Sie sind auf Seite 1von 11

DC201200 DOI: 10.

2118/201200-PA Date: 18-May-20 Stage: Page: 1 Total Pages: 11

Empirical Random Kick Model and


Casing Reliability
A. T. Beck, University of São Paulo; D. L. Cecı́lio, Federal University of Rio Grande do Sul;
W. J. S. Gomes, Federal University of Santa Catarina; R. Tessari, Technical Federal University of Paraná;
and C. M. C. Jacinto, Petrobras

Summary
The development of exploratory wells is based on the interpretation of seismic readings by geotechnical field experts. Correlation
wells, when they exist, are located very far, leading to large uncertainties and to different interpretations by different experts. Explora-
tory wells are expensive and risky endeavors, in particular deep wells with high-temperature and high-pressure conditions. Reaching
reservoirs at record depths often requires challenging usual safety margins, which should be based on reliability analyses. In this paper,
we present an empirical random kick model to combine the opinions of different specialists and the reliability analysis of casing subject
to kick loads. The empirical kick model has a random kick volume and intensity, gas gravity, and influx depth. Three tubular strength
models are considered: Barlow formula, von Mises stress, and Klever-Stewart model (Klever and Stewart 1998; Stewart and Klever
1998), with random yield stress and tube thickness. A typical 19,000 ft well is considered as a case study example, and the First-Order
Reliability Method (FORM) is used as a probabilistic solver. Results show the differences between safety factors and reliability indices
obtained for the three strength models. Sensitivity factors show that yield stress and model error are the most relevant random variable
at shallow depths. Kick volume and influx depth are found to be the most relevant random variables of the kick model, especially at
greater depths. The proposed empirical kick model is shown to have a significant impact on casing reliability, in comparison with a
kick in which uncertain kick variables are assumed with uniform distribution.

Introduction
The development of exploratory wells carries great risks and significant uncertainties. The geotechnical stratification is evaluated by
means of seismic responses, which record the depths at which there are transitions in rock density. Geotechnical field experts are
responsible for the interpretation of seismic data, which is highly subjective. This involves estimating the actual type of rock at every
depth, including rock density, porosity, and fracture strength, and anticipating regions of possible abnormal pressures. Correlation
wells, for which the geotechnical stratification has already been established, are often many kilometers away, in case of development
(wildcat) wells, making the interpretation of seismic responses even more subjective. Hence, it is expected that two experts, given the
same raw seismic readings and correlation well data, will arrive at different estimates for geotechnical stratification, expected abnormal
pressure, and reservoir depths. This leads to significant uncertainties in well development, which has a strong effect on risks.
One of the most challenging loads for exploratory wells is the kick, which occurs when drilling through a zone of unexpected abnor-
mal pressures. If formation pressure becomes greater than the drilling fluid hydrostatic column, the hole is invaded by formation fluid
and gas in unknown proportions. Invading gas travels up the drilling column, transmitting bottom pressures to the surface. A kick can
occur at any depth between the shoe of the previous casing, and the target depth of the next phase. The influx volume usually depends
on the time it takes the personnel to detect the problem and close the wellbore. A kick can cause rupture of the installed casing because
of the excessive internal pressure. If undetected, a kick can cause the release of hydrocarbons and a blowout at the drilling rig. Hence,
kick loading is at the same time potentially severe, of dramatic consequences, and subject to significant uncertainties. The relevant kick
load variables, which need to be quantified by the experts, are the influx depth (anticipated depth of abnormal pressures), kick volume,
initial density of invading fluid, and specific gravity of invading gas.
It is currently widely accepted that optimization of well design involves the quantification of uncertain variables. In exploratory wells,
in particular, there is significant uncertainty about the drilling window (formation pore pressures and rock fracture gradients). Exploratory
wells are very costly to develop, and decisions on the number of phases have significant effects on safety and costs. Following Kaiser
(2009), the cost of casing corresponds to 10 to 20% of the cost of a completed well. Increasing the number of casing strings from three to
four increases costs by 10 to 20%, increasing the number of strings from four to five has an effect of 20 to 30%. However, a smaller
number of casing strings requires reducing safety margins with respect to fluid overbalance (increasing chances and potential volume of a
kick) and with respect to fracture gradients (increasing chances of an underground formation fracture). This type of optimization action
requires moving from a prescriptive design approach (with given safety factors) to a quantitative risk and reliability-based design.
The oil and gas industry has recognized that uncertainties in the casing design problem can be addressed in a more objective way
using structural reliability analyses. Several papers have been published on the subject (Adams 1995; Adams and Glover 1998; Adams
et al. 1998; Adams and Hodgson 1999), but they mainly address the uncertainties in the strength side of the problem. This includes
casing manufacturing uncertainties and casing rupture and collapse models, as reported in ISO 10400 (2011).
An extensive literature review revealed few papers addressing uncertainties in the loading side of the problem (Wylie and Visram
1990; Adams et al. 1993; Lewis et al. 1995; Adams and Glover 1998; Tallin et al. 2000; Mason and Chandrasekhar 2005; Guan et al.
2018). Wylie and Visram (1990) evaluated the probabilities of kick occurrence for land wells in Alberta, Canada. Adams et al. (1993)
described the uncertainties in pore pressures and kick volumes for the development and exploratory wells of BP. Adams and Glover
(1998) reported on a joint industry project study that derived probability distributions and occurrence probabilities for kick volume and
intensity, but statistics on these variables were not presented. Adams and Hodgson (1999) presented a qualitative study on kick varia-
bles. Tallin et al. (2000) estimated probability occurrence of kicks and their magnitude for exploratory wells in Sultanate of Oman.
Dedenuola et al. (2003) developed a comprehensive stochastic kick model using data from Shell wells at the Niger Delta. Mason and
Chandrasekhar (2005) developed probabilistic models for pore pressure, fracture gradients, kick volume, and intensity using data from

Copyright V
C 2020 Society of Petroleum Engineers

Original SPE manuscript received for review 7 August 2019. Revised manuscript received for review 27 February 2020. Paper (SPE 201200) peer approved 2 March 2020.

2020 SPE Drilling & Completion 1

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 2 Total Pages: 11

various wells around the world, but the raw data were not provided by the authors. The preceding references construct statistics for kick
variables based on averaged historical data and using well-known statistical tools. These statistics are strictly valid for the basins and
operational crews from which the data were derived. The statistics may not be valid for specific wells in which parameters deviate sig-
nificantly from the average.
By contrast, in this paper we propose a procedure to combine the opinion of different specialists, when estimating pore pressures,
fracture gradients, and other kick parameters for a specific well. This procedure does not depend on the existence of a large database of
observed kicks for the basin under analysis or reflect operational procedures of a given company. The procedure proposed herein goes
in line with the proposal of Guan et al. (2018), who suggested adjusting an arbitrary Gaussian distribution to pore pressures and fracture
gradients estimated by specialists. In this paper, however, we combine the opinion of any number of specialists without the assumption
of a Gaussian distribution. The model proposed herein can also be used as a complementary tool, when some (limited) statistics are
available for the basin being explored.
We advance toward the risk- and reliability-based design of wells by proposing an empirical random model of kick loads, which
combines the opinions of any number of experts, and by developing the reliability analysis of casing strings subject to kick loads. The
empirical kick model is presented in the section Empirical Kick Load Model. Random models and variables describing the burst
strength of line pipe are discussed in the section Burst Strength Models and Limit States. The solution of the casing reliability problem
by FORM and failure of the distributed system are discussed in the section Solution of Reliability Problem by FORM. A typical case
study well is presented in the section Typical Well of the Case Study, and results of the reliability analyses are presented in the section
Results of Reliability Analysis. We finish the paper with Concluding Remarks. This paper results from a major research and develop-
ment project, sponsored by PETROBRAS, with the objective of developing expertise and software for the probabilistic design of well
casing (Beck et al. 2017).

Empirical Kick Load Model


Kick loading is critical in all drilling operations, but it is of special concern in exploratory wells, for which correlation wells either do
not exist or are located at great distances. Pressures and axial loads are computed for the initial condition (installed casing) and during
occurrence of the actual kick.

Initial Conditions. In initial conditions, the annular pressure Pe and the internal pressure Pi acting in the casing are equal and corre-
spond to the hydrostatic load exerted by the drilling fluid of the previous phase (before casing is placed in the wellbore):

Pi ¼ Pe ¼ qmud gh; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð1Þ

where qmud is the drilling fluid density, g is the standard gravity acceleration, and h is the current depth. The axial loads acting in initial
conditions are given by Eq. 2, where Ae and Ai are the pipe outer (perimetral) and internal (hollow) area, respectively; wpipe is the linear
weight of the tube in lbm/ft; and L is the length of the casing.

Fi ¼ Pi ðLÞAi  Pe ðLÞAe þ wpipe ðL  hÞ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ

Kick Loading. Kick is a phenomenon characterized by the invasion of the wellbore by formation fluids. The invasion occurs due to
insufficient mud weight to control formation pressures. Kick can happen at any depth [total vertical depth, kick (TVDk)] between the
previous casing shoe and the target depth of the next phase. The formation influx has an initial volume Vk, initial density qk, and gas
gravity ck. It is assumed that the bottomhole pressure described in Eq. 3 is constant and equal to the hydrostatic pressure of the drilling
fluid plus the kick intensity qki. The kick intensity is a density defined as the difference between the maximum anticipated formation
pressure and planned drilling fluid weight. The invader fluid is considered to be gas with a lower density than the drilling mud, and
thus, it climbs up to the surface (Fig. 1). In Guner et al. (2017), more details are given about gas kick migration in the wellbore pressure.
The equilibrium of hydrostatic pressures at the top of the bubble is given by

Pbh ¼ g TVDk ðqki þ qmud Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð3Þ

Pb ¼ Pbh  qmud gðTVDk  Hb  hÞ  qk g Hb : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð4Þ

It is also assumed that the number of moles of the invader fluid is constant, and thus, C ¼ nR:
Pbh Vk
C ¼ nR ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð5Þ
Zb Tb
The bubble height Hb changes as it rises to the surface. It depends on the pressure Pb and the temperature T, according to
CZT
Hb ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð6Þ
Pb A
where Z is the compressibility factor, computed with the Redlich-Kwong model, T is the temperature at the desired point in the well-
bore, Pb is the pressure at the top of the bubble, and A is the wellbore area. Zb and Tb are the compressibility factor and temperature,
respectively, at influx depth.
The formation influx density is given by
c M Pb
qk ¼ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð7Þ
ZRT

where M ¼ 28:9645 lb=lbmol is the air molar mass, c is the gas gravity, and R ¼ 10:7258 ft3 psi R1 lbmol1 is the universal
gas constant.
The pressure in a kick loading varies with depth, temperature, and compressibility factor and is an implicit function of Z and Pb.
Thus, for each depth h, a nonlinear system of equations has to be solved using Newton’s method. The solution provides pressure Pb and
compressibility factor Z at depth h.

2 2020 SPE Drilling & Completion

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 3 Total Pages: 11

Conductor 600 ft

Surface 2,000 ft
Drilling h
mud

Pb

Formation
TVDk fluid Hb
Intermediate 11,100 ft

Drilling TVDk-Hb-h
mud

19,000 ft
Pbh

Fig. 1—Schematics of the wellbore in a kick load.

The casing analyzed in this study is considered to be cemented throughout its length to simplify axial load calculations and avoid
buckling. Therefore, before drilling the next phase, the casing is cemented and fixed at wellbore walls and base. Later on, the drill bit is
lowered, and the drilling of the next phase begins, starting from the shoe of the current cemented casing. The axial loads are calculated
as the sum effects of temperature Ft, ballooning Fb, and initial conditions Fi:

Ff ðhÞ ¼ Fi ðhÞ þ Fb ðhÞ þ Ft ðhÞ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð8Þ

The axial force caused by the ballooning effect is


p
Fb ðhÞ ¼ ðDrr þ Drh Þ ðOD2  ID2 Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð9Þ
4
where OD and ID are the casing outer and inner diameters, respectively; and Drr and Drh are the radial and tangential stress differen-
ces, respectively, between the initial and current casing configuration given by

Drr ¼ rfinal
r  rinitial
r ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð10aÞ

Drh ¼ rfinal
h  rinitial
h : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð10bÞ

The actual stresses are calculated by

rr ¼ Pi ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð11aÞ

ID2 Pi þ OD2 ðPi  2Pe Þ


rh ¼  : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð11bÞ
ðID2  OD2 Þ
The axial force due to temperature effects is
p
Ft ðhÞ ¼ EaDT ðOD2  ID2 Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð12Þ
4
where a ¼ 6:9106 F1 is the thermal expansion coefficient, E ¼ 30106 psi is the steel Young’s modulus, and D T is temperature
change from initial geothermal condition to drilling condition.

Proposed Empirical Kick Model. Independent variables of the kick load model presented in the section Kick Loading are the kick
volume Vk, kick intensity qk, gas gravity ck, and influx depth TVDk. These variables are estimated subjectively by field experts. The
most conservative estimate for kick volume, for instance, is to assume the same volume as the wellbore or a wellbore full of gas. This is
a very conservative assumption, which is also very unlikely to happen. Estimates for influx depth depend on expert readings of the seis-
mic data. Estimates for kick intensity and gas gravity depend on proper anticipation of the type of hydrocarbons to be encountered
while drilling. For these variables, there is not enough data to construct a proper probabilistic model. The uncertainty in these variables
can be classified as epistemic, that is, related to a lack or imprecision of knowledge.
Evidence or Dempster-Shafer Theory provides the theoretical framework for handling nonobjective uncertainty. In engineering anal-
yses, it is widely accepted that a possibilistic representation of epistemic uncertainties is appropriate. Possibilistic representations
include, for instance, interval analysis (Moens and Vandepitte 2005; Hanss and Turrin 2010; Verhaeghe et al. 2013), fuzzy numbers
(Moller and Beer 2004), fuzzy randomness (Moller and Beer 2004, 2008), or generally imprecise probabilities (Moller and Beer 2004,
2008; Moens and Vandepitte 2005). The combined handling of possibilistic and probabilistic uncertainties, however, is quite

2020 SPE Drilling & Completion 3

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 4 Total Pages: 11

challenging, due to nested loops of uncertainty quantification approaches. Such nesting should be avoided in large problems, such as
the one presented here, in which reliability analysis has to be repeated for all casing depths. Therefore, an empirical distribution model
is adopted herein.
From the reading of seismic data records, and from eventual correlation wells, individual experts arrive at estimates of the opera-
tional window (pore pressure and fracture gradients at every depth) and other variables. Due to subjectivity of the interpretations,
experts are asked to provide a conservative and a nonconservative estimate of each variable. By convention, and as a way of guiding
the evaluation, the lower value (say, x10) of a variable X is associated with a probability of 10% (p10), and the larger value (x90) is asso-
ciated with a probability of 90%. The expert, however, does not have objective data to evaluate or check these probabilities; hence, the
values (x10 and x90) are more relevant that the corresponding probabilities. Which value among x10 and x90 is the conservative one
depends on the actual variable. For loading variables, x90 is the conservative value, which has a 10% probability of being exceeded in
the unfavorable sense.
We assign an empirical probability distribution function to the values estimated by the experts. When only the lower and upper
bounds of a (random) variable are known, the probability distribution that maximizes Shannon entropy of information is the uniform
distribution (Kapur and Kesavan 1992). However, the information provided by the experts, in theory, refers to the 10 and 90% probabil-
ities, not to the bounds of the variable. In theory, it would be possible to linearly extrapolate an empirical uniform cumulative distribu-
tion function (CDF) from known values (x10, p10) and (x90, p90), to obtain the variable bounds. However, any extrapolation of this kind
is arbitrary and dangerous in terms of structural reliability analyses. Accordingly, we prefer to consider the x10 and x90 values as the
bounds of the variable, from which to construct the empirical CDF.
The empirical kick model becomes more interesting when the opinions of two or more experts are combined. We consider that each
expert provides an independent estimate of the x10 and x90 values of each relevant variable. These values are organized in ascending
order, irrespective of being conservative or unconservative values. The smallest value becomes the lower bound of the variable, and the
largest value becomes the upper bound. The corresponding empirical probabilities are p0 and p100. The remaining values receive the
empirical CDF according to the number of points n, which is twice the number of experts:
i1
FX ¼ for 1  i  n: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð13Þ
n1
The probability density function fX(x) is obtained as the derivative of FX ðxÞ. An arbitrary empirical distribution, obtained with this
logic, is illustrated in Fig. 2. In the figure, the opinions of three experts (E1, E2, and E3) are combined. Notice that the empirical CDF
does not correspond exactly to the subjective probabilities (10 or 90%) of any of the experts; however, the raw information they convey
(x10 and x90) is logically combined. In the case study example (to follow), the opinions of three experts about kick variable volume Vk,
intensity qk, gas gravity ck, and influx depth TVDk are combined. The hypothetical x10 and x90 values obtained by each expert are pre-
sented in Table 1. The empirical distributions obtained from these values are illustrated in Fig. 3. The extreme values of kick variables
correspond to physical limitations. Kick volume, for instance, cannot be larger than the casing volume; influx depth cannot be deeper
than drilling target depth, or shallower than the casing shoe.

1.0 1.0

p90E3 p90E1
p90E2
0.8 0.8

0.6 0.6
CDF

PDF

0.4 0.4

0.2 0.2
p10E2 p10E3
p10E1
0 0
x10E2 x10E3 x90E2 x90E3 x90E1
x10E1

Fig. 2—Arbitrary empirical random variable constructed from the opinions of three experts. PDF 5 probability density function.

In the case study example, the effect of the empirical kick model proposed herein is investigated by comparing reliability indices
obtained with the empirical distributions illustrated in Fig. 3, with results obtained assuming a uniform distribution between the upper
and lower bounds shown in Fig. 3.

Burst Strength Models and Limit States


The resistance random variables are model errors for burst pressure Me, steel ultimate stress ru, and casing thickness t. All are consid-
ered to follow a Gaussian distribution. The load random variables are the kick volume Vk, the gas gravity ck, the kick intensity qki, and
the influx depth TVDk. The load variables are assumed to be empirically distributed.
The random variables describing empirical kick and pipe strength are grouped in vector X, given by
0 1 0 1
X1 Me
B X2 C B ru C
B C B C
B X3 C B t C
B C B C
B X4 C ¼ B Vk C: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð14Þ
B C B C
B X5 C B qki C
B C B C
@ X6 A @ ck A
X7 TVDk

4 2020 SPE Drilling & Completion

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 5 Total Pages: 11

Lo a di ng R a nd o m V a r i a b l e s
Vk (bbl) ρk (ppg) γgas (adm) TVDinflux (ft)
Expert p10 p90 μ p10 p90 μ p10 p90 μ p10 p90 μ
1 1 260 0.00 1.00 0.48 0.72 11,900 14,400
2 50 430 273.83 0.30 1.10 0.825 0.55 0.83 0.683 12,300 16,600 14,533
3 102 800 0.55 2.00 0.61 0.91 13,000 19,000

Table 1—Hypothetical X10 and X90 values by three experts for four kick variables (kick volume, kick intensity, gas gravity, and influx depth)
employed in the case study example.

1.0 0.0041 1.0 3.3

0.8 0.0033 0.8 2.7

0.6 0.0024 0.6 2.0

PDF
CDF

PDF

CDF
0.4 0.0016 0.4 1.3

0.2 0.0082 0.2 0.67

0 0 0
0 200 400 600 800 0.5 0.6 0.7 0.8 0.9
Kick Volume (bbl) Gas Gravity

1.0 2.0 1.0 0.0005

0.8 1.6 0.8 0.0004

0.6 1.2 0.6 0.0003


CDF

PDF
PDF

CDF

0.4 0.8 0.4 0.0002

0.2 0.4 0.2 0.0001

0 0 0 0
0 0.5 1.0 1.5 2.0 12,000 13,000 14,000 15,000 16,000 17,000 18,000 19,000
Kick Intensity (ppg) Influx TVD (ft)

Fig. 3—Empirical distributions of kick volume Vk, kick intensity qk, gas gravity ck, and influx depth TVDk used in the case
study example. PDF 5 probability density function.

Barlow. The Barlow formula computes the casing burst resistance PR considering only the yield stress and the thickness.

2 ry
PR ðry ; tÞ ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð15Þ
OD=t

The Barlow limit-state equation is the difference between the Barlow pressure resistance and the pressure in a kick loading:

GðXÞ ¼ PR ðry ; tÞ  ½Pb ðVk ; ck ; qki ; TVDk Þ  Pe : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð16Þ

The Barlow safety factor is

PR ðry ; tÞ
SFB ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð17Þ
Pi  Pe

von Mises. The von Mises yield criterion is used to represent plastic failure in metals. The von Mises stress is given by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rmises ¼ r2z þ r2r þ r2h  rr rh  rz rh  rz rr ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð18Þ

where rr and rh are given in Eq. 11. The axial stress rz is obtained from the actual loads resulting from the combination of temperature,
ballooning, and initial conditions effects:
Ff
rz ¼ p : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð19Þ
ðOD2  ID2 Þ
4

2020 SPE Drilling & Completion 5

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 6 Total Pages: 11

The limit-state equation assumes that failure occurs if von Mises stress resulting from a kick is greater than the yield stress:

GðXÞ ¼ ry  rmises ðt; Vk ; ck ; qki ; TVDk Þ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð20Þ

The von Mises safety factor is


ry
SFVM ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð21Þ
rmises

Klever-Stewart. The Klever-Stewart model (Klever and Stewart 1998; Stewart and Klever 1998) for ductile rupture of pipes computes
casing burst resistance PR taking into account axial forces. This makes it more accurate than the Barlow model because in practice all
casings are under combined loads (i.e., pressures and axial loads). The model is implemented following ISO 10400 (2011, page 26).
The limit-state equation is given by

GðXÞ ¼ Me PR ðru ; tÞ  Pb ðVk ; ck ; qki ; TVDk Þ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð22Þ

The Klever-Stewart safety factor is


PR ðru ; tÞ
SFKS ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð23Þ
Pi  Pe
In ISO 10400 (2011), several models for burst strength were compared in terms of accuracy. The Klever-Stewart model was found
to be one of the best. Statistics for the model error random variable Me were obtained, as shown in Table 2. Model error statistics are
obtained by comparing burst strength in experimental tests with model predictions. Model error is known to be a significant source of
uncertainty in engineering. In this paper, we consider the Klever-Stewart as the most accurate model. However, we also compute safety
factors and reliability indices by Barlow and von Mises equations, for completeness and for reference.

Var iab l e Symbol D is t r i b ut i on Mea n C V ( %)


Model error ME Normal 1.0040 4.70
Ensemble Ultimate stress σu Normal 1.1000 4.22
properties Wall thickness t Normal 1.0069 2.59
Governing Ultimate stress σu Normal 1.0938 3.33
properties A Wall thickness t Normal 0.9876 0.20
Governing Ultimate stress σu Normal 1.10 4.22
properties B Wall thickness t Normal 0.9856 0.10

Table 2—Random strength variables for burst pressure ISO 10400 (2011). CV ¼ coefficient of variation.

Strength Random Variables. The Klever-Stewart burst model (Klever and Stewart 1998; Stewart and Klever 1998) is based on four
interconnected concepts:
• An equation based on equilibrium and plasticity for ductile rupture of the tube with known diameter and wall thickness
• A strength reduction factor to account for wall thickness reductions originating in imperfections not detected during quality
control
• A criterion for minimum tenacity
• A transition to necking under very large tensile loads
The Klever-Stewart burst strength equation is presented in Rahman and Chilingarian (1995) and ISO 10400 (2011). In the ISO
10400 (2011) document, an ample set of statistics is presented for the ultimate stress of pipeline material and for manufacturing toleran-
ces and deviations. These statistics are classified in three groups: ensemble, governing A, and governing B properties. The ensemble
properties are averages for all manufacturers and all line pipe classes. The governing A and B are more critical subsets, which reflect a
smaller group of manufacturers, subgroup of larger deviations from nominal measures, and more critical combination of factors. Statis-
tics taken from ISO 10400 (2011) for rupture stress and wall thickness are shown in Table 2.
Uncertainties in rupture stress and wall thickness are propagated through the Klever-Stewart burst strength model. Seventeen tubu-
lars were considered in the analysis, and it was found that the subset of statistics called governing properties B (ISO 10400 2011) lead
to the smallest burst strengths and to the worst results for pipe burst reliability analysis. From ISO 10400 (2011), the ultimate stress is
defined as ru ¼ 1.14 ry.

Solution of Reliability Problem by FORM


Failure Probability for Individual Limit States. For each limit-state function presented in the section Burst Strength Models and
Limit States, the failure probability is evaluated as
ð
Pfi ¼ P½gi ðxÞ  0 ¼ fX ðxÞdx; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð24Þ
gi ðx0Þ

where fX ðxÞ denotes the joint probability density of the random variable vector X.
In the so-called FORM, Eq. 24 is solved indirectly by means of a mapping from the space of the original random variables
(X -space) to the so-called standard Gaussian U-space, where all components of vector u are independent and identically distributed
standard Gaussian random variables (Beck and Melchers 2018):

u ¼ T½x; gui ðuÞ ¼ gi ½T 1 ðuÞ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð25Þ

6 2020 SPE Drilling & Completion

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 7 Total Pages: 11

This mapping can be accomplished by means of the Nataf or Rosenblatt transformations. In the standard space, the joint probability
density fU ðuÞ is rotationally symmetric; hence, the point u over the limit-state function gui ðu; tÞ, which is the closest to the origin, rep-
resents the most probable failure point, also known as the design point. This feature also allows the search for the design point to be
cast as a constrained optimization problem:

ui ¼ arg minjjujj; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð26Þ

subject to gui ðuÞ ¼ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð27Þ

From Eq. 27, bi ¼ jju jj is the Hasofer-Lind reliability index, which becomes the distance between u and the origin of the standard
space. The FORM method, therefore, consists of finding the design point u for each limit state and approximating the original limit-
state function gui ðu; tÞ by a tangent hypersurface at the design point. The first-order approximation of the failure probability becomes

Pfi ¼ P½gi ðxÞ  0 ¼ P½gui ðUÞ  0  Uðbi Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð28Þ

where P[] is the probability operator. The FORM approximation assumes that the most probable failure point is unique, which is true in
a large number of applications and when limit-state functions are not excessively nonlinear. In this paper, this assumption is verified by
means of the Monte Carlo simulation.
In Eq. 28, b is the reliability index, which is a linear geometrical measure of the failure probability. For convenience, in this paper
we report bs instead of failure probabilities. The reliability index has an order of magnitude that is similar to that of the traditional
safety factor, to which bs are compared.
Many algorithms can be used for solving the constrained optimization problem in Eq. 27. The classical Hasofer, Lind, Rackwitz,
and Fiessler algorithm, and its improved version (the iHLRF algorithm) produced unstable results, in particular due to the presence of
load variables with uniform (or uniform by parts, in case of the empirical model) distributions. Uniform distributions are known to pro-
duce instabilities in the search for the design point (Youn and Choi 2004; Beck and Silva 2016). Hence, a novel algorithm proposed by
Jin-Xin and Ping (2011) has been used herein. The gradient vector rgu ðuk Þ of the limit-state function with respect to u evaluated at uk
is evaluated analytically using Mathematica software (Version 12.1, Wolfram Research, Champaign, Illinois, USA). An important sub-
product of the FORM solution are the so-called sensitivity factors a2i , which are obtained from

rgu ðuk Þ
a¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð29Þ
jjrgu ðuk Þjj

where a ¼ fa1 ; a2 ; …; an g. Because a is a unitary vector, the sensitivity factors show the relative contribution of each random variable
toward the failure probability, for linear or approximately linear limit-state functions.

Failure Probability for the Distributed System. The casing string forms what is known as a distributed system, in which failure can
occur at any depth. Distributed systems require a space-variant reliability analysis, constructed from a failure rate, which is integrated
over the system length (Beck and Melchers 2018; Napa-Garcia et al. 2018). The correlation between failure probabilities evaluated at
different depths plays a crucial role in such models. Division of the casing string in segments of arbitrary length, to use a system reli-
ability formulation, is questionable and again requires evaluation of correlation between failure probabilities evaluated for different seg-
ments. To circumvent these problems, in this paper we assume full correlation between failure probabilities evaluated at different
depths. This assumption is based on the existence of common strength variables (model error, ultimate stress, and pipe thickness) and
common load variables (kick load). Because the ultimate stress is found to be a relevant variable (case study to follow), this assumption
could be reevaluated if correlation information were available for the yield stress at different points of a tube, or between different
tubes. With the assumption of full correlation, failure probability for the series system problem (along the casing string) becomes the
failure probability of the weakest link, or the maximum failure probability evaluated at any depth.

Typical Well of the Case Study


As an application example, the well illustrated in Fig. 4 is considered. This is a theoretical well, taken from reference Rahman and
Chilingarian (1995). Strength of the intermediate casing is considered for a hypothetical kick occurring during drilling of the next phase
(11.1 to 19,000 ft). The analytical formulation for axial loads and kick pressures is verified by means of comparison to an existing com-
mercial software. Safety factors are also verified before the reliability analysis.
The data considered in this deterministic verification was: drilling mud qm ¼ 17.9 ppg, kick intensity qki ¼ 0.825 ppg, gas
gravity cg ¼ 0.683, and kick volume Vk ¼ 273.83 bbl. The casing in the study has outer and inner diameters of OD ¼ 13.375 in. and
ID ¼ 11.937 in., self-weight of 86 lbm/ft, and steel grade of 125,000 psi.
The temperature profiles considered in the example are illustrated in Fig. 4. In initial conditions, the reservoir temperature is 392 F,
and the surface temperature is 39.2 F. During next-phase drilling, the reservoir temperature is also 392 F, but the surface temperature
is 86 F.
The analytical formulation for axial loads and internal and external pressures is verified by means of comparison with solutions obtained
using the WellcatÔ (Halliburton Landmark, Houston, Texas, USA) software. The axial loads are illustrated in Fig. 4, in terms of initial con-
dition (casing placed into wellbore) and after being cemented, under kick loading. Nominal kick pressures are illustrated in Fig. 4. A very
good agreement is obtained between our formulation and the Wellcat software for axial loads and kick pressures, as observed.
The Barlow, von Mises, and Klever-Stewart safety factors are shown in Fig. 4. Minimal safety factors, at the bottom of the well, are
very close to industry practice (1.5 to 2.0). Safety factors become larger toward the surface because the same line pipe is used for the
whole intermediate casing. This is also usual industry practice.

Results of Reliability Analysis


Results for Empirical Kick Model. Reliability of the intermediate casing, for rupture due to a random kick occurring during drilling
of the next phase (11.1 to 19,000 ft), is considered. The wellbore is discretized every 555 ft and, at each depth, the FORM method is

2020 SPE Drilling & Completion 7

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 8 Total Pages: 11

used to compute the reliability index and sensitivity factors of each random variable. Results obtained for the Barlow, von Mises, and
Klever-Stewart burst strength models are compared for the empirical kick model of Table 1 and Fig. 3, in terms of reliability indices
and sensitivity factors.

Safety Factors Kick Pressures


0 0
Pi
–2,000 –2,000 Pe
ΔP
–4,000 –4,000 Pe –Wellcat
Depth (ft)

Depth (ft)
Pi –Wellcat
–6,000 –6,000 ΔP–Wellcat

–8,000 –8,000
von Mises
–10,000 Klever-Stewart –10,000
Barlow
–12,000 –12,000
1.5 2.0 2.5 3.0 0 3,000 6,000 9,000 12,000
psi

Axial Force Temperatures


0 0
Final
–2,000 –2,000 Initial

–4,000 –4,000
Depth (ft)

Depth (ft)

–6,000 –6,000
Fi
–8,000 Ff –8,000
Fi –Wellcat
–10,000 Ff –Wellcat –10,000

0 200 400 600 800 50 100 150 200 250


Force (lbf × 1000) °F

Fig. 4—Loading conditions during kick loading, considering the means of random variables.

Fig. 5 shows safety coefficients, reliability indices and sensitivity factors for the Barlow and von Mises limit states. Fig. 5 considers
a casing with 11,100-ft depth under a kick-loading event taking place when the next phase is being drilled at 19,000 ft with a 17.9 ppg
mud weight. The mud in the casing annulus has a 12-ppg weight. The same results are shown in Fig. 6 for the Klever-Stewart model.
In all cases, we observe that reliability indices grow with safety factors, as expected, although the relationship is clearly nonlinear. A
general agreement can be observed between safety factors and reliability indices obtained for the Barlow, von Mises, and Klever-
Stewart limit states. Some relevant differences are observed in terms of sensitivity factors. For Barlow and von Mises, the most-
relevant variable is the ultimate stress at shallower depths. Farther down the well, the kick volume becomes the most relevant source of
uncertainty. For the Klever-Stewart model, the model error variable is the most relevant, at shallower depths, with a trend similar to
ultimate stress, and the kick volume becomes the most relevant variable at greater depths. Sensitivity factors are evaluated from Eq. 29.
In Figs. 5 and 6, we observe that as safety factors vary between 1.2 and 3, reliability indices vary from 1.6 to more than 8. The target
annual reliability index in civil engineering is 3.7 for high consequences of failure and high cost of the safety measure. Hence, we
observe in Figs. 5 and 6 that there is ample margin for optimizing the intermediate casing of the studied well because reliability indices
of more than 5 have little contribution to actual well safety. This optimization, however, also depends on other factors, such as the
appropriateness of using one (or more) different casings in the upper parts of the intermediate casing string.

Results for Uniform Kick Model. The results shown in Figs. 5 and 6 show that loading uncertainty is certainly very relevant for well
casing reliability studies. To evaluate the effect of the empirical kick model proposed here, results for the Klever-Stewart strength
model are also computed for a uniform kick model; in this model, kick variables (Vk, qk, ck, and TVDk) have the same lower and upper
bounds shown in Table 1, but a uniform distribution is assumed in this range. Fig. 6 also shows the results for a random kick given by
uniform random variables. Reliability indices are larger, and sensitivity factors for load variables are reduced, confirming that the
empirical kick model produces a significant effect. Clearly, such an effect will depend largely on expert opinion, as reflected in Fig. 3
and Table 1.

Casing with Reduced Ultimate Stress. Reliability of the casing string is also evaluated for a situation with reduced material strength
(derating). The line pipe with original ry ¼ 125,000 psi is replaced by pipes of reduced strength (ry ¼ 110,000 psi). Results are shown
in Fig. 7 for the Klever-Stewart strength model and for the empirical and uniform random kick models. Reliability indices and safety
factors are reduced, as expected, in comparison to the stronger pipe case. The contribution of strength variables is reduced and that of
load variables is increased, especially for the empirical kick model.

8 2020 SPE Drilling & Completion

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 9 Total Pages: 11

Barlow Barlow
0 0

k
–2,000 –2,000
Vk
ρk
–4,000 –4,000
Depth (ft) σy

Depth (ft)
t
–6,000 –6,000
TVDk
–8,000 –8,000

Reliability index –10,000


–10,000
Safety factor

1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

von Mises von Mises


0 0

–2,000 –2,000

–4,000 –4,000
Depth (ft)

Depth (ft)
–6,000 –6,000 k
Vk
–8,000 ρk
–8,000
σy
–10,000 Safety factor t
–10,000
Reliability index TVDk
–12,000
1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Fig. 5—Safety factors and reliability index (left), and sensitivity coefficients (right) during kick loading. Barlow (both upper plots)
and von Mises (both lower plots).

Klever-Stewart Klever-Stewart
0 0

–2,000 –2,000

–4,000 –4,000 k
Depth (ft)

Depth (ft)

Vk
–6,000 –6,000 Me
ρk
–8,000 –8,000 σy
t
Reliability index
–10,000 –10,000 TVDk
Safety factor

–12,000 –12,000
1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Klever-Stewart Klever-Stewart
0 0

–2,000 –2,000

–4,000 –4,000
Depth (ft)

Depth (ft)

k
–6,000 –6,000 Vk
Me
ρk
–8,000 –8,000
σy
Reliability index t
–10,000 –10,000
Safety factor TVDk
–12,000 –12,000
1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Fig. 6—Safety factors and reliability index (left), and sensitivity coefficients (right) during kick loading. Empirical (both upper
plots) and uniform (both lower plots) kick models. This figure shows the results considering a yield stress of ry 5 125,000 psi.

2020 SPE Drilling & Completion 9

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 10 Total Pages: 11

Klever-Stewart Klever-Stewart
0 0

–2,000 –2,000

–4,000 k
–4,000
Depth (ft) Vk

Depth (ft)
–6,000 Me
–6,000
ρk
–8,000 σk
–8,000
t
–10,000 Reliability index TVDk
–10,000
Safety factor
–12,000 0
1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Klever-Stewart Klever-Stewart
0 0

–2,000 –2,000

–4,000 –4,000
Depth (ft)

Depth (ft)
–6,000 k
–6,000
Vk
Me
–8,000 –8,000 ρk
σk
–10,000 Reliability index –10,000 t
Safety factor
TVDk
–12,000 –12,000
1 2 3 4 5 6 7 8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Fig. 7—Safety factors and reliability index (left), and sensitivity coefficients (right) during kick loading. Empirical (both upper
plots) and uniform (both lower plots) kick models. This figure shows the results considering a reduced strength material
(ry 5 110,000 psi).

Concluding Remarks
We addressed the reliability of casing strings of exploratory wells, which are subject to significant uncertainties. Reliability analysis is
well-accepted as a tool to optimize well casing design, but most work done so far has addressed the strength side of the problem, includ-
ing manufacturing tolerances and uncertainty of strength prediction models. Few papers in the literature address the large uncertainties
appearing in the loading parameters, with kick loading being a concern. An extensive literature review revealed some papers building
kick parameter statistics from averaged historical data. Such models are valid for the basins and operational crews for which the data
were obtained. Such models may not be valid for individual wells departing significantly from average parameters. In this paper, we
proposed a novel empirical model in which the opinion of any number of specialists can be combined, in the reliability analysis of a
specific well, in a basin for which historical kick data are insufficient or unavailable. The proposed empirical kick model was shown to
fit seamlessly into conventional reliability analyses. Equations for axial loads and internal and external pressures were checked in com-
parison with the results from a commercial software. Three burst strength models were compared and shown to provide similar results
in terms of safety factors and reliability indices but with different results in terms of sensitivity factors (random variable contributing
factors). The Klever-Stewart model was preferred because it is known to be one of the most precise models for burst strength, and
because model error statistics are known. Results have shown that uncertainty in load variables is significant for bursts under kick
loads, especially at greater well depths. The empirical kick model was shown to have a significant effect, in comparison to results con-
sidering uniform distributions for random kick variables. It was also shown that for lower-strength pipes, reliability indices are reduced
and sensitivity factors change significantly along the well depth.

Acknowledgments
We thankfully acknowledge financial support from ANP-Brazilian National Agency of Petroleum, Natural Gas and Biofuels GRANT
CENPES No. 0050.0089855.14.2.

References
Adams, A. 1995. QRA for Casing/Tubing Design. Paper presented at the Seminar of Norwegian HPHT Program, Stavanger, Norway.
Adams, A. and Glover, S. 1998. An Investigation into the Application of QRA in Casing Design. Paper presented at the SPE Applied Technology Work-
shop on Risk Based Design of Well Casing and Tubing, Houston, Texas, USA, 7–8 May. SPE-48319-MS. https://doi.org/10.2118/48319-MS.
Adams, A. and Hodgson, T. 1999. Calibration of Casing/Tubing Design Criteria By Use of Structural Reliability Techniques. SPE Drill & Compl 14 (1):
21–27. SPE-55041-PA. https://doi.org/10.2118/55041-PA.
Adams, A., Parfitt, S., Reeves, T. et al. 1993. Casing System Risk Analysis Using Structural Reliability. Paper presented at the SPE/IADC Drilling Con-
ference, Amsterdam, The Netherlands, 22–25 February. SPE-25693-MS. https://doi.org/10.2118/25693-MS.
Adams, A. J., Warren, A. V. R. and Masson, P. C. 1998. On the Development of Reliability-Based Design Rules for Casing Collapse. Paper presented at
the SPE Applied Technology Workshop on Risk Based Design of Well Casing and Tubing, The Woodlands, Texas, USA, 7–8 May. SPE-48331-MS.
https://doi.org/10.2118/48331-MS.
Beck, A., Gomes, W., Silva, J. C. et al. 2017. Projeto Probabilı́stico de Revestimento de Poços, Relatório Final Consolidado. Final consolidated technical
report, Universidade de São Paulo, São Paulo, Brazil.
Beck, A. and Melchers, R. 2018. Structural Reliability Analysis and Prediction. New York, New York, USA: Wiley.

10 2020 SPE Drilling & Completion

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010


DC201200 DOI: 10.2118/201200-PA Date: 18-May-20 Stage: Page: 11 Total Pages: 11

Beck, A. T. and Silva, C. R. Jr. 2016. Strategies for Finding the Design Point under Bounded Random Variables. Struct Saf 58: 79–93. https://doi.org/
10.1016/j.strusafe.2015.08.006.
Dedenuola, D., Iyamu, E., and Adeleye, O. 2003. Stochastic Approach to Kick Tolerance Determination in Risk Based Designs. Paper presented at the
SPE Annual Technical Conference and Exhibition, Denver, Colorado, USA, 5–8 October. SPE-84174-MS. https://doi.org/10.2118/84174-MS.
Guan, Z., Sheng, Y., Xi, C. et al. 2018. Oil & Gas Drilling Risk Analysis Utilizing Quantitative Risk Assessment. J Appl Sci Eng 21 (4): 541–546.
https://doi.org/10.6180/jase.201812_21(4).0005.
Guner, S., Elshehabi, T., and Bilgesu, H. I. 2017. The Effects of Gas Kick Migration on Wellbore Pressure. Paper presented at the AADE National Tech-
nical Conference and Exhibition, Houston, Texas, USA, 11–12 April. AADE-17-NTCE-132.
Hanss, M. and Turrin, S. 2010. A Fuzzy-Based Approach to Comprehensive Modeling and Analysis of Systems with Epistemic Uncertainties. Struct Saf
32 (6): 433–441. https://doi.org/10.1016/j.strusafe.2010.06.003.
ISO 10400, Petroleum and Natural Gas Industries—Equations and Calculations for the Properties of Casing, Tubing, Drill Pipe and Line Pipe Used as
Casing or Tubing. 2011. Geneva, Switzerland: International Organization for Standardization.
Jin-Xin, G. and Ping, Y. 2011. A Robust Iterative Algorithm for Structural Reliability Analysis. Struct Multidiscip Optim 43: 519–527. https://doi.org/
10.1007/s00158-010-0582-y.
Kaiser, M. 2009. Modeling the Time and Cost To Drill an Offshore Well. Energy 34 (9): 1097–1112. https://doi.org/10.1016/j.energy.2009.02.017.
Kapur, J. and Kesavan, H. 1992. Entropy Optimization Principles with Applications. Waltham, Massachusetts, USA: Academic Press.
Klever, F. J. and Stewart, G. 1998. Analytical Burst Strength Prediction of OCTG with and without Defects. Paper presented at the SPE Applied Tech-
nology Workshop on Risk Based Design of Well Casing and Tubing, The Woodlands, Texas, USA, 7–8 May. SPE-48329-MS. https://doi.org/
10.2118/48329-MS.
Lewis, D., Brand, P., Whitney, W. et al. 1995. Load and Resistance Factor Design for Oil Country Tubular Good. Paper presented at the Offshore Tech-
nology Conference, Houston, Texas, USA, 1–4 May. OTC-7936-MS. https://doi.org/10.4043/7936-MS.
Mason, S. and Chandrasekhar, S. 2005. Stochastic Kick Load Modeling. Paper presented at the SPE High Pressure/High Temperature Sour Well Design
Applied Technology Workshop, The Woodlands, Texas, USA, 17–19 May. SPE-97564-MS. https://doi.org/10.2118/97564-MS.
Moens, D. and Vandepitte, D. 2005. A Survey of Non-Probabilistic Uncertainty Treatment in Finite Element Analysis. Comput Methods Appl Mech Eng
194 (12–16): 1527–1555. https://doi.org/10.1016/j.cma.2004.03.019.
Moller, B. and Beer, M. 2004. Fuzzy Randomness—Uncertainty in Civil Engineering and Computational Mechanics. Berlin Heidelberg, Germany:
Springer-Verlag.
Moller, B. and Beer, M. 2008. Engineering Computation Under Uncertainty—Capabilities of Non-Traditional Models. Comput & Struct 86 (10):
1024–1041. https://doi.org/10.1016/j.compstruc.2007.05.041.
Napa-Garcia, G., Beck, A., and Celestino, T. 2018. Risk Analysis of Fractured Rock Mass Underground Structures. Georisk: Assess Manage Risk Eng
Sys Geohazards 12 (9): 123–134. https://doi.org/10.1080/17499518.2017.1397275.
Rahman, S. and Chilingarian, G. 1995. Casing Design, Theory and Practice, Developments in Petroleum Science. Amsterdam, The Netherlands:
Elsevier.
Stewart, G. and Klever, F. J. 1998. Accounting for Flaws in the Burst Strength of OCTG. Paper presented at the SPE Applied Technology Workshop on
Risk Based Design of Well Casing and Tubing, The Woodlands, Texas, USA, 7–8 May. SPE-48330-MS. https://doi.org/10.2118/48330-MS.
Tallin, A., Paslay, P., Cernocky, E. et al. 2000. Risk Assessment of Exploration Well Designs in the Oman Ara Salt. Paper presented at the SPE Annual
Technical Conference and Exhibition, Dallas, Texas, USA, 1–4 October. SPE-63130-MS. https://doi.org/10.2118/63130-MS.
Verhaeghe, W., Desmet, W., Vandepitte, D. et al. 2013. Interval Fields To Represent Uncertainty on the Output Side of a Static FE Analysis. Comput
Methods Appl Mech Eng 260: 50–62. https://doi.org/10.1016/j.cma.2013.03.021.
Wylie, W. and Visram, A. 1990. Drilling Kick Statistics. Paper presented at the SPE/IADC Drilling Conference, Houston, Texas, USA, 27 February–2
March. SPE-19914-MS. https://doi.org/10.2118/19914-MS.
Youn, B. and Choi, K. 2004. An Investigation of Nonlinearity of Reliability-Based Design Optimization Approaches. J Mech Des 126 (3): 403–411.
https://doi.org/10.1115/1.1701880.

2020 SPE Drilling & Completion 11

ID: jaganm Time: 16:57 I Path: //chenas03.cadmus.com/Home$/jaganm$/SA-DC##200010

Das könnte Ihnen auch gefallen