefficients of regressor variables
Indeed, one way to assess the success of the regression is the closeness
of these fitted values, namely
1
,
2
,
3
,,
n
to the actual observed Y values Y
1
, Y
2
,
Y
3
,, Y
n
.
23
Table 1 Example of Data Table for Multiple Linear Regression
Source: Montgomery (2001)
Two main points distinguish multiple regression analysis from other
techniques:
- In multiple regression analysis, there is only one dependent or
response variable and many independent or regressor variables. In simple
regression, there is only one independent variable; in factor analysis, cluster
analysis and most other multivariate techniques, there are many dependent
variables.
- In multiple regression analysis, the independent variables may be
correlated. In ANOVA, all the independent variables are arranged to vary
completely independently of each other.
This means that multiple regression analysis is useful in the following
general class of situations. While dependent variable is being observed, other
independent variables are also observed. These other variables are empirical
observations not under experimental control. The purpose is to find out which if any
of these independent variables is significantly correlated with the dependent
variable, taking into account the various correlations that may exist between the
independent variables. So typically multiple regression analysis is used to analyze
data that come from natural or empirical rather than experimental situations.
24
This makes it very useful in social psychology, social science and
biological field work. However it is inherently a technique of correlation, it cannot
of itself tell anything about the causalities that may underlie the relationships it
describes. Also, as with all statistical inference, the data should be a random
sample from some specified population; the technique will allow researchers draw
inferences from the sample to that population but not to any other.
7.4 Fitting Multiple Regression Model
Montgomery (2001) provides overview of fitting regression model as
follows:
In many problems there are two or more variables that are related, and
it is of interest to model and explore this relationship. For example, in a chemical
process the yield of product is related to the operating temperature. The chemical
engineer may want to build a model relating yield to temperature and then use the
model for predication, process optimization, or process control.
Regression models are frequently used to analyze data from unplanned
experiments, such as might arise from observation of uncontrolled phenomena or
historical records. Regression methods are also very useful in designed experiments
where something gone wrong.
8. Risk Management
Louvar and Louvar (1998) explain risk management as a simple, systematic,
logical, and orderly approach to help managers make decisions that effectively
reduce risks. The method prevents quick and ineffective decisions, made before
problems are identified decisions that would treat symptoms rather than the true
problem.
25
A risk management process is directed toward moving potential accidents
from region A to region B. Each potential accident is moved by lowering the
frequency or reducing the consequences of the accident. The value of a specific
project is based on comparing the reduced dollar cost of the consequences to the
dollar cost of achieving the reduction.
In a risk management model, the frequencies may be computed using Fault
Tree Analysis [FTA] or event tree analysis and the consequences could be
determined using cost benefit analysis. Alternatively, the frequency may be
determined through panel discussions with experts such as plant managers or
environmental experts, and the consequences may be determined by an of
historical events.
8.1 Risk Assessment
In their book, Crowl and Louvar (2002) suggest that risk assessment to
include identification and analysis of the probabilities. Consequence analysis
describes the expected damage. This includes loss of life, damage to the
environment or capital equipment, and days outage.
8.2 Quantitative Risk Analysis [QRA]
CCPS (2000) defines QRA as a method that identifies where
operations, engineering, or management systems can be modified to reduce risk.
The complexity of a QRA depends on the objectives of the study and available
information. Maximum benefits result when the QRA are used at the beginning of a
project, conceptual review and design phases, and are maintained throughout the
facilitys life cycle.
26
The QRA method is designed to provide manager with a tool to help
them evaluate the overall risks of a process. The QRAs are used to evaluate
potential risk when qualitative methods cannot provide an adequate understanding
of the risks. The QRA is especially effective for evaluating alternative risk
reduction strategies.
The major steps of a QRA study include:
1) defining the potential event sequences and potential incidents,
2) evaluating the incident consequences, the typical tools for this step
include dispersion modeling and fire & explosion modeling,
3) estimating the potential incident frequencies,
4) estimating the incident impacts on people, environment, and property,
and
5) estimating the risk combining the impacts and frequencies, and
recording the risk using a graph.
8.3 Event Tree Analysis [ETA]
The following is extracted from Chemical Process Safety
Fundamental with Applications, Crowl and Louvar (2002):
Even trees begin with an initiating event and work toward a final result.
This approach is inductive. The method provides information on how a failure can
occur and the probability of occurrence.
27
When an accident occurs in a plant, various safety systems come into
play to prevent the accident from propagating. These safety systems either fail or
succeed. The event tree approach includes the effects of an event initiation followed
by the impact of the safety systems. The typical steps in an event tree analysis are:
1) identify an initiating event of interest,
2) identify the safety functions designed to deal with the initiating
event,
3) construct the event tree, and
4) describe the resulting accident event sequences.
If appropriate data are available, the procedure is used to assign
numerical values to the various events. This is used effectively to determine the
probability of a certain sequence of event and to decide what improvements are
required.
This is used most successfully to modify the design to improve safety.
The difficulty is that for most real processes the method can be extremely detailed,
resulting in a huge event tree. If a probabilistic computation is attempted, data must
be available for every safety function in the event tree.
8.4 Cost Benefit Analysis [CBA]
The following CBA introduction is extracted from Cost Benefit
Analysis, Integrated Environmental Management, DEAT (2004) as follows:
In the private section, economic profit is often used as an indicator of
economic efficiency. This need not be valid; the market may be distorted or the
decision may involve non-profit projects or the introduction of the new government
28
policies. The CBA offers an alternative test of efficiency for such situation. As its
name suggests, the CBA simply compares all the expected present and future
benefits of a project or policy with its present and future costs. In general future
costs and benefits appear less important than present one, for this reason the CBA
attaches a progressively lower weight to cost and benefits the further in the future
they appear.
8.5 Cost Benefit Analysis Definition
The term cost benefit analysis is used frequently in business planning
and decision support, activities. However, the term itself has no precise definition
beyond the implication that both positive and negative impacts are going to be
summarized and compared. The term also has no universally agreed spelling. It is
written as cost benefit, cost/benefit, or cost-benefit, for instance. Because the term
cost benefit analysis does not refer to any specific approach or methodology, the
business person who is asked to produce one should take care to find out what is
expected or needed. The term covers several varieties of business case analysis,
such as:
- Return On Investment [ROI]
- Financial Justification
- Cost of ownership analysis
All of these approaches to cost benefit analysis attempt to predict the
financial impacts and other business consequences of an action. All these
approaches have the same structural and procedural requirements for building a
strong, successful business case. They differ primarily in terms of:
- How are cost and benefit defined in practical terms?
- Which costs and benefits are included for analysis?
- Which financial metrics are important for decision makers and planners?
29
Financial justification, in other words, is a business case analysis that
helps decision makers decide whether or not to go forward with a proposed action.
It is distinguished from other business case approaches only by the special emphasis
on financial decision criteria. Otherwise, a strong financial justification has the
same characteristics as other kind of business cases.
There are many definitions of the CBA in the readily available
literature. In various ways, they all say about the same thing:
Benefit-cost analysis is a method of evaluating the relative merits of
alternative public investment projects in order to achieve efficient allocation of
resources. It is a way of identifying, portraying and assessing the factors which
need to be considered in making rational economic choices. In principle, it entails
little more than adjusting conventional business profit-and-loss calculations to
reflect social instead of private objectives, criteria, and constraints in evaluating
investment projects.
1) The purpose of the CBA is to improve or ensure efficiency of
resources allocation so as to increase economic and perhaps social welfare. It is a
valuable tool, but by definition it cannot incorporate certain important aspects into
the analysis, e.g. political non-efficiency objectives.
2) Formal cost-benefit analysis is in principle a rigorous, quantitative,
and data-intensive procedure, which requires identification of all nontrivial effects,
categorization of these effects as benefits or costs, quantitative estimation of the
extent of each benefit or cost associated with an action, translation of these into a
common metric such as dollar, discounting of future costs and benefits into the
terms of a given year, and summary of all costs and benefits to see which is greater.
The logic of cost-benefit analysis also demands that these sums be compared across
alternatives, a point neglected even by many of its proponents.
30
3) The central issue in cost-benefit analysis is the aggregate gain or
loss to society as a whole from a particular decision, and not the identification of
winners and losers. In the economist's terms, the technique is concerned with the
efficient allocation of resources, not the distribution of income.
In the 1950s and 1960s the CBA was developed in the United States in
connection with multiple use of water resources. In the 1950s to the 1970s it was
extended into fields such as manpower programming, transportation and health
analysis. The initial cost-benefit guide for the Government of Canada was prepared
in 1961 as a background paper for resources for tomorrow conference - Sewell et al
(1965). Since the 1960s, the CBA has increasingly also became a tool of internal
government management.
It is also frequently a requirement of public regulatory process. For
example, the CBA plays a major role in Regulatory Impact Assessment [RIA]
undertaken by the US Environmental Protection Agency [EPA].
31
METHODOLOGY
1. Overview of Prediction Model
The first objective of this research is to carry out a study of the use of the
multiple regression equation in predicting the erosion-corrosion rate in the blinded
flowline tees of gas production system.
In this section, the multiple regression analysis technique is applied to
predict the erosion-corrosion rate. The ability to predict the erosion-corrosion rate
provides decision makers with a tool to set up a risk based inspection program for
blinded flowline tees inspections.
2. Qualitative and Quantitative Prediction Method
Nilkitsaranont (2004) explains the Qualitative and Quantitative terms as
listed below:
2.1 Qualitative methods are generally characterized by the fact that little or
no quantitative information is available and the forecast is mainly based on
sufficient qualitative knowledge. They involve subjective estimation through the
opinions of the experts. In this thesis, erosion/corrosion rate is based on historical
measurement data and therefore obviously does not fall into this category.
2.2 Quantitative forecasting procedures rely almost entirely on historical
data. Statistical methods clearly define how the forecast is determined, the logic is
clearly stated and the operations are mathematical.
32
3. Inspection Program
Figure 9 shows a cross section of a used blinded flowline tee, an arrow mark
points to the most eroded-corroded area. This eroded-corroded area is measured for
thickness evaluation by means of Non-Destructive Testing [NDT] technique.
Figure 9 Cross Section of Blinded Flowline Tee
Source: Unocal Thailand, Ltd. (n.d.)
In the past the thickness was measured by means of the radiography
technique. At the present, the ultrasonic thickness gauging has been used and is
very effective. In this research, the erosion-corrosion data records were measured
by PANAMETRICS

36DL PLUS.
33
Figure 10 Ultrasonic Thickness Gauging
Source: Unocal Thailand, Ltd. (n.d.)
Figure 10 is an illustration of the application of the ultrasonic thickness
gauging technique. Specifications of the PANAMETRICS

36DL PLUS, are


shown in Appendix B Appendix Table B5.
4. Data Collection
The motivation behind this research is to correlate the historical data sets to
predict the erosion-corrosion rate. This statistical approach was well accepted and
was approved by company management because it is very cost effective compare
with other approaches.
Fifty eight historical data sets were randomized from 422 erosion-corrosion
records of the CS blinded flowline tees collected from Erawan, Baanpot, Satun and
Platong remote platforms during the year 1998 to 2002. The total gas production
was produced from 582 wells. The 58 data sets are used as response variable input
data in multiple regression analysis. The details of the data sets are listed on the
Appendix Table B1 in Appendix B.
34
5. Model Diagnostics
Another fundamental research concern is how to measure the suitability of a
particular prediction model for a given data set. In most prediction situations,
accuracy is treated as the overriding criterion for selection of a forecasting method.
In many instances, the word accuracy refers to goodness of fit, which in turn refers
to how well the forecasting/prediction model is able to reproduce the data that are
already know.
5.1 Measuring of Predicting Accuracy Parameters
Fitting a regression model requires several assumptions, estimation of
the model parameters requires the assumption that the errors are uncorrelated
random variables with mean zero and constant variance. Test of hypotheses and
interval estimation require the errors be normally distributed. In addition, the
assumption was made that the order or the model is correct.
5.2 Residuals Analysis
The difference between the fitted model and the observation is called
the error or residual which is explained by Nilkitsaranont (2004) as follows:
The residual are informative in demonstrating the adequacy of the
regression model. A plot of the residuals will indicate whether the regression model
is a good fit to the data. Any structure to the residuals plot will indicate that there
are patterns in the data that are not modeled by the regression line Montgomery
and Runger (2002). Analysis of the residuals is frequently helpful in checking the
assumption that the errors are approximately normally distributed with constant
variance, and in determining whether additional terms in the model would be
useful.
35
The plots will usually look like one of the four general patterns shown
in figure 11. Pattern (a) represents the ideal situation, but patterns (b), (c) and (d)
represent anomalies. If the residuals appear as in (b), the variance of the
observations may be increasing with time, or the magnitude of the variance may be
increasing with time. Pattern (c) also indicates the inequality of the variance.
Pattern (d) also indicates model inadequacy.

Figure 11 Plots of Residuals
Source: Montgomery and Runger (2002)
36
The following is extract from the SPSS Training Department (1999):
5.3 Mean Error
The mean error is the mean difference between the actual series value
and the fitted value. The mean error only provides a guide as to whether the model,
on average, over or under predicts, and error adjustments are necessary. If the mean
error is positive, then the model has a tendency to under predict. Alternatively, if
the mean error is negative, then the model has a greater tendency to over predict.
The criticism of this measure is that positive and negative errors
cancel each other out when averaging is performed, and this is often undesirable.
The mean absolute error was developed to address this weakness.
5.4 Mean Absolute Error
The mean absolute error, also referred to as the Mean Absolute
Deviation [MAD], takes the average of the errors and is often used as the
primary measure of fit. This is a considerably higher error than that suggested by
the mean error, and better represents the magnitude of the typical error.
5.5 Mean Percent Error
As its name suggests, the mean percent error expresses the error as a
percentage of the values in the original series. This number is determined by
dividing the error for each point by the actual series value at that point, averaging
these ratios, and finally multiplying by 100. This number has the advantage of
expressing the error on a percentage basis, which is often desirable. However, it
shares with the mean error summary, the weakness that positive and negative errors
cancel each other out.
37
5.6 Mean Absolute Percentage Error [MAPE]
Both the mean error and the mean absolute error must be interpreted
with reference to the original units of measurement. The MAPE is sometimes
preferred because it is a percentage and thus is a relative measure. Also positive and
negative errors do not cancel. MAPE is obtained by taking the absolute error for
each point, dividing it by the actual series value, averaging these ratios across all
the points, and multiplying by 100. The MAPE is often used as it measures the
percentage prediction.
5.7 Sum of Squared Error [SSE]
Another measure of forecasting/predicting accuracy is the sum of
squared error, obtained by squaring each of the errors and summing them. Since the
errors are squared, the positive and negative errors do not cancel. Minimizing the
SSE is also known as the least squares method and is integral to the derivation of
the best model in Ordinary Least Squares Regression.
SSE increases monotonically with the length of the series since the
squared errors are summed, which again makes it a difficult number to interpret.
For these reasons, the following measures derived from the SSE are reported.
5.8 Mean Squared Error [MSE]
MSE divides the SSE by the degrees of freedom of error. It is thus a
variance measure of the error and does not increase with the length of the series, as
does the SSE. However, because its metric is still based on the square of the
original units, the square root of this measure - root mean squared error is more
often reported.
38
5.9 Root Mean Squared Error [RMS] or [RMSE]
The RMS is equal to the square root of the MSE. The MSE can be seen
as the variance of the error terms, and the root mean square error can be thought of
as the standard deviation of the error terms. A small RMS is preferred since this
signifies that the error terms do not have a large spread, that is, the errors
concentrate near zero, which means a good fit. One of the differences between the
measures based on absolute error values of MAD and MAPE and measures based
on the squared value of the error is that the latter penalizes a forecast much more for
extreme deviations or a large error than it does for small ones. For instance, the
MAD calculation counts an error of 2 as twice as much as an error of 1, where the
MSE calculation squares an error of 2 and thus counts it as 4 times as much as an
error of 1. Thus, adopting the criterion of minimizing mean squared error implies
that one would prefer to have several small deviations from the forecast value rather
than one large deviation.
5.10 Standard deviation of the residuals
The variance of the estimated error or residuals is given by:
s
2
= (Y
i
-
i
)
2
(5)
n-2
The denominator of the above expression denotes the degrees of
freedom, which can be defined as the number of data points minus the number of
estimated parameters variance -
0
and
1
. The standard deviation, s, is simply the
square root of the variance.
This measure is useful when two models give similar high values for
R
2
or coefficient of determination. There could be a large difference in the standard
deviation of the residuals for the two models.
39
5.11 Coefficient of determination [R
2
] and Adjusted R
2
The coefficient of determination is the most general method for
assessing the goodness of fit. R
2
is then defined as the ratio of the explained
variation over the total variation. In Regression the square of the correlation
between Y - the response variable and

- the estimated Y value based on the set of
explanatory variables is denoted R
2
. This statistic is often called the coefficient of
determination.
R
2
varies between 0 and 1, where 0 indicates complete lack of fit and 1
indicates a strong linear relationship between Y and x value. For example, if R
2
is
equal to 0.9, this means that the model accounts for 90% of the variation in the
value Y.
There are several misconceptions about R
2
, the R
2
does not measure
the appropriateness of the model. Even if Y and x are related in a nonlinear fashion,
R
2
will often be large. For example, R
2
for the regression in figure 11 (b) will be
relatively large, even though the linear approximation is poor. Finally, even though
R
2
is large, this does not necessarily imply that the regression model will provide
accurate prediction of future observations Montgomery and Runger (2002).
Because the R
2
always increases as terms are added to model, some
regression model builders prefer to use the Adjusted R
2
Statistic defined as:
Adjusted R
2
= 1-(n-1/n-p)(1- R
2
)
2
(6)
In general, the Adjusted R
2
Statistic will not always increase as
variables are added to the model. In fact, if unnecessary terms are added, the value
of Adjusted R
2
will often decrease.
40
5.12 Significance level
In hypothesis testing, the significance level is the criterion used for
rejecting the null hypothesis. The significance level is used in hypothesis testing
as follows: First, the difference between the results of the experiment and the null
hypothesis is determined. Then, assuming the null hypothesis is true; the
probability of a difference that large or larger is computed. Finally, this
probability is compared to the significance level. If the probability is less than or
equal to the significance level, then the null hypothesis is rejected and the
outcome is said to be statistically significant. Traditionally, researchers have used
either the .05 level, sometimes called the 5% level or the .01 level, although the
choice of levels is largely subjective. The lower the significance level, the more
the data must diverge from the null hypothesis to be significant. Therefore, the .01
level is more conservative than the .05 level. The Greek letter alpha - is often
used to indicate the significance level. This significance test can be applied to the
regression model because the primary assumption of regression is that the
residuals or errors should be normally distributed.
5.13 Normal Distributions
Normal distributions are a family of distributions that have the same
general shape. They are symmetrical, with scores more concentrated in the middle
than in the tails. Normal distributions are sometimes described as bell-shaped. The
height of a normal distribution can be specified mathematically in terms of two
parameters: the mean [m] and the standard deviation [s].
41
Figure 12 Normal Distribution
Source: Montgomery and Runger (2002)
Figure 13 illustrates how Multiple Regression Analysis in MINITAB

program works. The historical data is the input and the output is the estimates of b
0
,
b
1
, and b
2
.
Figure 13 How MINITAB

Works Multiple Regression Analysis


Source: Montgomery (2001)
42
5.14 Hypothesis Testing in Multiple Regression Model
Montgomery (2001) provides the following hypothesis testing in
multiple regression. In multiple linear regression problems, certain tests of
hypotheses about the model parameters are helpful in measuring the usefulness of
the model. There are several important hypothesis-testing procedures.
These procedures require that the errors

in the model be normally


and independently distributed with mean zero and variance
2
.
As a result of this assumption, the observations Y
i
are normally and independently
distributed with mean
0
+
j
k
=1
j
x
ji
and variance
2

.
5.15 The Use of p-value in Hypothesis Testing
The p-value [P] approach has been adopted widely in practice. The
p-value is the probability that the test statistic will take on value that is at least as
extreme as the observed value of the statistic when the null hypothesis H
0
is true.
Thus a p-value conveys information about the weight of evidence against H
0
, and so
a decision maker can draw a conclusion at any specified level of significance. More
formally, p-value is defined as the smallest level of significance that would lead to
rejection of the null hypothesis H
0
.
It is customary to call the test statistic and the data significant when
the null hypothesis H
0
is rejected; therefore, the p-value can be interpreted as the
smallest level at which the data are significant. Once the p-value is known, the
decision maker can determine how significant the data are without the data analyst
formally imposing a pre-selected level of significance.
It is not always easy to compute the exact p-value for a test. However,
most modern computer programs for statistical analysis report p-value, and they can
be obtained on some handheld calculators.
43
6. Risk Management and Decision-Making
Referring to HSE (1999), UK Health & Safety Executives [HSE] views on
decision-making on safety issues have most recently been explained in a Reducing
Risks, Protecting People.
The HSEs approach is based on a Tolerability Of Risk [TOR] framework
Appendix Figure C1 in Appendix C. It applies to risk in a broad sense, including
not just the risks of individual harm and societal risks, but also the perception of
hazards and associated ethical and social considerations, such as aversion to large
multiple-fatality accidents. It divides into 3 regions:
1) Unacceptable risks regarded as unacceptable except in extraordinary
circumstances such as wartime, whatever their benefits. Activities causing such
risks would be prohibited, or would have to reduce the risks whatever the cost.
2) Tolerable risks that are tolerated in order to secure benefits. In this
region, risks are kept As Low As Reasonably Practicable [ALARP], by adopting
reduction measures unless their burden in terms of cost, effort or time is grossly
disproportionate to the reduction in risk that they achieve.
3) Broadly acceptable risks that most people regard as insignificant.
Further action to reduce such risks is not normally required.
Simplified diagram of ALARP Risk Framework is illustrated in Appendix C
Appendix Figure C1.
Below flowchart - figure 14 illustrates the processes of Risk Analysis, Risk
Assessment, and Risk Management. This flowchart is a very good tool for risk
communication purpose because it can differentiate between Risk Analysis, Risk
Assessment, and Risk Management concepts in terms of scopes and objectives. The
flowchart can also be utilized to explain this thesiss process overview in a
simplified method.
44
Figure 14 Risk Analysis, Risk Assessment, and Risk Management Processes
Source: Det Norske Veritas (2001)
6.1 Framework for Risk Related Decision Support
The UK Offshore Operators Association, UKOOA (1999) has
developed a framework to assist risk-related decision-making which helps decision-
makers choose an appropriate basis for their decisions. Below figure 15 shows a
risk-related decision support framework.
45
Figure 15 Risk-Related Decision Support Framework
Source: UKOOA (1999)
The above framework takes the form of spectrum of decision bases,
ranging from those decisions dominated by purely engineering concerns to those
where company and societal values are the most relevant factors.
Down the right-hand edge of the framework are typical characteristics
with indicate the decision context; these can be used to help the user determine the
context for a specific decision.
Once this level has been identified, reading horizontally across the
framework shows the suggested balance of decision bases to be taken into account
in the decision. Some means of calibrating or checking the decision basis are shown
on the left-hand side of the framework.
46
6.2 Overview of QRA and CBA
The second objective of this thesis is to conduct the QRA and the CBA
for evaluation of the materials to be used as the gas production blinded flowline
tees. The quantitative findings of the QRA and the CBA can be presented and
communicated to all stakeholders.
International Maritime Organization IMO (1997) develops a Formal
Safety Assessment [FSA] which consists of five-step process, involving hazard
identification, risk assessment, development of risk control options, cost-benefit
assessment, and making recommendations for decision-making.
The purpose of FSA is to help develop risk-based regulation, and hence
it should not be confused with risk assessment used in support of a safety case,
although it uses many of the same techniques.
Figure 16 illustrates the FSA flowchart.
Figure 16 Flowchart of Formal Safety Assessment
Source: IMO (1997)
47
In the case of financial justification, firstly the ETA is used as
quantitative risk assessment tool to quantify likelihood or probabilities and
consequences. Then values of probabilities and consequences are multiplied for risk
exposure calculation. Finally, cost benefit analysis is applied to compare the cost
between the CS and the DSS blinded flowline tees.
6.3 ETA Process and Construction
The following is extract from Marine Risk Assessment by Det Norke
Veritas (2001);
The ETA is a logical representation of the various events that may
follow from an initiating event e.g. a flowline failure. It uses branches to show the
various consequences models. It may be used to quantify system failure event to
various consequences models. It may also be used to quantify system failure
probabilities, where several contributory causes can only arise sequentially in time.
Construction starts with the initiating event and work through each
branch in turn. A branch is defined in terms of questions e.g. protective device fails.
The answers are usually binary e.g. yes or no, but there can also be multiple
outcomes e.g. 100%, 20% or 0% in the operation of a control valve. Each branch is
conditional on the appropriate answers to be previous ones in the tree.
Usually an event tree is presented with the initiating events on the left
and the outcomes on the right. The questions defining the braches are placed across
the top of the tree, with upward branches signifying yes and downward ones for no.
Quantification of event tree is relatively simple, and is readily
performed by hand, although spreadsheets or computer models are increasing used
to automate the multiplication task. A probability is associated with each branch,
being conditional probability of the branch i.e. the answer yes or no to the branch
question given answers of all branches leading up to it.
48
In each case, the sum of the probabilities of each branch must be unity.
The probabilities of each outcome are the products of the probabilities at each
branch leading to them. The sum of probabilities for all outcomes must be unity as
well. This provides a useful check on the analysis.
Figure 17 illustrates an example of the ETA applied to a marine hazard.
Figure 17 Event Tree Analysis of Platform Collision Probability
Source: Det Norke Veritas (2001)
The ETA has proved to be useful tool for major accident hazard
assessment and was used by the Health & Safety Executive for the assessment of
risks to the public from serious accidents. In major accidents, the ETA is used for
evaluation of possible consequences following a release of toxic/flammable vapor
cloud from a process and to analyze the effects on plant, personnel, general public
and the environment.
49
6.4 Financial Justification
The purpose behind almost any risk assessment is to support some
decision making on safety matters.
Hence it is desirable for risk assessment to produce a clear view on the
issues of operational, economic, social, political and environmental factors. The
decision-making process must take account of the values of the company and
society, and may rely on engineering judgment, good practice and codes &
standards.
To answer the question of how safety is safe enough, the ALARP
principle is used as the risk criteria - HSE (1999).
6.5 Cost-Benefit Analysis
Det Norske Veritas (2001) explains that CBA is a technique for
comparing the costs and benefits of a project, developed to help appraise public
sector projects. In safety assessment, it is usually used to assess additional safety
measures on a project by comparing the cost of implementing the measure with the
benefit of the measure, in terms of the risk-factors cost of the accidents if would
avert.
In a conventional CBA, future costs and benefits are converted to
present values, discounting those that occur in the future. Discounting financial
quantities is justified because money is always more useful now than in the future,
due to the opportunities to invest and make it grow. Discounting risks to life in the
same way is much more questionable.
50
It can be argued that it is better to reduce risks now than in the future,
and so immediate risk reductions should be valued more highly than future ones.
However, when considering the benefits of a given safety measure, it is not clear
that the lives of present workers are any more valuable than the lives of future
workers.
In fact, given the progressive increase in real terms of the Value of
Preventing a statistical Fatality [VPF] used in decision-making, the reverse may be
true, Discounting the cost of future fatalities is widely regarded as unethical.
In offshore industry, VPFs for decision-making purposes have been in
the range of 1m 10m, although few have been published.
In order to ensure a bias in favour of safety, it is preferable to calculate
risks and costs ratio, known as the Implied Cost of Averting a Fatality [ICAF] from
the lifetime risk benefits with no discounting and the present value of costs with
conventional discounting:
Present value of lifetime cost of measure
ICAF = Reduction in lifetime statistical fatalities
Below is equation to calculate present value:
PV= C
o
+ C
k
{1- (1 + r)
-L
} / r (7)
Where:
PV = present value of cost
r = discount rate per year; for example, 7% per year = 0.07
L = project life in years
C
o
= initial cost
C
k
= cost in year k; for k=1 to L
51
CMPT (1999) suggests that if the ICAF is less than 1m, the measure
would be cost-effective, and hence reasonably practicable even if individual risks
are low, and would normally be adopted.
If ICAF are in the range of 1m 10m, the measure would not be
cost-effective, but might be considered reasonably practicable, especially if the
individual risks are high in the ALARP zone.
If the ICAF exceeds 10m, the measure would not be considered
reasonably practicable, and the money could usually be spent more effectively on
other safety measures.
52
RESULTS AND DISCUSSION
Results
1. Results of Multiple Regression Analysis
A combination of flowline failure incidents and factors identified in the API
RP 14 E erosion modeling are analyzed to correlate with the historical field data
sets. Carbon Dioxide or CO
2
has been identified as a primary cause of the flowline
failure incidents.
The above mentioned factors or regressor variables are as follows:
- density fluid density at flowing pressure and temperature in lb/ft
3
.
- mixture fluid flow rate in barrels/day
- CO
2
% content in fluid flow
In this multiple regression model, a response variable is erosion-corrosion
rate in millimeters per year or mm/year.
1.1 Fitting Multiple Regression Model by Using MINITAB
MINITAB

is a computer software program which is being recognized


as statistical analysis tool in research study, explained by Siroratanasret (2003).
The statistical software MINITAB

is used to analyze the input data for


selecting the most fitted regression model. The table 2 shows the MINITAB output
of a three-regressor multiple regression analysis. The three regressors are the fluid
flow, the fluid density, and CO
2
% content.
53
Table 2 Results of the three-regressor multiple regression analysis
In regression, the goal is to get an estimate which best predicts the
outcome. The outcome in this case is erosion-corrosion rate in mm/year. If all
predictor variables are measured and recorded, the best guess at the erosion-
corrosion rate is given by above multiple regression equation in table 2.
Each of the estimates, coefficients indicated with Coef, has a Standard
Error [SE]. This is a measure of how variable the estimate is likely to be. The 95%
Confidence Intervals [CI] of the coefficient is normally applied to most of the
studies.
Next part of the model is T which is given by Coef / SE Coef. T is not
very much useful on its own but it does provide P or p-value which is called the
probability of resulting occurring.
Now F is examined, F is equals to 195.64 with Degree of Freedom
[DF] = 3 and 54. F is a test statistic that is used in ANOVA and regression. F is
given by MSregression / MSerror, where MS stands for Mean Squares.
54
The Mean Squares are given by the Sums of Squares [SS] divided by
the DF. The Sum of Squares is main concept of regression which is also called
ordinary least squares regression. These are the squares that make the SS. The SS is
short for Sums of Squared deviation from the regression.
The first Sum of Squares is the total Sum of Squares. These are
calculated by finding the residual difference between each value and the mean,
squaring it, and then adding them up.
The mean is like a regression equation with no predictor variable. To
determine a value of the mean, the calculation can be performed as illustrated in
table 3 below:
Table 3 Example of the Mean Value Calculation
In regression equation, a predicted variable for each case is calculated
to find the residual. Next the residuals are squared and added them up to obtain the
error sum of squares [SSerror]. The point of regression is to find the values for the
equation that give the error sum of squares, this is the least squares that was
mentioned earlier. This will always be lower than the total sum of squares [SStotal]
but the question is how much lower. SStotal and SSerror are determined to calculate
the regression sum of squares [SSregression] by subtracting error from the total:
55
SSregression = SStotal SSerror (8)
This gives the three SS that are listed on table 2. The trouble with Sums
of Squares is that they are dependent on the number of records, and the number of
variables, so it is a must to divide the regression by the regression DF, which is
given by k, where k is the number of regressor variables, and error DF is given by
N - k -1.
The SSregression is divided by the SSerror to obtain F value, the
smaller the error the larger the F. In addition, more predictors, for the same value of
R
2
gives a lower F. F has a distribution that depends on the DF. Therefore, the DF
must be reported when F is reported. In this case, F = 195.64 which is larger than F-
value from the Appendix Table B4 with DF = 3 and 54.
The above mentioned model from MINITAB output has a very good R
2
and Adjusted R
2
. This is the most common way to statistically consider a model.
The null hypothesis is usually the hypothesis of zero effect. There are
two sets of null hypothesis that the predictions are no better that chance. This tested
using the F test of the R
2
. The above MINITAB output easily rejected this null
hypothesis.
Second, is the set of null hypotheses about the parameter estimates or
coefficients. For mixture and CO
2
regressors, the null hypothesis of no effect is
rejected because their p-values are zero which is less than 0.05.
From figure 18, the regression plot of erosion-corrosion rate in
mm/year versus the fluid density regressor shows very low results of R
2
= 6.9 %
and Adjusted R
2
= 5.2%. The p-value of fluid density regressor is 0.376 which is
larger than 0.05 so the null hypothesis of no effect cannot be rejected.
56
Figure 18 Regression Plot of mm/year Response Variable versus density Regressor
Figure 19 Plot of Residuals versus Fitted Values of density Regressor
From figure 18 and figure 19, the Regression Plot of mm/year
Response Variable and Residuals Versus Fitted Values Plot of density Regressor,
the fluid density regressor shows a very narrow range of scatter plots because the
majority of the fluid flow is natural gas.
57
Another factor that keeps the fluid density in the very narrow range of
scatter plots is the gas production goal. The amount of produced water in the fluid
flow rate is being kept at the minimum because the costs of the treatments of the
produced water are quite high. The water shut-off technique has been applied
down-hole to prevent excessive water from flowing into the hydrocarbon flow.
The fluid density regressor shows a lack of symmetry of the normal
distribution with p-value is larger than 0.05. So it can be concluded that the fluid
density is not a significant regressor.
During piping design phase, the fluid density factor identified in the
API RP 14E is used for erosional velocity calculation. Since the C is a constant in
the API RP 14E erosion equation, the fluid density factor plays an important role in
designing of piping network and related process equipment.
1.2 Test for Significance of Regression
The test for significant of regression is a test to determine if there is a
linear relationship between the response variable Y and a subset of the regressor
variables x1, x2, x
k
.
The appropriate hypotheses are
H
0
:
1

=
2
= ... =
k

= 0
H
A
:
j

0 for at least one j
Rejection of H
0
in equation 1.1 implies that a least one of the regressor
variables x1, x2, x
k
contributes significantly to the model. The test procedure
involves an analysis of variance partitioning of the total sum of square SS
T
into a
sum of squares due to the model or to regression and a sum of square due to
residual or error, say:
58
SS
T
= SS
R
+ SS
E
(9)
Now if the null hypothesis H
0
:
1

=
2
= ... =
k

= 0 is true with the
following test procedure:
F
0
= SS
R
/ k = MS
R
/MS
E
(10)
SS
E
/

(n-k-1)
The F-Statistic is always positive. If the test statistic is below the
critical value, the null hypothesis will not be rejected otherwise H
0
will be rejected.
To reject H
0
if F
0
exceeds F
,k,n-k-1
. Alternatively, the p-value approach
is used to test hypothesis and thus, reject H
0
if the p-value for the statistic F
0
is less
than . The test is summarized in below table 4:
Table 4 Results of the two-regressor multiple regression analysis
59
From table 4, the above equation determines the best erosion-corrosion
prediction rate in mm/year. This is given by R
2
. R
2
and Adjusted R
2
are very high
i.e. 91.5% and 91.1%. The p-value of 0.000, a very high R
2
and Adjusted R
2
indicate that the prediction model is highly significant.
The table 4 also includes the computed value of T, the p-value called
significance level, and the decision should be given the specified value of . Notice
that the computed p-value may differs slightly from manually calculated value and
that the p-value is reported to be P=0.0000. Many software packages will not report
an actual p-value less than 0.0001 and instead return a default value. That is the
case here.
1.3 Hypothesis Testing for MINITAB Output
Once the regression is fitted, there are 2 tests to determine how well the
model fit the data as listed below:
1) Test for Significant of Regression or the overall fit
2) Tests on Individual Regression Coefficients
Testing individual terms, if an independent variable does not contribute
significantly to predicting the value of Y, the coefficient of that variable will be 0.
The test of these hypotheses determines whether the estimated coefficient is
significantly different from 0.
The following tests are performed to test the multiple regression
analysis in table 4:
- For testing H
0
:
0

= 0 versus H
A
:
0


0, p-value from the output is =
0.000 < 0.05 at the 95% confident level, H
0
is rejected.
60
- For testing H
0
:
1

=
2
= 0 versus H
A
: at least one of
1

,
2


0.
F-value from the MINITAB output is 294.15 is larger than F
(k, n-k-1)
value which is 3.96 in Appendix Table B4 at the = 0.05 level, H
0
is rejected. So it
can be concluded that at least one of
1

,
2


0.
- For testing H
0
:
1

= 0 versus H
A
:
1


0,
p-value from the output is = 0.000 < 0.05 at the 95% confident level,
H
0
is rejected.
- For testing H
0
:
2

= 0 versus H
A
:
2


0,
p-value from the output is = 0.000 < 0.05 at the 95% confident level,
H
0
is rejected.
Based upon the above tests, the overall regression is significant. Also,
each of two regressors has the p-value below 0.05. So it can be concluded that the
mixture and CO
2
regressors are significant.
A test statistic is F, which equals to 294.15, is more than previous
model shown in the table 2. As mentioned earlier, the higher the F, the smaller the
errors.
1.4 Scatter Plots and Regression Line
A scatter plot is a visual representation of the relationship between x
and Y variables. First the x and Y axes are drawn with equally spaced markings to
include all values or that variable that occur in the data. The pair of x and Y data
will be represented as vectors or points in this graph. The point is plotted by finding
the intersection of the x and Y scores of that pair of values. The regression line is
drawn by plotting upon the x and values.
61
Figure 20 Scatter Plot of mm/year versus mixture Regressor
Figure 21 Scatter Plot of mm/year versus CO
2
Regressor
From figure 20, the erosion-corrosion rate in mm/year versus the fluid
flow in barrels - mixture indicates a strong positive linear relationship. From figure
21, the erosion-corrosion rate in mm/year versus CO
2
content indicates a positive
linear relationship. Therefore, a multiple regression is quite certain.
62
Figure 22 Scatter Plot of mixture Regressor versus CO
2
Regressor
Figure 22 shows a slightly correlation between mixture regressor and
CO
2
regressor. Since this estimated multiple regression equation is to be used only
for predictive purpose so multicollinearity is usually not a serious problem. If
multicollinearity is extreme, the MINITAB program will refuse to carry out the
analysis.
To determine the multicollinearity problem, a Variance Inflationary
Factor [VIF] is widely used for comparing a calculated VIF to a set value of 10.
The following is the VIF equation:
VIF
j
= 1 / (1-r
2
j
) (11)
r
12
= x
1
, x
2
(( x
2
1
) ( x
2
2
))
0.5
Where
r
2
j
= coefficient of correlation between x
1
, x
2
63
The calculated VIF is equal to 5.3, less than the set value of 10.0.
Therefore, the multicollinearity is not a problem for the two-regressor multiple
regression model, as suggested by Pongsakornrungsilp (2004).
Figure 23 shows the plot of standardized residual against the predicted
values which indicates that the residuals appear to have a constant variance.
Figure 23 Residuals Versus the Fitted Values Plot - response is mm/year
Figure 24 Normal Probability Plot of the Residuals - response is mm/year
From figure 24, the normal probability plot is likely to be a straight
line. So it can be assumed that model error terms are approximately distributed.
64
Below figures 25 and 26 show Regression Plot of CI and Prediction
Intervals [PI] correlated between mm/year response variable and mixture
regressor, then CO
2
regressor respectively. Details of MINITAB output are
included in Appendix Table B2.
Figure 25 Plot of CI and PI mm/year versus mixture Regressor
Figure 26 Plot of CI and PI mm/year versus CO
2
Regressor
65
Results of prediction model can be used to determine erosion-corrosion
rates. The ability to predict the erosion-corrosion rates, thus, enables a risk based
inspection program to be implemented. The results also satisfy the first objective
which is to determine a correlation of significant parameters affecting the erosion-
corrosion rate of the CS blinded flowline tees.
In addition, the data table in Appendix Table B3 can be calculated to
determine amount of the fluid flow through each of the gas flowlines.
The following is MINITAB output of multiple regression equation:
mm/year = -1.39 + 0.000092(mixture) + 0.111(CO
2
) (12)
(1.423) = -1.39 + 0.000092(16,147 barrels) + 0.111(11.96 %)
As per above multiple regression equation (12), using mean figures of
mixture in barrels and CO
2
in % content, indicates that the CS blinded flowline tee
will last approximately 7 years. This calculation is based on an assumption that a
thickness of the new CS blinded flowline tee and a minimum thickness allowance
are 15.5 millimeters and 5.5 millimeters respectively.
2. Results of ETA and CBA
Below is extraction from Quantitative Risk Assessment to assess the need
for Fixed Fire Suppression Systems on Wellhead Platforms in the Gulf Of Thailand
(1999):
2.1 Assumption for ETA and CBA Calculations:
- The wellhead platforms are assumed to be equivalent in design.
- There are 16 flowlines per wellhead platform.
66
- The retirement life of the platforms is assumed to be 20 years.
- The discount rate = 7% or 0.07, the 7% is used in this study is based
on companys previous study.
- Production loss per one day/platform = $58,230 or =$3,639/flowline.
- Only fire hazards that may result in significant injury of loss of life
from release of hydrocarbons from flowline is considered.
- Company Goal is ALARP which is equal to 1E-04.
- Helicopter operations, construction, and drilling operations are not
included.
2.2 Risk Data
To provide a relative measure of the expected future incident
frequencies, this analysis assumes estimated frequency of future incidents is
proportional to past incident rates during the year 1991-1998 as per the companys
incident reports.
Although this approach uses limited data in the future risk estimates, it
has benefit of being based on actual incidents. Also, the data is based on the designs
and operating practices of the company rather than generic data. The flowline
failure incidents are extracted from all incident reports.
67
2.3 Preliminary Hazards Analysis [PHA]
Referring to Quantitative Risk Assessment to assess the need for Fixed
Fire Suppression Systems on Wellhead Platforms in the Gulf Of Thailand (1999),
PHA related to flowline failure incidents is caused by erosion-corrosion and
consequences are hydrocarbon releases, fire, and explosion.
2.4 ETA Result
The ETA is based on binary logic, in which an event either has or has
not happened or a blinded flowline tee has or has not failed. It is valuable in
analyzing the consequences arising from a failure or undesired event.
The below ETA construction begins with an initiating event, which a
failure of a blinded flowline tee, releases hydrocarbon fluid. The consequences of
this failure are followed through a series of possible paths.
Each path is assigned a probability of occurrence as per highlighted
items in Appendix C Table C2, and the probability outcomes can be calculated.
In this case, hydrocarbon release is protected by the shutdown or shut-
in system. The shutdown system will either activate or it will not. If the shutdown
system succeeds, the alarm will either work correctly or it will not. There is only
one branch in the tree that indicates all the subsystems have succeeded.
68
Figure 27 ETA Diagram of the CS Blinded Flowline Tee Failure
The above ETA diagram - figure 27 illustrates the failure calculation for the
CS blinded flowline tee, total probabilities are as follows:
- No production loss 1 = 0.590
- Production Loss 2 = 0.027
- Production Loss 3 = 0.347
- Hydrocarbon Release 4 = 0.036
- Total Probabilities = 1.000
Based upon Quantitative Risk Assessment to assess the need for Fixed
Fire Suppression Systems on Wellhead Platforms in the Gulf Of Thailand (1999),
the company incidents statistics of the CS flowline failure incidents were 6
incidents during the year 1991 1998 which equal to 0.01 failure of frequency per
year per platform. There were 72 platforms in that period.
The loss of a remote or wellhead platform was determined to have
business interruption or contractual ramification, but the primary risk is determined
to be the risk to personnel. The maximum fire damage and business interruption
expense estimated was to be $16 million dollars. This estimate assumed that the
69
company opts to replace platform in nearby location, drill the wells, remove the
damaged structure, and perform wells abandonment. This is a very conservative
estimation of damages. Based on Gulf of Mexico risk data, the study assumed that a
major uncontrolled fire had a 5% probability of causing major topsides damage and
a 1% probability of causing complete platform destruction.
Table 5 Risk Level Calculation in $ / Year
From table 5, Risk Level Calculation in $ dollars per year is based on
below equation:
Risk = Initiating Event Frequency x Probabilities x Consequences (13)
To ensure the bias in favor of safety, the 5% probability of major
platform topsides damage is used to multiply the total platform destruction,
probability of Risk Calculation of Individual, and Group Risks. Another assumption
is that a total of 13 personnel are onboard the platform during the major fire
incident. According to current activities on the remote platform, the 13 personnel
are the maximum Personnel On Board [POB]. So the probability of the evacuation
is equal to the following equation:
Failure Rate of the CS flowline (0.01) x Failure Rate of the valve
(0.036) x (5 % Probability of the Major Fire Incident (0.05)
= 0.000018 Evacuation per year or 1.8E-05
70
In addition to the table 5, another ETA is created to calculate the rate of
individual and group risks. Details of the ETA and Risk Calculation are shown in
Appendix Figure C3 and Appendix Table C1.
The outcome fatalities and difference weather cases are based on North
Sea statistics due to no fatalities statistics available in Thailand for offshore Oil &
Gas industry. To determine the risk level, the calculated results are used to compare
against the risk regions on the FN curve in Appendix Figure C2. The calculated
Fatality Accident Rate [FAR] is in the negligible region.
2.5 CBA Result
The results of the ETA risk estimation indicate that there is no need to
add safety measures to reduce the risk exposures because the calculated FN curve is
already in the negligible region. This leads to the conclusion that the risks are
broadly acceptable and no further risk reduction measures need to be considered,
provided appropriate diligence is applied to maintain the risks in this region.
From Appendix Table C3 in Appendix C, the cost comparison indicates
that the CS is more cost-effective.
2.6 Sensitivity Analysis Result
In order to response to a fluctuation of cost ratios, a sensitivity analysis
is carried out to reflect the change of the cost ratios.
71
Table 6 Cost-Benefit Comparison Summary
From table 6, the project life periods are 2, 4, 6, 8, 10, 12, 14, 16, 18,
and 20 years respectively. The cost-ratios between the DSS and the CS are 8, 5, 4,
3, and 2 times respectively.
Table 6 shows that the CS is the material of choice when cost ratio is
equal to 5 times or higher. Once the cost ratio is equal to 3 or 4 times, the CS can be
used only for shorter project life periods of 11 years, and 19 years respectively. If
the cost ratio is equal to 2 times, then the DSS material is recommended because of
the complexity of the CS flowline system such as sand probe shutdown loop and
inspection frequency etc.
Discussion
1. Discussion of mixture and CO
2
Regressors
Based upon the MINITAB output of multiple regression analysis, the fluid
flow and the CO
2
variables are significant in predicting the erosion-corrosion rate
while the fluid density is not.
Campell (2004) explains the effects of velocity and carbon dioxide to the
erosion-corrosion rate as follows:
- If the velocity is too high, erosion, erosion-corrosion may results. These
72
problems can especially severe if the water contains large amounts of suspended
solids, and are usually worst at changes in direction, such as elbows and blinded or
plugged tees.
- Partially protective corrosion product films are constantly removed or
eroded away, leaving a bare metal surface which eagerly corrodes.
- High velocities are usually not an issue in systems where the fluid must be
pumped considerable distances, as the pump horsepower required to overcome
friction losses at higher velocities tends to be excessive. However, high velocities
are common in some compact offshore water injection systems where distances are
short and the volumes injected are large.
1.1 Dissolved Carbon Dioxide
When CO
2
dissolves in water it forms carbonic acid, from which the H
+
ions decrease the pH of the water and increase its corrosivity. It usually results in
pitting. Corrosion caused by CO
2
dissolved in water is called sweet corrosion.
CO
2
+ H
2
O H
2
CO
3
H
2
CO
3
2H
+

+ CO
=
3
Fe + H
2
CO
3
FeCO
3
+ H
2
Fe + 2H
+
+ CO
=
3
Fe
=

+ CO
=
3
+ H
2
The formation of a layer of corrosion product on the surface of a
corroding metal usually results in decreased corrosion rates. The corrosion product
acts as a rather poor coating and partially protects the surface.
Removal of the corrosion product by erosion due to high velocities,
turbulence, or the abrasive action of suspended solids can result in increased
corrosion rates by continually exposing fresh metal to the electrolyte. This form of
corrosion is called erosion-corrosion and is common cause of failure. Carbon steels
are particularly susceptible to this form of attach when the corrosion products which
form soft and easily removed, such as iron carbonate.
73
Erosion-corrosion normally occurs in localized areas such as bends,
elbows and blinded or plugged tees.
Several equations have been proposed for the calculation of minimum
velocity for the onset of erosion-corrosion but none have proven to date - Craig
(1985).
Based upon the multiple regression equation, it can be concluded that
erosion-corrosion rate is generally much more severe when the fluid flow rate and
CO
2
content are high.
1.2 Degradation of Materials
The CS and low alloys steel are highly sensitive to flow-induced
erosion-corrosion. At high velocities this may result in accelerated corrosion
degradation originating from increased mass transfer, degradation of protective
corrosion product layers and reduced corrosion inhibitor efficiency. For Corrosion
Resistant Alloys, high fluid velocities will in general not induce or accelerate
corrosion degradation.
2. Rules for using Multiple Regression Analysis
In this research, the multiple regression analysis technique has proved to be
a very effective tool in predicting equipment failure in a very cost effective manner.
On the other hand, there are some additional rules that have to be obeyed in order to
apply a multiple regression analysis properly:
- The dependent variable should be measured on an interval or continuous
scale. In practice an ordinal scale is usually good enough. If it is only measured on a
nominal scale, the other techniques such as Discriminant analysis and Logistic
regression must be used.
74
- The independent variables should be measured on interval scales.
However, most ordinal scale measurement will be acceptable in practice; 2-vauled
categorical variables can be used directly; and there is way of dealing with k-valued
categorical variables, by dummy variables. The k usually stands for any integer
greater than 2.
- The distributions of all the variables should be normal. If they are not
roughly normal, this can often be corrected by using an appropriate transformation
e.g. taking logarithms of all the measurement.
- The relationships between the dependent variable and the independent
variable should be linear. That is, it should be possible to draw a rough straight line
through an x-y plot of the observed points. If the line looks curved, but is
monotonic meaning increases or decreases all the time, things are not too bad and
could be made better by transformation. If the line looks U-shaped, there is a need
to take special steps before regression can be used.
- There must be no interactions, in the ANOVA sense, between independent
variables the effect of each on the dependent variable must be roughly independent
of the effects of all others. However, if interactions are obviously present, and not
too complex, there are special steps that can be coped with the situation.
- Although the independent variables can be correlated, there must be no
multicollinearity.
- There are also requirements on the distributions of error that needs to be
checked.
75
3. Application of Multiple Regression Equation
It is noteworthy to address that application of this multiple regression
equation to predict the erosion-corrosion rate in hydrocarbon production systems,
one should remember that this equation is based on specific data sets, specific
reservoir and gas production parameters. In other hydrocarbon production systems,
a response variable and regressors may not have a linear regression relationship.
The relationship could be quadratic, cubic etc.
4. QRA ETA Discussion
The ETA is, used as QRA in this research, very informative and useful. The
ETA technique bridges the gap between the risk and the CBA.
4.1 Strengths and Weaknesses of ETA
The ETA is a simple but effective technique, suitable for many
applications.
The strengths of the ETA are:
- It is widely used and well accepted.
- It is suitable for many hazards in QRA that arise from sequences of
successive failures.
- It is clear and logical form of presentation.
- It is simple and readily understood.
76
Its weaknesses are:
- It is not efficient where many events must occur in combination, as it
results in many redundant branches.
- All events are assumed to be independent.
- It loses its clarity when applied to system that do not fall into simple
failed or working states e.g. human error, adverse weather etc.
4.2 Individual Risk and Group Risk Discussion
The results from a QRA may be expressed as:
- Individual risks the risk experienced by the individual on the remote
platform. This usually refers to the risk of death, and may be expressed as an
individual risk per annum or a FAR per 100 million exposed hours.
- Group risks the risk experienced by the whole group of personnel
working on the remote platform. This usually refers to the risk of death, and is
usually expressed as an average number of fatalities per platform or installation-
year, known as annual fatality rate, potential loss of life, rate of death etc.
Alternatively, it may be expressed as an FN curve, showing the cumulative
Frequency [F] of events involving N or more fatalities. The FN curve is shown in
Appendix Figure C2 - Appendix C.
4.3 CBA Impact Analysis Tool
The CBA is only one of several techniques useful in impact analysis. In
the area of environmental regulation, for example, it is most useful where scientific
results i.e. dose-response relationships can be translated into costs and benefits with
reasonable certainty. The technique of risk benefit analysis is usually more useful
77
where uncertainty prevents such translation. Other techniques, such as cost
effectiveness analysis, may be useful where the problem is the most efficient way of
achieving a given objective.
A number of concepts are critical to an understanding of both the
applications and limitations of the CBA. In general, it should be noted that the CBA
is a formal, abstract and rigorous technique used to develop a bottom-line number
which is supposed to indicate how much a particular project, policy or regulation
will add to, or detract from, the overall well-being of society.
The bottom-line number can be either net benefits, the benefit per cost
ratio, or the internal rate of return, depending on what the analysis is supposed to
find out. If the problem is the evaluation of a single project, net benefits is often
used. If the problem is which of a number of investment options makes the largest
contribution to social welfare, the benefit per cost ratio is the preferred measure. If
the problem is to determine how a project compares with all alternative investment
opportunities in society, the internal rate of return should probably be used.
4.4 Limitation of CBA
The CBA is theoretically lodged in welfare theory and partial
equilibrium analysis, which are essentially static and rigorously formal bodies of
economic theory. While adequate for many purposes, the CBA has some important
drawbacks which reflect its origins, particularly the assumption that only a few
factors under analysis can be varied while all others such as prices in general, the
distribution of income and wealth, tastes and preferences, and technology are held
constant.

It should also be noted that, because the CBA is essentially incremental
or marginal in its approach, it has no place for cumulative effects assessment. It is
forward looking, focusing on additions to the economy, and regards past economic
78
choices as irrevocable. Therefore, the analysis is carrying no weight in making
current choices.
4.5 The Use of QRA and CBA
The QRA and the CBA are sometimes applied to demonstrate that
measures which have not been adopted are not reasonably practicable. This
negative approach results in the QRA being used to explain why some safety
measures are not adopted, while quantitative arguments are considered sufficient to
explain why other measures are adopted. The suggested solution to this is that,
where it is appropriate to use the QRA, it should be used to evaluate all major
safety measures, including those that are adopted as well as those that are not. This
will avoid creating the negative impression.
Below Figure 28 illustrates the method of design and operate the plant
with risk As Low As Reasonably Practicable principle.
Figure 28 Design and Operate Plant with Risk - ALARP
Source: HSE (1999)
79
The results or benefits are risk-based or performance based inspection
program which is cost-effective. Current practice to inspection and maintenance
planning is for the most industries based on tradition and prescriptive rules. New
technology for taking risk based decision is emerging within a broad range of
sectors, and has proven to be a very efficient tool.
The risk based maintenance/inspection approach should be associated
with company objectives towards plant availability and safety. Risk based methods
like Reliability, Availability and Maintainability [RAM], Reliability Centered
Maintenance [RCM] and Risk Based Inspection [RBI] can be used to rank systems
and equipment and develop strategies for maintenance and inspection.
The RBI uses risk as a basis to prioritize and manage the efforts of an
inspection program. An effective risk based program results in a reduced level of
risk for a given level of inspection activity. In any operating plant, a relatively large
percentage of risk is usually associated with a small percentage of equipment. The
RBI methodology permits the shift of inspection and maintenance resources to
provide a higher level of coverage on high-risk items and an appropriate effort on
lower risk equipment.
In addition to the results of prediction model, the QRA, and the CBA
are used as decisive factors for the CS or the DSS material selection.
4.6 Benefits of Risk Management Process
It is a common experience in performing risk assessments that the
process of performing a risk assessment yields greater benefits than the final risk
results. The relatively small importance of risk results arises from the uncertainties
that are inevitably attached.
80
The results tend to be more important in a novel application of risk
assessment, where risk estimates have not previously been available. As more and
more risk assessments of similar installations/plants or activities are carried out, the
differences in the risk results are often seen to arise more from differences in
methodology than from actual differences in the installations, and their significance
decreases.
The much larger importance of the risk assessment process arises from
the creative yet systematic though process that is necessary to produce risk
estimates.
4.7 QRA and CBA Risk Presentation
One of the most difficult issues in the CBA of safety is how to balance
costs with risks, when the two are in different units. Many types of risks can easily
be expressed in monetary terms for example, risks of property damage or business
interruption. But risks to life are much more difficult to value. Risks of damage to
the environment pose an even greater problem in this respect.
The standard approach to the CBA of risks to life is to convert them
into equivalent costs. The monetary valuation of risks to life is often described as
placing a value on life. This phrase is convenient but distasteful, because no amount
of money can compensate an individual for loss of their life. In fact, the CBA
places a value on averting a statistical fatality.
Presentation of this difficult concept can be improved by using the term
value of preventing a statistical fatality. This emphasizes that what is being valued
is the reduction in risk to many lives, rather than the actual lives that are at risk of
being lost.
81
CONCLUSIONS AND RECOMMENDATIONS
Conclusions
This thesis represents a study of the application of the multiple regression
analysis in a real-world scenario of the erosion-corrosion prediction in the
hydrocarbon production blinded flowline tees using a simplified approach.
The data sets from the company records are used to analyze the historical
trend. From the results of the two-variable multiple regression model, it can be
concluded that the fluid flow and percentage of CO
2
content are significant
variables affecting the erosion-corrosion rates in the CS blinded flowline tees.
The QRA and the CBA impose a discipline on the analysts to consider the
safety of an installation or activity in great detail, to think about what might go
wrong and what is available to prevent or mitigate it, and to consider the relevance
of previous accident experience.
When performed properly, this process yields a great understanding of the
installation and its safety features, often with useful insights into ways that safety
must be improved. Even if no new cost-effective safety measures are identified, the
process provides reassurance that an important and reasonably practical step has
been taken to anticipate what might go wrong and what could be done to prevent it.
A particular benefit arises from the consideration of the role of safeguard i.e.
risk reduction measures incorporated into the design in achieving acceptable safety.
Consideration of these is particularly important in the HAZID and in the semi-
quantitative approach to risk, as well as some approaches to the QRA, although it
tends to be obscured in the approaches based on historical frequencies. This
provides important input to the safety management system, for example by
suggesting the performance standards required from key safeguards, highlighting
training needs, providing input to emergency planning, emergency drills etc.
82
The CBA forms an important link between the QRA and general safety
management. This explains why most modern safety management systems include
a requirement for a risk assessment & risk management, and why it is such an
important component of the offshore safety regime.
The QRA and The CBA can also be used as tools for communication and
education to closing the gap between management and frontline workforce. Without
proper risk presentation and communication, the use of the CS instead of the DSS
may be perceived by the frontline workforce as an attempt to lower safety standard
because the CS cost is much less than the cost of the DSS material. To ensure
correct perception of this process, the QRA and the CBA applications should be
presented and communicated to all affected personnel.
In conclusion, the QRA and the CBA allow an estimate of the benefits of
safety measures, in terms of the risk-factored cost of the accidents they would avert.
The QRA and the CBA bridge the gap between engineering and finance. This is
because expected value, defined by a multiplication of probability and consequence,
is a term that is already used in the finance community.
To highlight the second objective of this research, a graph of the sensitivity
analysis is plotted by utilizing the data from table 6. Figure 29 illustrates the
sensitivity graph highlighting the cost ratios of the CBA. The selected cost ratios
are 8, 5, 4, 3, and 2 respectively.
83
Figure 29 Sensitivity Analysis - DSS versus CS Cost-benefit Comparison
At this time the DSS material cost is approximately 8 times higher than the
CS material. For example, the project life is 10 years will equate to the saving of
$17,126 dollars per flowline or $274,016 dollars per platform. If the project life is
20 years then the saving is $12,195 dollars per flowline or $195,120 dollars per
platform.
According to the companys forecast of Thailands natural gas production
trend reflecting supply and demand, six additional wellhead platforms will require
to be built annually to maintain the sustainable gas production rate and meet the
forecast.
The results of the quantified risk level and the CBA, as applied to the
problem of erosion-corrosion management, show that the risk of using carbon steel
is negligible. This practice is broadly accepted as a practical way to manage cost.
This leads to the conclusion that the cost-benefit of using carbon steel outweighs
the cost-benefit of using duplex stainless steel in many erosion-corrosion situations.
84
Recommendations
The overall concept of the process, outlined in this thesis can be extended to
cover other aspects of the topic, providing that the following recommendations are
being considered:
1. Due to the complexity of erosion-corrosion phenomena, additional
parameters such as Temperature, Pressure, and pH Value etc. should be taken into
accounted for the multiple regression analysis. Then Stepwise regression
procedures in MINITAB

program should be applied to select the best fitted model.


2. Det Norske Veritas [DNV] Recommended Practice O501 - Erosive
Wear in Piping Systems - revision 1999 should be applied to the multiple regression
analysis in comparison to the API RP 14E. The DNV RP O501 contains additional
factors which are not accounted for in the API RP 14E such as sand, H
2
S, and O
2
etc.
3. The multiple regression analysis technique should be applied to predict
erosion-corrosion rates of offshore oil flowline, pipeline, and water injection
systems. In the case of oil flow regime, the fluid density factor should be included
in the multiple regression analysis.
4. The computational prediction models of erosion-corrosion rates should
be cross referenced with the multiple prediction equation to provide comparison of
results from the different models. The results can be used to calibrate the erosion-
corrosion prediction equation for blinded flowline tee.
5. The multiple regression analysis technique can be applied to other types
of processing equipment e.g. pressure vessels. It is also applicable to the prediction
of equipment failures in chemical and petrochemical processing facilities.
85
6. The concept of QRA and the CBA can be applied to all high value
projects in order to provide company management with a tool to make informed
decision based on quantitative findings.
Emotions play a role in every decision that humans make. It plays a greater
role as the complexity and the risk of making decision goes up. It would be a
challenging task for all stakeholders in providing decision makers with a standard
that enables factual information to be used as a basis of the decision making
process.
86
LITERATURE CITED
American Petroleum Institute API RP 14E. 1991. Recommended Practice for
Design and Installation of Offshore Production Platform Piping
Systems. 5
th
edition. (reaffirmed 2000).
Barton, N.A. 2003. Erosion in elbows in hydrocarbon production systems:
Review document. Research Report 115.
Campbell, J.M. 2004. Corrosion Management: Production/Processing
Operation. Version:W22_UnocalThailand_01.01.04, John M. Campbell
and Company, 1215 Crossroads Blvd. Norman, Oklahoma U.S.A.
CCPS. 2000. Guidelines for Chemical Process Quantitative Risk Analysis.
2
nd
Edition. Center for Chemical Process Safety of the American Institute
of Chemical Engineers, AICHE, New York
CMPT. 1999. A Guide to Quantitative Risk Assessment for Offshore
Installations. Centre for Maritime and Petroleum Technology, London.
Craig, B. 1985. Critical Velocity Examined for Effects of Erosion-Corrosion.
Oil & Gas Journal. 99.
Crowl, D.A. and J.F. Louvar. 2002. Chemical Process Safety Fundamentals
with Applications. Second Edition, Prentice Hall PTR, Prentice-Hall Inc.
DEAT. 2004. Cost Benefit Analysis, Integrated Environmental Management,
Information Series 8. Department Affairs and Tourism (DEAT), Pretoria.
Det Norske Veritas. 2001. Marine risk assessment. Prepared for the Health
and Safety Executive. Offshore Technology Report 2001/063.
87
Hattakitjumreon, P. 2000. Stress and Deformation Analysis of Pipe Tees in the
UNOCAL Pipeline System. Master of Engineering Thesis, Chemical
Engineering Program, King Mongkuts University of Technology Thonburi.
HSE. 1999. Reducing Risks, Protecting People. Discussion Document.
Health & Safety Executive.
Igland, R.T. and T. Moan. 2000. Reliability of pipelines during laying, considering
ultimate strength under combined loads. Journal of Offshore Mechanics and
Arctic Engineering. 122: 40-66.
IMO. 1997. Interim Guidelines for the Application of Formal Safety
Assessment [FSA] to the IMO Rule-Making Process. Marine Safety
Committee MCS/Circ.829. International Maritime Organization, London.
Louvar, J.F. and B.D. Louvar. 1998. Health and Environmental Risk
Fundamentals with Applications. Volume 2, Prentice Hall PTR, NJ.
Mochizuki, M., M. Hayachi and T. Hattori. 2000. Residual stress distribution
depending on welding sequence in multi-pass welded joint with X-shaped
groove. Journal of Pressure Vessel Technology. 122 : 27-32.
Montgomery, D.C. 2001. Design and Analysis of Experiments. 5
th
Edition.
John Wiley & Sons, Inc.
Montgomery, D.C. and G.C. Runger. 2002. Applied Statistics and Probability
for Engineers. 3
rd
Edition. John Wiley & Sons, Inc.
Nilkitsaranont, P. 2004. Development of a Diagnostics and Prognostics
Approach to Stationary Gas Turbine Applications. Master of Science
Thesis, Thermal Power, Cranfield University, School of Engineering.
88
Pongsakornrungsilp, S. 2004. Business Statistics. 1
st
Edition. Pearson Education
Indochina Ltd., Bangkok, Thailand.
Quantitative Risk Assessment to Assess the Need for Fixed Fire Suppression
Systems on Wellhead Platforms in the Gulf Of Thailand. 1999. Prepared
by AcuTech for Unocal Thailand Limited, Bangkok, Thailand.
Salama, M.M. and E.S. Venkatesh. 1983. Evaluation of API RP14E erosional
velocity limitations for offshore gas wells: 371 376. OTC 4484, OTC
Conference, Houston.
Sewell, W.R.D., J. Davis A.D. Scott and D.W. Ross. 1965. Guide to Benefit-Cost
Analysis. Queens Printer, Ottawa.
Shadley, J.R., S.A. Shirazi, E. Dayalan, M. Ismail and E.F. Rybicki. 1996.
Erosion-corrosion of a carbon steel elbow in a carbon dioxide
environment. 52 (9): 714 723.
SPSS Training Department. 1999. Time Series Analysis and Forecasting with
SPSS Trends 9.0, SPSS Inc.
Siroratanasret, P. 2003. Quantitative Study of VOC Emission from Plastic
Printing Industry. Master of Engineering Thesis, Safety Engineering
Program. Kasetsart University.
Sritham, S. 1999. Simulation of Pressure Distribution of ERAWAN Pipelines
Circuit. Master of Engineering Thesis, Chemical Engineering Program,
King Mongkuts University of Technology Thonburi.
Thampanitchawong, P. 1999. Study of Corrosion Phenomena in Pipeline
System with Multiphase Flow Simulation Model. Master of Engineering
Thesis, Chemical Engineering Department. Kasetsart University.
89
UKOOA. 1999. A Framework for Risk Related Decision Support. UK
Offshore Operators Association, London.
Venkatesh, E.S. 1986. Erosion damage in oil and gas wells. Proc. Rocky
Mountain Meeting of SPE. Billings, MT.
90
APPENDICES
91
APPENDIX A
Overview of Oil and Gas Development Process
92
Overview of Oil and Gas Development Process is illustrated in below figures:
The D-Cycle Gas Engine in Appendix Figure A1, which consists of
Drilling, Discovery of 3-D Seismic Survey, Delineation and Development or Gas
Production, illustrates the process of oil and gas exploration and production in the
Gulf of Thailand.
Appendix Figure A1 D-Cycle Gas Engine - Gas Development Overview
Source: Unocal Thailand, Ltd. (n.d.)
The gas development cycle or process begins with an exploration activity by
means of seismic survey to collect geological data then the data will be interpreted
to evaluate for potential of oil and gas reservoir.
Appendix Figure A2 illustrates a survey boat that is employed to conduct a
seismic survey activity in the Gulf of Thailand.
93
Appendix Figure A2 A Survey Boat at Seismic Survey Location
Source: Unocal Thailand, Ltd. (n.d.)
If the result reveals a high potential for oil and gas reservoir, then a jackup
rig will be mobilized to drill an exploration well in order to evaluate the reservoir.
Appendix Figure A3 illustrates a typical jackup rig that is utilized to drill
exploration, delineation, development, and infill wells in the Gulf of Thailand.
94
Appendix Figure A3 Jackup Drilling Rig
Source: Unocal Thailand, Ltd. (n.d.)
Upon reaching total depth of the well as per well path planning, a well
logging and a drill/tubing string testing are performed to determine pay zones, flow
rate, and amount of condensate and gas in the reservoir. The testing results will be
analyzed and financial evaluated for economic justification.
The next step is to drill delineation wells to gather more geological data to
evaluate the quantity of hydrocarbon in the reservoir and quality of the reservoir in
terms of permeability and porosity. Then, an offshore platform will be constructed
onshore and it will be transported to offshore location by a barge. When the
platform arrives at its location, a construction barge will be utilized to install the
platform at the offshore location.
95
Most of wellhead and processing platforms are constructed at Laem
Chabang in Chon Buri Province, Thailand as shown in Appendix Figure A4.
Appendix Figure A4 Construction of Unocal Offshore Remote Platform
Source: CUEL Company
Once the installation of the wellhead platform at offshore location is
completed, a tender-assisted rig is utilized to drill development wells as per well
path planning shown in Appendix Figure A6. This tender-assisted rig can drill the
development and infill wells only. So in term of multipurpose usage, the jackup rig
is the rig of choice because it can drill all types of wells. But the tender-assisted rig
has its advantage over the jackup rig due to its ability to drill faster, more efficient
with less $ day rate.
96
Appendix Figure A5 A Tender-assisted Rig while drilling at Remote Platform
Location
Source: Unocal Thailand, Ltd. (n.d.)
Appendix Figure A6 shows a typical well path for well planning of
development wells and infill wells in the Gulf of Thailand at the offshore platform
location.
97
Appendix Figure A6 Typical Well Path West-East Cross Section
Source: Unocal Thailand, Ltd. (n.d.)
98
The following figures show various types of the offshore platforms.
Appendix Figure A7 Offshore Platforms in the Gulf Of Thailand
Source: Unocal Thailand, Ltd. (n.d.)
99
Appendix Figure A8 Offshore Remote or Wellhead Platform
Source: Unocal Thailand, Ltd. (n.d.)
Production facilities and process equipment are fabricated and installed onto
the platform at Laem Chabang CUEL yard in Chon Buri province. Upon
completion of drilling operation, construction crew will hook up the flowline onto
the wellhead components. Then well services crew will perforate the production
string called tubing to allow a well to flow from reservoir to process equipment.
The commingle flow is undersea piped to the Central Processing Platform for
further processing.
Appendix Figure A9 below illustrates the process equipment and simplified
process diagram of a remote platform.
100
Appendix Figure A9 Simplified Diagram of Process Overview of Remote
Platform
Source: Unocal Thailand, Ltd. (n.d.)
101
Appendix Figure A10 Wellheads Orientation after completion of Drilling
Operation
Source: Unocal Thailand, Ltd. (n.d.)
Appendix Figure A10 shows wellheads installation and orientation after the
rig left the location.
On the Central Processing Platform, the gas, condensate, and produced
water are separated by means of density & gravity. The produced water is treated
by the following methods:
1. Inject the produced water back to formations deep-well injection.
2. Treat the produced water until it is safe to discharge overboard.
The condensate flow is stabilized at the stabilizer and is sent to Floating
Storage Operation Unit. The condensate is shipped onshore by offloading tankers.
102
The gas flow is processed at Glycol Dehydration Unit to converse a wet gas
flow into a dry gas flow. The dry gas flow is compressed at the Compression
Platform to boost up the pressure. Then the compressed dry gas is undersea piped to
the gas separation plant onshore at Mathaput in Rayong Province.
Appendix Figure A11 Wellhead components, choke and Blinded Tee
Source: Unocal Thailand, Ltd. (n.d.)
From above simplified diagram, Appendix Figure A11 illustrates wellhead
components, a choke and a blinded tee used for reducing erosion and flow cut
problems.
The below Appendix Figure A12 illustrates a blast Joint, a production
flowline with a sand probe shutdown diagram.
103
Appendix Figure A12 Flowline & Sand Probe Shutdown Diagram
Source: Unocal Thailand, Ltd. (n.d.)
Appendix Figures A12 and A13 illustrate production flowline with a sand
probe shutdown system. The sand probe is embedded in the most eroded-corroded
area in the blinded flowline tee and it will be activated by erosion-corrosion process
that removes the amount of steel until it reaches the depth of the sand probe. The
sand probe will activate a valve to route the gas pressure to shutdown the wing
valve. In the mean time, it will activate a solenoid to send alarm signal to control
room at central processing platform. The signal also sets off an alarm flag on that
remote platform.
104
Appendix Figure A13 Production Flowline & Sand Probe Shutdown System
Source: Unocal Thailand, Ltd. (n.d.)
105
APPENDIX B
Data Sets, Data Tables, and Specifications of Ultrasonic Thickness Gauging
106
Appendix Table B1 Erosion-corrosion Data Sets of Blinded Flowline Tees
Erosion-corrosion Data Sets of Gas Production Blinded Flowline Tees
Flowline # Erosion-Corrosion
Rate in mm/year
Fluid Flow in
Barrels - mixture
Fluid density in
lb/ft
3
CO
2
- %
content
1 0.61 9779 29.23 9.5
2 0.94 10472 29.05 11.5
3 0.83 19125 29.58 7.9
4 1.14 17684 29.06 11
5 1.46 12602 29.07 11.7
6 0.71 10945 29.48 9.6
7 3.53 32577 29.01 14.2
8 1.81 13937 29.03 11.1
9 0.88 9890 29.02 11.6
10 1.74 22065 30.6 10.8
11 0.7 14198 29.33 9.6
12 0.28 10663 29.35 6.4
13 3.88 30611 29.1 16.1
14 0.77 16354 29.1 8.9
15 1.95 14318 29 14.1
16 1.34 13672 29.07 12.2
17 0.77 13308 29.1 10.7
18 2.21 18120 29.17 11.8
19 2.68 25672 29.82 13.9
20 4.66 48577 29.1 14.8
21 1.8 13478 29 11.6
22 0.61 13359 29.1 11.6
23 0.58 11408 29.65 12
24 1.9 13734 29.26 14.1
25 2.71 25396 29.02 11.4
26 1.25 18424 29.01 11.1
27 1.52 19267 29.07 13.9
28 0.38 14387 31.89 6.8
Source: Unocal Thailand, Ltd. (n.d.)
107
Appendix Table B1 (Continued)
Erosion-corrosion Data Sets of Gas Production Blinded Flowline Tees
Flowline # Erosion-Corrosion
Rate in mm/year
Fluid Flow in
Barrels - mixture
Fluid density
in lb/ft
3
CO
2
- % content
29 1.65 16981 29.16 10.2
30 4.63 40081 29.1 23.4
31 4.86 50588 29.05 17
32 1.62 22446 29.2 13.9
33 2.07 14243 29.09 16.7
34 0.53 8574 30.16 10.8
35 1.59 13383 29.03 14.6
36 0.39 9914 30.17 11.9
37 2.57 21306 29.31 20.1
38 1.68 19673 29.13 10.8
39 0.56 9986 29.32 10.8
40 2.05 19208 29.14 10
41 0.83 6674 29.49 10.1
42 0.41 16876 31.07 10.4
43 0.55 8254 29 9.5
44 0.35 6970 29.14 11.6
45 0.71 11932 29.12 11.9
46 0.6 13006 29.18 11.2
47 0.33 6068 29.09 10.6
48 0.24 9658 29 8.8
49 5.08 45331 29.06 18
50 0.17 3329 30.05 8.7
51 0.45 5615 29 9.9
52 0.23 4687 34.41 7.8
53 0.27 4522 29.14 8.1
54 0.43 9435 29.27 10
55 0.54 6745 29.49 13.5
56 0.76 8084 29.28 14.4
57 0.94 10295 29.39 14.5
58 1.7 18640 29.31 14.8
Source: Unocal Thailand, Ltd. (n.d.)
108
Appendix Table B2 is the output of the predicted values for new observation
for each data set. Confidence and Prediction Intervals are included with 95.0 %
level.
Appendix Table B2 Table of CI and PI of MINITAB Output
Predicted Values for New Observations:
New Obs Fi t SE Fi t 95. 0%CI 95. 0%PI
1 0. 5640 0. 0627 0. 4383, 0. 6897 - 0. 1890, 1. 3169
2 0. 8498 0. 0569 0. 7358, 0. 9638 0. 0987, 1. 6009
3 1. 2422 0. 1052 1. 0315, 1. 4530 0. 4705, 2. 0139
4 1. 4549 0. 0552 1. 3443, 1. 5654 0. 7043, 2. 2054
5 1. 0671 0. 0521 0. 9628, 1. 1715 0. 3175, 1. 8168
6 0. 6819 0. 0613 0. 5591, 0. 8047 - 0. 0706, 1. 4343
7 3. 1749 0. 0935 2. 9874, 3. 3623 2. 4092, 3. 9405
8 1. 1227 0. 0506 1. 0214, 1. 2241 0. 3735, 1. 8720
9 0. 8076 0. 0592 0. 6889, 0. 9263 0. 0558, 1. 5594
10 1. 8339 0. 0721 1. 6893, 1. 9785 1. 0776, 2. 5903
11 0. 9799 0. 0636 0. 8524, 1. 1073 0. 2267, 1. 7331
12 0. 3003 0. 1061 0. 0877, 0. 5130 - 0. 4719, 1. 0725
13 3. 2060 0. 0890 3. 0275, 3. 3844 2. 4425, 3. 9695
14 1. 0996 0. 0787 0. 9418, 1. 2573 0. 3406, 1. 8585
15 1. 4911 0. 0699 1. 3511, 1. 6312 0. 7357, 2. 2466
16 1. 2207 0. 0519 1. 1166, 1. 3249 0. 4711, 1. 9704
17 1. 0206 0. 0526 0. 9153, 1. 1260 0. 2708, 1. 7705
18 1. 5837 0. 0507 1. 4822, 1. 6853 0. 8345, 2. 3330
19 2. 5090 0. 0666 2. 3756, 2. 6424 1. 7547, 3. 2632
20 4. 7072 0. 1740 4. 3585, 5. 0560 3. 8870, 5. 5274
21 1. 1363 0. 0503 1. 0354, 1. 2372 0. 3871, 1. 8855
22 1. 1254 0. 0505 1. 0241, 1. 2266 0. 3761, 1. 8746
23 0. 9911 0. 0567 0. 8776, 1. 1047 0. 2401, 1. 7421
24 1. 4377 0. 0718 1. 2937, 1. 5816 0. 6815, 2. 1939
25 2. 2058 0. 0799 2. 0456, 2. 3659 1. 4463, 2. 9652
26 1. 5338 0. 0561 1. 4214, 1. 6461 0. 7829, 2. 2846
27 1. 9223 0. 0576 1. 8068, 2. 0377 1. 1710, 2. 6735
28 0. 6859 0. 1084 0. 4687, 0. 9031 - 0. 0876, 1. 4594
109
Appendix Table B2 (Continued)
Predicted Values for New Observations:
New Obs Fi t SE Fi t 95. 0%CI 95. 0%PI
29 1. 3015 0. 0620 1. 1772, 1. 4258 0. 5488, 2. 0542
30 4. 8850 0. 1863 4. 5117, 5. 2583 4. 0541, 5. 7160
31 5. 1360 0. 1732 4. 7890, 5. 4831 4. 3165, 5. 9555
32 2. 2135 0. 0591 2. 0950, 2. 3320 1. 4617, 2. 9652
33 1. 7733 0. 1130 1. 5469, 1. 9998 0. 9972, 2. 5495
34 0. 5981 0. 0607 0. 4764, 0. 7198 - 0. 1542, 1. 3504
35 1. 4611 0. 0806 1. 2996, 1. 6226 0. 7014, 2. 2208
36 0. 8432 0. 0610 0. 7208, 0. 9655 0. 0908, 1. 5955
37 2. 7983 0. 1532 2. 4913, 3. 1053 1. 9950, 3. 6016
38 1. 6148 0. 0630 1. 4886, 1. 7411 0. 8618, 2. 3678
39 0. 7275 0. 0568 0. 6136, 0. 8413 - 0. 0236, 1. 4785
40 1. 4833 0. 0717 1. 3397, 1. 6269 0. 7272, 2. 2394
41 0. 3462 0. 0663 0. 2133, 0. 4792 - 0. 4079, 1. 1004
42 1. 3141 0. 0594 1. 1951, 1. 4332 0. 5623, 2. 0660
43 0. 4243 0. 0646 0. 2949, 0. 5537 - 0. 3293, 1. 1778
44 0. 5401 0. 0707 0. 3985, 0. 6818 - 0. 2156, 1. 2959
45 1. 0280 0. 0546 0. 9187, 1. 1373 0. 2776, 1. 7784
46 1. 0486 0. 0510 0. 9463, 1. 1508 0. 2992, 1. 7979
47 0. 3463 0. 0691 0. 2079, 0. 4848 - 0. 4088, 1. 1015
48 0. 4751 0. 0696 0. 3357, 0. 6145 - 0. 2803, 1. 2304
49 4. 7656 0. 1476 4. 4698, 5. 0615 3. 9665, 5. 5648
50 - 0. 1158 0. 0799 - 0. 2760, 0. 0444 - 0. 8753, 0. 6437
51 0. 2270 0. 0699 0. 0870, 0. 3670 - 0. 5285, 0. 9825
52 - 0. 0915 0. 0835 - 0. 2588, 0. 0759 - 0. 8525, 0. 6695
53 - 0. 0732 0. 0811 - 0. 2357, 0. 0892 - 0. 8332, 0. 6867
54 0. 5880 0. 0599 0. 4680, 0. 7081 - 0. 1640, 1. 3401
55 0. 7307 0. 0931 0. 5441, 0. 9174 - 0. 0347, 1. 4962
56 0. 9534 0. 1001 0. 7529, 1. 1540 0. 1845, 1. 7224
57 1. 1671 0. 0915 0. 9836, 1. 3505 0. 4024, 1. 9318
58 1. 9649 0. 0689 1. 8268, 2. 1029 1. 2098, 2. 7200
110
The following is Appendix Table B3 which contains the Values of
Predictors for new observation for each data set. The regressors of this multiple
regression equation are shown namely fluid flow or mixture in barrels and CO2 in
percentage content respectively.
Appendix Table B3 Table of Values of Predictors for New Observations
New Obs mi xt ur e CO2
1 9779 9. 5
2 10472 11. 5
3 19125 7. 9
4 17684 11. 0
5 12602 11. 7
6 10945 9. 6
7 32577 14. 2
8 13937 11. 1
9 9890 11. 6
10 22065 10. 8
11 14198 9. 6
12 10663 6. 4
13 30611 16. 1
14 16354 8. 9
15 14318 14. 1
16 13672 12. 2
17 13308 10. 7
18 18120 11. 8
19 25672 13. 9
20 48577 14. 8
21 13478 11. 6
22 13359 11. 6
23 11408 12. 0
24 13734 14. 1
25 25396 11. 4
26 18424 11. 1
27 19267 13. 9
28 14387 6. 8
29 16981 10. 2
30 40081 23. 4
31 50588 17. 0
32 22446 13. 9
33 14243 16. 7
34 8574 10. 8
35 13383 14. 6
36 9914 11. 9
37 21306 20. 1
111
Appendix Table B3 (Continued)
New Obs mi xt ur e CO2
38 19673 10. 8
39 9986 10. 8
40 19208 10. 0
41 6674 10. 1
42 16876 10. 4
43 8254 9. 5
44 6970 11. 6
45 11932 11. 9
46 13006 11. 2
47 6068 10. 6
48 9658 8. 8
49 45331 18. 0
50 3329 8. 7
51 5615 9. 9
52 4687 7. 8
53 4522 8. 1
54 9435 10. 0
55 6745 13. 5
56 8084 14. 4
57 10295 14. 5
58 18640 14. 8
112
Appendix Table B4 Table of Percentage Points of the F-Distribution
Source: Montgomery (2001)
Appendix Table B4 contains Table of Percentage Points of the F-
Distribution which is used to determine F value in comparison to a calculated F-
statistic.
Appendix Table B5 contains the specifications of the Ultrasonic Thickness
Gauging - The PANAMETRICS

36DL PLUS Model.


113
Appendix Table B5 Specifications of Ultrasonic Thickness Gauging
114
Appendix Table B5 (Continued)
115
APPENDIX C
Risk Framework, ETA, and Risk Calculation and Cost Comparison
116
Risk Framework, Risk Calculation, and Cost Comparison are
illustrated in the following Appendix Figures and Appendix Tables:
The Tolerability of Risk framework has been used widely for offshore
installations. To apply it, the duty holder must first ensure that the risks are not
unacceptable, and must then show that the risks are either ALARP or broadly
acceptable.
Appendix Figure C1 Tolerability of Risk Framework
Source: HSE (1999)
117
HSE (1999) has specific risk criteria to indicate the boundaries between
zones. Although these are intended to be guideline, not rigid criteria to be complied
with in all cases, in practice most offshore operators have adopted criteria based
closely on them.
Appendix Figure C2 illustrates how to use the calculated Frequency and
Number of Fatalities to determine the risk criteria which has three regions as listed
on the Chart of FN Curve.
Appendix Figure C2 Frequency and Number of Fatalities FN Curve
Source: CMPT (1999)
118
Appendix Figure C3 contains Risk Calculation of ETA for Evacuation
Event.
Appendix Figure C3 ETA and Risk Calculation for Evacuation Event
Source: CMPT (1999)
Appendix Table C1 ETA and Risk Calculation for FN Curve
119
Appendix Table C2 contains Failure Rate Data for various selected process
components used for probability calculation in Event Tree Analysis.
Appendix Table C2 Failure Rate Data in Process Components
Instrument Failures/Year
Controller 0.29
Control valve 0.60
Flow measurement fluids 1.14
Flow measurement solids 3.75
Flow switch 1.12
Gas-liquid chromatograph 30.6
Hand valve 0.13
Indicator lamp 0.044
Level measurement liquids 1.70
Level measurement solids 6.86
Oxygen analyzer 5.65
pH meter 5.88
Pressure measurement 1.41
Pressure relief valve 0.022
Pressure switch 0.14
Solenoid valve 0.42
Stepper motor 0.044
Strip chart recorder 0.22
Thermocouple measurement 0.52
Thermometer measurement 0.027
Value positioner 0.44
Pump failure 0.026
Valves failure 0.026
Motor-operated 0.36
Solenoid 0.36
120
Appendix Table C2 (Continued)
Mechanical hardware Failures/Year
Air-operated 0.036
Check, failure to open 0.036
Relief, failure to open 0.0036
Electrical hardware Failures/Year
Motor failure 0.0086
Transformers, open/shorts 0.0086
Relays, failure to energize 0.011
Circuit breakers, failure to transfer 0.36
Limit switch failure 0.036
Torque switches, failure to operate 0.036
Pressure switch, failure to operate 0.036
Manual switch, failure to operate 0.011
Battery power supply failure 0.0026
Solid state device failure 0.0086
Diesels, failure to start 10.8
Diesels, Failure to run 8.6
Human error Failures/demand
Instrument, failure to operate 0.0086
Selection of key-operate switch 10
-4
Selection of hand switch 10
-3
Reading of labels 3 10
-3
Correct action under stress 0.2
Correct action under stress in 60 sec 1.0
Source: Louvar and Louvar (1998)
121
Appendix Table C3 contains the detailed calculation of the Cost
Comparison between the CS and The DSS materials. At this time the cost of the
DSS is equal to 8 times of the CS cost. Project life of 10 years is used as an
example for calculating the cost comparison.
Appendix Table C3 Cost Comparison between the CS and the DSS
122
Appendix Table C3 (Continued)
The CS estimated time replacement used is 5 years which is very
conservative figure due to a mean replacement period of 7 years. But the 5 year
figure is used because it provides marginal on the safety side. By the same token,
the estimated time replacement for the DSS is none. The inspection periods of the
two materials, are assumed by the following estimates:
- One inspection per 2 years for the CS.
- One inspection per 4 years for the DSS.

You might also like