Measurement 127 (2018) 546–553

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Prediction analysis of downhole tubing leakage location for offshore gas T

production wells
⁎
Shengnan Wua, Laibin Zhanga, Jianchun Fana, , Ximing Zhangb, Yangfan Zhouc, Deguo Wanga
a
College of Mechanical and Transportation Engineering, China University of Petroleum-Beijing, Beijing, China
b
China North Vehicle Research Institute, Beijing, China
c
Beijing Municipal Institute of Labour Protection, Beijing, China

A R T I C LE I N FO A B S T R A C T

Keywords: Industry experience has shown that downhole tubing leakage is the most challenging issue resulting in sustained
Offshore gas well casing pressure in natural gas production wells. This article presents a novel tool based on pressure difference
Sustained casing pressure and probability distribution for predicting the tubing leakage location for an offshore gas production well with
Tubing leakage prediction sustained casing pressure. Combined with the formation information, well structure, annular fluid level and
Bayesian inference
dynamic production parameters of a gas well, the model of pressure difference between tubing and production
Probabilistic distribution
casing annulus is established. However, these pressure-based predictions suffer from various sources of un-
certainties, such as the variations in reservoir conditions, and measurement errors. Bayesian inference is in-
troduced to handle these uncertainties in leakage location forecasting effectively. Dynamic confidence intervals
under two kinds calculation modes are applied to calculate the leakage depth within a certain range. The
probabilistic distribution of leakage location is statistically predicted under different wellhead pressure differ-
ence using Monte Carlo simulation. A case study on a specific offshore gas well with the field data is presented to
illustrate the feasibility of the proposed method and to demonstrate that the downhole tubing leakage location
prediction contributes to the safety and integrity management of offshore production wells in an economical and
convenient way.

1. Introduction prediction is necessary to promote integrity management of offshore

gas production wells in a safe way.
In a steady-state production gas well, the wellhead pressure of Many theoretical models have been developed for SCP prediction
casing annulus should be zero after a small volume of fluid caused by and annulus pressure calculation. These models mainly focus on the gas
thermal expansion effects has been bled. If it rebuilds to the same migration behaviors in the cement and the mud of annulus, with some
pressure level when the needle valve is closed, the casing is considered influences on the annulus pressure behavior considered, such as well-
to be exhibiting sustained casing pressure (SCP) [1]. SCP may be prone bore temperature, formation pressure and casing deformation [6,7].
to pose a serious threat to well integrity and production safety without The typical mathematical model has been developed by [1] to predict
being well handled. The well integrity failures and other problems have wellhead pressure in casing-casing annulus with limitation of applica-
been investigated in [2], mostly accounting for 18% in 406 Norway’s tion in tubing-production casing one. Yang et al. [8] have established
offshore wells. High annulus pressure will cause a collapse failure of the model for the interactions among the casing-cement-formation
casing [3] or liner [4]. If the casing with SCP fails, the next outer casing system, taking into account deformation of the casing. A semi-steady
string generally would not be able to withstand the pressure, which state model has been presented to predict annulus pressure buildup
may result in serious accidents even with huge economic losses. When considering temperature effects and other factors in subsea wells[9].
the measured wellhead pressure is greater than maximum allowable Based on the pressure prediction model, the risk of wells with SCP can
wellhead operating pressure, the casing pressure needs to be managed be quantitatively estimated by utilizing the early-time pressure buildup
on a case-by-case basis [5]. Downhole tubing leakage is a major con- data [2]. A pressure-balance-based approach has also been proposed to
tributing source which is responsible for SCP in the tubing-production determine temperature and pressure distributions of tubing and an-
casing annulus, with the risk of extra production disturbance and nulus fluid [10].
wellbore integrity failure. Therefore, downhole tubing leakage Conventional downhole leakage is detected by logging techniques,

⁎
Corresponding author.
E-mail address: fanjianchun@cup.edu.cn (J. Fan).

https://doi.org/10.1016/j.measurement.2018.05.084
Received 1 July 2017; Received in revised form 9 May 2018; Accepted 23 May 2018
Available online 24 May 2018
0263-2241/ © 2018 Elsevier Ltd. All rights reserved.
S. Wu et al. Measurement 127 (2018) 546–553

testing and analytical approaches. Ultrasonic logging method for casing transfer coefficient [22].
leak detection has been developed by an ultrasonic signature [11]. Combining Eqs. (1) and (2), based on the energy conservation
Temperature logging [12] and the temperature analytical thermal principle, the fluid temperature distribution in the tubing can be given
model [13] have been presented for wellbore leakage of carbon dioxide by [23]:
with large leaks. Pulse testing in frequency domain has been used to
Cpm Gt (k cem + Rto U0 TD ) ⎞−1
detect and characterize the location and geometry of the carbon dioxide Cpm Gt (kcem + Rto U0 TD ) ⎡ Δz·⎛ ⎜
2πRto U0 k ce m
⎟ ⎤⎛
g
Tfo = Tho + ⎢1−e ⎝ ⎥⎜
⎠ −
leak [14]. The acoustic detection technology based on Fourier trans- 2πRto U0 kcem ⎢ ⎥ ⎝ Cpm
form has been applied to gas flux detection [15]. Tracer technology has ⎣ ⎦
−1
been also used in detection of dam leaks [16] and gas lifted wells under Δz· ⎛ Cpm Gt (k cem + Rto U0 TD ) ⎟⎞
+ ϕ+g⎟⎞ + e ⎝
⎜
2πRto U0 k ce m ·(Tfi−Thi )
an appropriate tracer selected with the purpose of small leakage point ⎠
⎠ (4)
location [17]. Other approaches such as machine learning and data-
driven models have also been used for incident prediction and detec- Combining Eqs. (2) and (3), the casing temperature at a certain
tion, such as Gaussian Process Regression model [18], and Bayesian depth can be expressed as [8]:
approaches [19,20].
Rto U0ln(R cem / R co)(Tf −Th )
However, the main weakness of the mentioned methods is their Tco = Th +
inability to predict tubing leakage location, especially in the circum- kcem (5)
stance of the tubing leakage which is considered as the main source of
SCP. These conventional tools for SCP detection inevitably interfere 2.1.2. Calculation of wellbore pressure
with normal operations, and the leak detection, and the location tech- The calculation of wellbore pressure is divided into two parts:
nique without downhole operation is therefore needed for remediation Tubing pressure and annulus pressure. According to the conservation
of the gas wells with SCP at lower costs and risks. Incorporating the law of mechanical energy, the pressure gradient of gas-liquid two-phase
effects of uncertainties on downhole tubing leakage location seems to flow in the length of tubing is expressed [10]:
be missing in the existing analytical models due to complicated re-
2
servoir information and measurement errors. Therefore, this paper dP 2fρ vm dv
= ρm g− m −ρm vm m
presents a newly tool for prediction of the downhole tubing leakage dz d dz (6)
location. A calculation model of tubing and annulus pressure profiles is where P is the flow pressure, g is the gravitational acceleration, m/s2,
established on heat transfer theories, mass conservation equation, and ρm is the average density of mixture fluid in the tubing, kg/m3; f is the
multiphase flow theories. Then the position of the tubing leakage in the friction coefficient; vm is the velocity of mixture fluid in the tubing, m3/
production tubing is determined by the pressure difference. Tubing kg; d is the inner diameter of tubing, m; and z is the well vertical depth
leakage location within a certain range is obtained by the calculation of (for a directional well with angle of inclination θ, z = z cosθ.)
dynamic confidence intervals. In addition, a probability-based model The critical parameter of gravity pressure gradient is the average
integrating Bayesian inference, and Monte Carlo simulation is devel- density of the gas-liquid flow in the tubing, which can be expressed as:
oped and applied in an offshore field gas well to predict the likelihood
1000γo ⎞ 1 1000γw ⎞
of tubing leakage location.
ρm = ⎛ ⎜ ⎛ ⎟ ⎞+⎛ ⎛ WOR ⎞ (1
⎜ ⎟

⎝ Bo ⎠ ⎝ 1 + WOR ⎠ ⎝ B w ⎠ ⎝ 1 + WOR ⎠
2. Proposed methodology P 293.15 1
−α ) ⎛ ⎞⎛ ⎞⎛ ⎞α
⎝ 0.101325 ⎠ ⎝ T ⎠ ⎝ Z ⎠ (7)
2.1. Theoretical model for wellbore pressure distribution
where α is the gas void fraction which is continually changing as the
fluid flows from tubing to the ground [21]. γo is oil relative density, γw
2.1.1. Calculation of wellbore temperature
is water relative density, WOR is the water-oil ratio, Bo is oil volume
As the gas pressure in the tubing decreases during the middle and
factor, Bw is water volume factor, T is tubing temperature, and Z is
later stages of production, the condensate oil and water, as well as
compressibility coefficient of natural gas.
formation water which is precipitated will gradually gather into the
The mixture velocity can be obtained by:
well and be carried to the offshore platform by gas [21]. Therefore, the
flow in the tubing is considered to be the gas-liquid two-phase flow. (
Ql GLR−Rs 1 + WOR
1
)
0.101325 T
Assuming that the heat transfer from high temperature fluid in the vm = ⎛ ⎞⎛ ⎞Z
86400A ⎝ P ⎠ ⎝ 293.15 ⎠
tubing to outer edge of wellbore cement sheath is in a steady state,
Ql ⎛ 1 WOR ⎞
while that from outer edge of wellbore cement sheath to the formation + Bo + Bw
is in an unsteady state [8]. The heat gradient transferred from high 86400A ⎝ 1 + WOR 1 + WOR ⎠ (8)
temperature fluid to cement sheath surface, that from cement sheath where GLR is gas liquid ratio, m /m . Rs is solubility of gas in oil, m3/
3 3

surface to formation, and that from casing to cement sheath surface can m3. A is tubing area, m2.
be expressed respectively as: The tubing pressure at the different well segment with depth Δz will
dQ1 2πRto U0 be evaluated as
= (Tf −Tcem)
dz Gt (1) fQl Mt2 2
⎡
ΔP = 10−6 ⎢Δz ⎛⎜ρm g + ⎞ ⎛ vm ⎟⎞ ⎤
⎟ + ρm Δ ⎜
9 5
9.21 × 10 ρm d ⎠ ⎥
dQ2 2πkcem ⎣ ⎝ ⎝ 2 ⎠⎦ (9)
= (Tcem−Th)
dz Gt TD (2)
where Ql is total liquid production rate, m /d, and Mt total mass of fluid
3

dQ3 2πkcem associated with 1 m3 of liquid, kg/m3.

= (Tco−Tcem) Similar to the tubing, the annulus pressure profile in the axis di-
dz ln(R ce / R co) (3)
rection can be calculated on the basis of the energy equation. Since the
where Tf , Tcem , and Th are the temperature of wellbore fluid, cement, and gas in the annulus is relatively static, the friction pressure gradient and
formation respectively, K. Gt is total mass flow rate, kg/s. TD is the time acceleration pressure gradient are assumed to zero. The annular pres-
function of transient heat transfer. kcem is the thermal conductivities of sure distribution can be divided two parts: The pressure in tubing-
cement sheath. Rto , R co and R ce represent outside diameter of tubing, production casing annulus above liquid level and that below liquid level
casing and cement sheath respectively. U0 denotes the overall heat considering the hydrostatic pressure of the liquid column.

547
S. Wu et al. Measurement 127 (2018) 546–553

By integrating Eq. (6) from the bottom to the top of the well seg- 1 (ri−μ−3σ )2 ⎤
f (ri |μ + 3σ , σ ) = exp ⎡−
ment with depth Δz, the annulus pressure above liquid level is calcu- 2πσ 2 ⎢
⎣ 2σ 2 ⎥
⎦ (18)
lated as:
The probability prediction of downhole tubing leakage under the
Mg gg Δz
Pc o = Pci e ZRT (10) symptom parameter ri is given by:
P (ri |A) P (A) P (A)
where Pco and Pci represent the outlet and inlet pressure of well seg- P (A |ri ) = =
∑ P (ri |A) P (A) (1−P (A))·exp(4.5−3(ri−μ)/ σ ) + P (A)
ment, respectively. Mg is gas relative molecular weight.
The annulus pressure below liquid level is given by: (19)

Pc = Pct + ρg (z−z t ) (11)

2.3. Prediction analysis process
where Pct is the pressure of gas column at liquid level, ρ is the liquid
density, kg/m3, z is the well vertical depth, m, zt is the depth of liquid Tubing corrosion penetration, thread connection failure, packer seal
level, m (for a directional well with angle of inclination θ, z = z cosθ). malfunction and gas lift valve leakage may lead to the high-pressure gas
leaking into the tubing-production casing annulus and then accumulate
2.2. Bayesian inference-based prediction analysis with uncertainties at wellhead, causing a serious SCP problem. Wellhead annulus pressure
increases with the accumulation of gas until the pressures difference of
Bayesian inference [24], an artificial intelligence approach, has the tubing and the production casing annulus is almost 0 at a certain
been applied to develop condition monitoring solutions that can deal point, according to the pressure equalization principle. Such a point
with uncertainty of faults and fault symptoms. It can combine an ex- will be located as a leakage point after leakage. If there is no SCP, such a
pert’s belief with evidence obtained from field measurements for point will be defined as the initial pressure point. Three steps are used
probability prediction [25]. The joint distribution of the data given that to illustrate the prediction analysis process of tubing leakage location in
the distribution of each observation Yi depends on a covariate mea- the following.
surement ri through some unknown function f and parameter θ is ex-
pressed Step 1: Calculation of tubing leakage location in the steady pro-
n duction state
P (Y |r1, ⋯, rn, θ) = P (Y |f (r1), ⋯, f (rn ), θ) = ∏ P (Yi |f (ri ), θ)
i=1 (12) An iterative model [10] is introduced to calculate the pressure dif-
A Gaussian process [18] is a collection of random variables, any ference distribution using computational mesh due to the non-linear
finite number of which have a joint Gaussian distribution. Intuitively, it distribution of wellbore temperature and pressure along the well depth.
can be considered as a generalization of the Gaussian probability dis- It divides the well into several segments (Δz ) along the depth direction,
tribution over functions, and can be described as: in which the pressure and temperature are considered as constant. The
pressure (Pc1, Pt1) and temperature (Tc1, Tt1) of the first segment are equal
f (r ) ∼ GP (u (r ), k (r , r ′)) (13) to those of the wellhead. The values of annulus pressure Pci∗ and tubing
where the mean function μ(r) represents the expected value of the pressure Pt∗i are then assumed to the cell i. Based on thermophysical
function f(r) at the input point r. The covariance function k(r, r′) can be parameters of the fluid, the annulus pressure Pci and tubing pressure Pti
considered as a measure of the confidence level for μ(r). For a one- are obtained by calculating the annulus temperature Tci and tubing
dimensional Gaussian process, the probability density can be de- temperature Tfi by using Eqs. (4), (5), (9) and (11). Check if |Pci∗−Pci | and
termined by two parameters: Mean μ and variance σ. |Pti∗−Pti | is less than a certain value. If not, assume new values to annulus
pressure Pci∗ and tubing pressure Pt∗i . If so, calculate ΔP = |Pci−Pti |. When
1 (r −μ)2 ⎤ it meets the requirement ΔP ≈ 0 , the depth of leakage location zl will be
f (r |μ, σ ) = exp ⎡−
2πσ 2 ⎢
⎣ 2σ 2 ⎥⎦ (14) acquired, and it is equal to z i = Δz·i . Otherwise, go to next cells. In
addition, the probability distribution of pressure difference ΔP can be
The likelihood function is numerically equal to:
obtained based on Eq. (19) to check the possibility of tubing leakage
n
location.
L (θ |r1, ⋯, rn ) = ∏ f (ri |θ)
i=1 (15)
Step 2: Calculation of dynamic confidence intervals (DCI) for pre-
Two kinds of states are used to describe the situation of wellbore dicting tubing leakage location
pressure difference between tubing and annulus under leakage occur-
ring: Pressure difference is more than 0 (denoted as N, meaning that The confidence interval [26] is intended to indicate a clear interval
there is no leakage) and that is almost 0 (denoted as A, meaning that of estimates as well as to better assess the leakage location of the gas
there is a leakage point). The probability agrees that P(N) = 1 − P(A). wells with given the confidence level that the true value will lie in this
Therefore, the probability in state A under the given parameters ri is interval. It is assumed that the true value of p could be estimated by
expressed as: pavge. A DCI can be obtained through the normal approximation of p
P (ri, A) P (ri |A) P (A) with the mean pavge and sample variance S2(pavge) with dynamic data
P (A |ri ) = = window n [27]. It should be calculated to improve the confidence in the
P (ri ) ∫ P (ri |A) P (A) dA (16)
assessment results. The approximation of the confidence intervals under
Assume that the state N and A of the tubing leakage events given a given confidence level 100(1 − α)% is generated by [28]:
parameter ri are following Gaussian rules. According to Bayesian in- 100(1 − α)% confidence lower bound
ference, if the event is in the state N and P (ri |N ) is assumed to satisfy
S 2 (pavge )
N (ri |μ, σ ), the likelihood estimate is expressed as: pL ≥ pavge −tα /2 (n−1)
n (20)
1 (r −μ)2
f (ri |μ, σ ) = exp ⎡− i 2 ⎤ 100(1 − α)% confidence lower bound
2πσ 2 ⎢
⎣ 2σ ⎦ ⎥ (17)
S 2 (pavge )
If the event is in state A and P (ri |A) is assumed to satisfy pU ≤ pavge + tα /2 (n−1)
N (ri |μ + 3σ , σ ), the likelihood estimate is expressed as: n (21)

548
S. Wu et al. Measurement 127 (2018) 546–553

where tα/2 (n−1) indicates the 100(1 − α)th percentile of the t-dis- circuits can test the depth of annulus liquid level, pressure and tem-
tribution. perature of annulus, and gas phase composition in annulus. The test
ideally contains three stages (bleed down, stabilized flow, and buildup)
Step 3: Calculation of probability distribution of different tubing which are essential for the unambiguous analysis of test results. In the
leakage location bleed down stage, the pressure bleeds to zero. In some cases, the
pressure does not bleed to zero, and the valve should stay open until the
The Monte Carlo simulation (MCS) is adopted in this paper to il- stabilized flow is reached. At the end of this stabilized flow stage, there
lustrate the probability distribution of different tubing leakage location. is a balance between inflow and outflow of gas in the annulus. In
MCS is a suitable and valuable approach which is widely applied to buildup stage, shut in the annulus and measure the pressure evolution
reliability analysis for real-world system in solving complex en- with time [32]. As the engineering analysis of SCP is a data driven
gineering problems to estimate the uncertainty for the measurand [29]. process, obtaining more data related to the casing pressure is the ob-
Such simulation methods can also provide a probabilistic approxima- jective of diagnostic test.
tion to the solution of a mathematical equation in consideration of any A real offshore gas production well with SCP in tubing-production
type of probability density function for the assessed random variables casing annulus is selected for analysis. The testing is performed ac-
[30]. MCS treats the problem as a series of real experiments, and the cording to the main test procedures, and the results are shown in the
necessary indicators such as failure probability can be estimated by following.
counting the number of times on an event that occurs in whole simu-
lated process. • The stable casing pressure at wellhead is measured before pressure
Generally, an algebraic relationship between the measurand Z and relief;
the parameters or individual factors ri that affect the measuring process • The depths of liquid surface before and after pressure relief are
through a model function, F, can be expressed as Z = F (r1, r2, ⋯, rn ) ; measured by acoustic detector. It is found that the two liquid level
Considering the available knowledge for the input random variables, values appear to very close and are considerably lower than the
their probability density functions will be assigned f(r), being assumed initial value after completion;
to follow Guassian (normal) distribution, rectangular (uniform), • Bleed-down/build-up test is conducted, and production casing
Triangular, etc. The probability distribution of a certain target area D pressures, versus time, is recorded. The casing pressure does not
for Z can be given by [31]: bleed to zero, but to a stable pressure through the 0.5-in. needle
valve over a 4-h period. By closing the valve, the pressure builds
P (Z ) ∝ ∫ F (r ) f (r ) dD (22) back up to initial pressure;
In fact, P(Z) is the propagation of n distributions by means of MCS. • Pressure of intermediate casing over time is recorded to keep close
to zero during the test;
In order to predict the probability distribution of different tubing
leakage location, the relationship between wellhead pressure difference • Natural gas with components similar to the production formation
gas is detected in the production casing annulus.
and leakage location in production gas wells is analyzed, and the
probability is obtained under MSC as follows:
The detailed parameters of the well, some of which are collected by
N the test system and from wellbore structures, are presented in Table 1.
P = SL
N (23) Based on the proposed method, the pressure difference profile be-
where N is a total number of samples, and NSL is a simulated number of tween the annulus and tubing and the probability distribution of tubing
leakage events. leakage location under the simplified well structure are built as shown
in Fig. 2. It is observed from Fig. 2(b) that the absolute value of the
pressure difference is close to zero at the well depth of 393 m if the
3. Case study
production gas well is in a stable state after leakage. In reality, this
location corresponds to the tubing, representing SCP caused by the
To predict leakage location, it is necessary to develop a ground test
tubing components failure, otherwise the leakage point would be at
system to monitor well parameters. The schematic of the test system is
another place. For instance, when the pressure of production casing
shown in Fig. 1. This system includes two main modules: Acoustic de-
annulus is almost zero at the wellhead after well completion, the
tector module of liquid level and annulus fluid characteristics mea-
pressure difference profile under no leakage is built as shown by red
surement module. Three gas circuits for testing are designed in the
line in Fig. 2(b). The value of the pressure difference is zero with the
system to test whether the annuluses are interconnected. All three gas

Fig. 1. Schematic of testing system.

549
S. Wu et al. Measurement 127 (2018) 546–553

Table 1
Detailed information of the illustration well.
Parameters Value Parameters Value

Diameter of tubing (mm) Φ73 × 5.51 Setting depth of packer (m) 2878.4
Diameter of casing (mm) Φ244.5 × 13.58 Depth of liquid level (m) 303.7
Formation temperature (°C) 127.23 Gas production rate (m3/d) 73,000
Geothermal gradient (°C/100 m) 3.67 Oil production rate (m3/d) 10
Tubing temperature(°C) 45.8 Water production rate (m3/d) 0.9
Tubing pressure (MPa) 15.3 Density of annular fluid (g/cm3) 1.05
Stable casing pressure (MPa) 14.6 Casing depth (m) 3244.2
Wellhead temperature(°C) 33.6

well depth of 1775 m, denoted as the initial pressure point. Fig. 2(c) remained constant, which equals the annulus pressure at this point. The
suggests that, in a stable production state, the probability at the well wellhead annulus pressure accordingly changes, and the corresponding
depth of 393 m is almost close to 1 while that of other positions is far value is calculated by ΔPL-ΔPC. It can be explained in detailed by Eqs.
less than one. The result obtained from Bayesian probability estimation (24)–(26). Abnormal increase of wellhead annulus pressure indicates
indicates that there is a great chance of leakage occurring in this lo- that there is another leakage location in the wellbore which needs to be
cation. It's important to note that the leakage location can be predicated paid more attention.
by the pressure difference distribution after detecting the increase of The pressure of leakage point before liquid level decreasing is given
wellhead annulus pressure. By pulling out the production string, it is by
confirmed that the tubing in this position is failed, indicating the pre-
PCW + PC + PL = PCL (24)
dicted leakage position is very close to the real one.
Actually, a well with SCP needs to be released regularly, so the The pressure of leakage point after liquid level decreasing is given
annular liquid level falls as a result of a series of pressure bleeding- by
down operations. Fig. 3 describes the effects of the annular liquid level
(PCW + ΔPCW ) + (PC + ΔPC ) + (PL−ΔPL) = PCL (25)
falling ΔH on the changes of wellhead annulus pressure. This reduced
column of liquid makes the pressure of gas and liquid column in the The increasing value of wellhead annulus pressure is expressed by
annulus increase ΔPC and decrease ΔPL respectively. This is because
ΔPCW = cosθρl g ΔH −ΔPC (26)
that the gas leaking from tubing to the annulus fills the space freed by
the dropping liquid. The tubing pressure at the leakage point has where PCL, PCW, PC, PL are the annulus pressure of the leakage point, the

Fig. 2. Prediction of leakage location given a specified well with (a) simplified annulus structure, (b) pressure difference, and (c) probability distribution.

550
S. Wu et al. Measurement 127 (2018) 546–553

Fig. 3. Change in wellhead annulus pressure.

annulus pressure at wellhead, the pressure of gas column, and the are predictable. Fig. 5 shows leakage location prediction analysis con-
pressure of liquid column before liquid level decreasing, respectively. sidering uncertainty. The wellhead pressure difference between tubing
ΔPCW, ΔPC, ΔPL are the increasing annulus pressure at wellhead, the and casing with different annulus pressure is presented in Fig. 5(a). It is
increasing gas pressure, and the decreasing liquid pressure in the an- seen from Fig. 5(b) that the depth of leakage location decreases with the
nulus after liquid level decreasing, respectively. ρl is the density of li- increase of wellhead pressure difference. Fig. 5(c) indicates that the
quid in the annulus, ΔH is the height of liquid column decrease, and θ is probability of tubing leakage occurring under wellhead pressure dif-
the angle between horizontal line and blue dot line. ference given the number of simulation at N = 1 × 104. It is found that
the uncertainty interval under different predicted probability can be
determined. Taking the approximate probability of 0.03 for instance,
4. Uncertainty analysis the result of wellhead pressure difference is among the area of
[−0.58 MPa, −1.36 MPa], and the corresponding leakage position is
Even though tubing and annulus pressure, temperature and liquid located in [381 m, 454 m]. The prediction probability using MCS is able
level are taken into account by the proposed prediction models and to generate reasonable results with uncertainty. Note that results from
make it adequate for leakage location and probability estimation, there both models agree well in the expanded uncertainty, mainly because of
are still considerable unexplained variables, such as the ignorance of the more linear model and the assumption of a normal distribution.
changes in formation and downhole fluid, as well as measurement er-
rors. Such ignorance will further result in uncertainties to affect deci- 5. Conclusions
sion-making. 95% DCI and MCS are therefore applied to illustrate the
uncertainties in leakage location. (1) A newly method is illustrated to predict downhole tubing leakage
Fig. 4 indicates DCI under two kinds calculation modes: Calculated location for a gas well with sustained casing pressure. Based on the
with pressure difference between tubing and annulus directly, as shown pressure difference between tubing and annulus with approxi-
in Fig. 4(a), and obtained with tubing pressure and annulus pressure mately zero, tubing leakage location is determined when the well is
respectively, as shown in Fig. 4(b). It is noted that 95% confidence in a steady state. The leakage point is identified next to tubing,
interval of tubing leakage location is [368 m, 418 m] using the previous which is confirmed by probability analysis and filed operation of
algorithm, while there are four confidence intervals [358 m, 418 m], pulling out the tubing. Thus, the validity of the proposed model is
[368 m, 428 m], [358 m, 428 m], and [368 m, 418 m] using the latter examined.
one. As can be observed, in this case, results from both algorithms have (2) Abnormal increase of wellhead annulus pressure indicates that
one same interval and three expanded uncertainties with 50 m, 60 m there is another leakage location in the wellbore which needs to be
and 70 m, mainly because of the latter algorithm generating two upper paid more attention. In addition, dynamic confidence intervals
limits and two lower limits. Compared to the analytic formula results, under two kinds calculation modes are used to predicate the
DCI can generate leakage location within a certain range. leakage location within a certain range with different expanded
An MCS model as a stochastic model is analyzed statistically and uncertainties. 95% confidence intervals of downhole tubing leakage
performed by program codes written by using Matlab software to de- location are less by estimating direct pressure difference algorithm
termine a likely outcome. As a starting point, wellhead pressure dif- than by indirect one.
ference and leakage location are specified as input entered to the (3) Downhole tubing leakage location prediction is analyzed under
model, and the probability density function is assumed following different wellhead pressure difference. The results indicate the
normal distribution. The corresponding range and possibility of values

551
S. Wu et al. Measurement 127 (2018) 546–553

Fig. 4. 95% confidence intervals of tubing leakage location given (a) pressure difference, and (b) tubing and annulus pressure.

Fig. 5. Leakage location prediction with uncertainty (a) leakage location with different wellhead annulus pressure, (b) relationship of leakage location and wellhead
pressure difference, (c) probability given wellhead pressure difference.

552
S. Wu et al. Measurement 127 (2018) 546–553

depth of leakage location decreases with the increase of wellhead estimation of CO2 wellbore leakage, Int. J. Greenhouse Gas Control. 67 (2017)
pressure difference. The Monte Carlo simulation is performed to 20–30.
[14] M. Shakiba, S.A. Hosseini, Detection and characterization of CO2 leakage by multi-
check that the prediction probability distribution of tubing leakage well pulse testing and diffusivity tomography maps, Int. J. Greenhouse Gas Control.
location is able to generate reasonable results with uncertainty in- 54 (2016) 15–28.
tervals. [15] X. Wang, Z. Guan, Y. Xu, Y. Tian, Signal analysis of acoustic gas influx detection
method at the bottom of marine riser in deepwater drilling, J. Process Control. 66
(2018).
Acknowledgment [16] D. Battaglia, F. Birindelli, M. Rinaldi, E. Vettraino, Fluorescent tracer tests for de-
tection of dam leakages: The case of the Bumbuna dam-Sierra Leone, Eng. Geol. 205
(2016) 30–39.
This work is supported by the National Key Research and [17] F.Y. Shnaib, M.S. Nadar, M.P. Sreekumar, K. Ponnuvel, L.M. Peacock, Successful
Development Plan of China (SQ2017YFSF060124, Application of CO2 Tracer Technology for Surveillance of Gas Lifted Wells, SPE
CCL2017RCPS0182RCN). The authors are grateful for the Editor and Production and Operations Conference and Exhibition, Society of Petroleum
Engineers, 2010.
reviewers’ helpful comments.
[18] Y. Liu, Y. Song, J. Keller, P. Bond, G. Jiang, Prediction of concrete corrosion in
sewers with hybrid Gaussian processes regression model, RSC Adv. 7 (2017)
References 30894–30903.
[19] Z. Wang, M.J. Small, A Bayesian approach to CO 2 leakage detection at saline se-
questration sites using pressure measurements, Int. J. Greenhouse Gas Control. 20
[1] Xu R. Analysis of diagnostic testing of sustained casing pressure in wells [Ph.D. (2014) 188–196.
Thesis]: Louisiana State University; 2002. [20] B. Cai, M. Xie, Y. Liu, Y. Liu, Q. Feng, Availability-based engineering resilience
[2] T. Rocha-Valadez, R.A. Mentzer, A.R. Hasan, M.S. Mannan, Inherently safer sus- metric and its corresponding evaluation methodology, Reliab. Eng. Syst. Safety 172
tained casing pressure testing for well integrity evaluation, J. Loss Prev. Proc. 29 (2018) 216–224.
(2014) 209–215. [21] A.R. Hasan, C.S. Kabir, Wellbore heat-transfer modeling and applications, J. Pet.
[3] Y. Jin, T. Haixiong, L. Zhengli, Y. Liping, H. Xiaolong, Y. De, et al., Prediction model Sci. Eng. 86 (2012) 127–136.
of casing annulus pressure for deepwater, Petro. Explor. Develop. 40 (2013) [22] H.J. Ramy, Wellbore Heat Transmission, J. Pet. Technol. 14 (1962) 427–435.
661–664. [23] X. Zhang, J. Fan, S. Wu, D. Liu, A novel acoustic liquid level determination method
[4] L.F. Eaton, W.R. Reinhardt, J.S. Bennett, Liner Hanger Trapped Annulus Pressure for coal seam gas wells based on autocorrelation analysis, Energies 10 (2017) 1961.
Issues at the Magnolia Deepwater Development, SPE/IADC Drilling Conference, [24] B. Cai, L. Huang, M. Xie, Bayesian networks in fault diagnosis, IEEE Trans. Indust.
Miami, 2006. Inform. (2017).
[5] API. Annular casing pressure management for offshore wells. Recommended Practic [25] B. Cai, H. Liu, M. Xie, A real-time fault diagnosis methodology of complex systems
90: API; 2006. using object-oriented Bayesian networks, Mech. Syst. Sign. Proc. 80 (2016) 31–44.
[6] Z. Zhang, H. Wang, Sealed annulus thermal expansion pressure mechanical calcu- [26] D. Grabaskas, M.K. Nakayama, R. Denning, T. Aldemir, Advantages of variance
lation method and application among multiple packers in HPHT gas wells, J. Nat. reduction techniques in establishing confidence intervals for quantiles, Reliab. Eng.
Gas Sci. Eng. 31 (2016) 692–702. Syst. Safety 149 (2016) 187–203.
[7] P. Oudeman, L.J. Bacarreza, Field trial results of annular pressure behavior in a [27] P. Liptai, B. Dolník, M. Pavlík, J. Zbojovský, M. Špes, Check measurements of
high-pressure/high-temperature well, SPE Drill. Completion (1995). magnetic flux density: Equipment design and the determination of the confidence
[8] J. Yang, H. Tang, Z. Liu, L. Yang, X. Huang, D. Yan, et al., Prediction model of interval for EFA 300 measuring devices, Measurement 111 (2017) 51–59.
casing annulus pressure for deepwater well drilling and completion operation, Pet. [28] S. Lee, M. Bolic, V.Z. Groza, H.R. Dajani, S. Rajan, Confidence interval estimation
Explor. Dev. 40 (2013) 661–664. for oscillometric blood pressure measurements using bootstrap approaches, IEEE
[9] J. Liu, H. Fan, Q. Peng, S. Deng, B. Kang, W. Ren, Research on the prediction model Trans. Instrument. Measure. 60 (2011) 3405–3415.
of annular pressure buildup in subsea wells, J. Nat. Gas Sci. Eng. 27 (2015) [29] E. Ferrario, N. Pedroni, E. Zio, Evaluation of the robustness of critical infra-
1677–1683. structures by Hierarchical Graph representation, clustering and Monte Carlo si-
[10] S. Wu, L. Zhang, J. Fan, X. Zhang, D. Liu, D. Wang, A leakage diagnosis testing mulation, Reliab. Eng. Syst. Safety (2016).
model for gas wells with sustained casing pressure from offshore platform, J. Nat. [30] K. Shahanaghi, P. Nakhjiri, A new optimized uncertainty evaluation applied to the
Gas Sci. Eng. 55 (2018) 276–287. Monte-Carlo simulation in platinum resistance thermometer calibration,
[11] J.E. Johns, D.N. Cary, J.C. Dethlefs, B.C. Ellis, M.L. McConnell, G.L. Schwartz, Measurement 43 (2010) 901–911.
Locating and Repairing Casing Leaks with Tubing in Place-Ultrasonic Logging and [31] M.Á. Herrador, A.G. González, Evaluation of measurement uncertainty in analytical
Pressure-Activated Sealant Methods, Offshore Europe Society of Petroleum assays by means of Monte-Carlo simulation, Talanta 64 (2004) 415–422.
Engineers, 2007. [32] M. Kazemi, A.K. Wojtanowicz, Development of Improved Testing Procedure for
[12] F. Zeng, G. Zhao, L. Zhu, Detecting CO 2 leakage in vertical wellbore through Wells with Sustained Casing Pressure, SPE Annual Technical Conference and
temperature logging, Fuel. 94 (2012) 374–385. Exhibition, Amsterdam, The Netherlands, 2014.
[13] Y. Mao, M. Zeidouni, I. Duncan, Temperature analysis for early detection and rate

553