J Fail. Anal. and Preven.

(2015) 15:873882
DOI 10.1007/s11668-015-0033-6

TECHNICAL ARTICLEPEER-REVIEWED

Probability of Detection and False Detection for Subsea Leak

Detection Systems: Model and Analysis
Alireda Aljaroudi . Faisal Khan . Ayhan Akinturk . Mahmoud Haddara .
Premkumar Thodi

Submitted: 22 September 2015 / Published online: 27 October 2015

ASM International 2015

Abstract Ensuring the integrity of subsea process components is one of the primary business objectives of the oil
and gas industry. Leak detection system (LDS) is one type
of system used to safeguard reliability of a pipeline. Different types of LDS use different technologies for detecting
and locating leaks in pipelines. One technology, which is
gaining wide acceptance by the industry, is the fiber opticbased LDS. This technology has great potential for subsea
pipeline applications. It is the most suited for underwater
applications due to the ease of installation and reliable
sensing capabilities. Having pipelines underwater in the
deep sea presents a great challenge and a potential threat to
the environment and operation. Thus, there is a need to
have a reliable and effective system to provide the assurances that the monitored subsea pipeline is safe and
functioning as per operating conditions. Two important
performance parameters that are of concern to operators are
the probability of detection and probability of false alarm.
This paper presents a probabilistic formulation of the

A. Aljaroudi (&)  F. Khan  M. Haddara

Memorial University, St. Johns, NL, Canada
e-mail: aaa515@mun.ca
F. Khan
e-mail: fikhan@mun.ca
M. Haddara
e-mail: mhaddara@mun.ca
A. Akinturk
National Research Council, St. Johns, NL, Canada
e-mail: ayhan.Akinturk@nrc-cnrc.gc.ca
P. Thodi
INTECSEA, WorleyParsons Group, St. Johns, NL, Canada
e-mail: premkumar.Thodi@intecsea.com

probability of detection and probability of false detection

for a fiber optic-based LDS.
Keywords Probability of detection (PD) 
Probability of false alarm (PFA) 
Leak detection system (LDS)  Oil and gas pipeline
Nomenclature
BSS
Brillouin-Stimulated scattering
CW
Continuous wave
LDS
Leak detection system
NP
Noise Power
PD
Probability of detection
PFA
Probability of false alarm
PMD
Probability of missed detection
SNR
Signal-to-noise ratio
c
Speed of light (Km/s)
d
Location of the temperature change
dP=
Temperature coefficient (mW/C)
dT
dP=
Strain coefficient (mW/le)
de
n
Refractive index
Aeff
Effective area of the fiber
Leff
Effective length of the fiber
gB
Gain
Po
Reference power
PB(measured) Measured Brillouin power
PCW
Input probe power
PP
Pulse power
ae
Strain coefficient expressed in MHz/le
aT
Temperature coefficient expressed in MHz/
C
De
Strain change
DT
Temperature change
DTmeasured
Measured temperature change

123

874

Dt
Va
vB
s
k
vo
Dt
Xth

J Fail. Anal. and Preven. (2015) 15:873882

Minimum detectable temperature change

Acoustic velocity
Brillouin frequency shift
Pulse width
Wavelength of the incident light
Reference Brillouin frequency at no strain
and at the ambient temperature (MHz)
Traveled time
Temperature change threshold

False detection results in excessive expenditure and

unnecessary mobilization of equipment and personnel to
the site that is thought to be leaking. On the other hand,
missed detection results in environmental and financial
liabilities and unfavorable impacts on reputation.
An overview of a fiber optic LDS and Brillouin-stimulated scattering (BSS) as a technique for temperature
sensing are provided in Fiber Optic-Based LDS section. Probability of False Alarm (PFA) and Probability
of Detection and Missed Detection sections, respectively,
discuss PFA and PD. Concluding remarks are provided in
Conclusion section.

Introduction
One of the key monitoring systems for subsea pipelines is
the LDS. Its performance should be assessed regularly to
ensure that its operability, functionality, and reliability are
maintained at all times and, more importantly, to ensure
that it does not miss to detect or falsely detect a pipeline
leakage. The consequences of such an incorrect diagnosis
may pose a threat to the environment and production.
Generally, the key factors that affect the performance of
a subsea LDS are missed detection and false detection. The
system may declare that a leak is happening somewhere
along the pipeline when in fact a leak is not present.
Likewise, the system may not declare that a leak is happening when in fact it is present. The latter scenario is
termed a missed detection, and the former is termed a false
detection or false alarm. Missed or false detection may not
completely place the system out of service; however, they
cause the system to fail partially. In either case, whether we
have total or partial failure, the performance of the system
will be in jeopardy. Once we know these failures and are
able to calculate the probability of their occurrences and
their consequences, we can evaluate the risk and its impact
on the environment and production.
Irrespective of which leak detection method is used, the
characteristics of the received signal are the results or
indications that determine the status of the pipeline. All leak
detection methods or systems have one common task: to
detect and declare if a leak has or has not occurred and,
based on the characteristics of the received signal, determine the quantity and location of the hydrocarbon leakages.
There is no established method for evaluating the
probability of detection (PD) and probability of false alarm
(PFA) for a fiber optic-based LDS. Hence, the main
objective of this paper is to formulate the PD and probability of false detection, alternatively called PFA, for a fiber
optic distributed sensing technique used for leak detection.
This is accomplished by adopting some concepts from
signal detection theory (SDT) and engineering probabilistic
methods.

123

Fiber Optic-Based LDS

One of the most promising condition-monitoring technologies is fiber optic distributed sensing, which can
perform ongoing sensing along the entire length of the
monitored structure. The fiber optic components act as a
sensor, providing sensing and prior warning capabilities in
real time and on a continuous basis.
Using distributed fiber optic sensing technology, the
vibration, strain, and temperature changes along the monitored object can be detected. Strain occurring on a pipeline
may give indication of the existence of cracks, and
detecting it in advance will enable maintenance personnel
to perform corrective actions in a timely manner. Applying
this technology will prevent structural failure that could
lead to a leak and eventually to an oil spill.
Generally, oil is transported through pipes at a temperature that is higher than its surroundings. In the event a leak
happens, the temperature of the surroundings will increase,
causing a portion of the light wave to scatter back to the
source, indicating the occurrence of a leak [1]. On the other
hand, when a gas pipeline starts leaking, the released gas
will cool down the surrounding area resulting in a cooler
temperature than the normal temperature. As a result, the
sensing cable will trigger an alarm indicating a leak [1].
This technology can provide accurate information in real
time about the status of the monitored structure, which can
significantly enhance the decision making about what
mitigation actions should be considered in the event a risk
or safety issue becomes imminent.
Scattering
When optical laser light propagates through the fiber, it
gets scattered in three different spectral forms as indicated
in Fig. 1, with different frequencies and intensities [2]. The
scattering is created due to impurities or changes of

J Fail. Anal. and Preven. (2015) 15:873882

875

Scaering
Brillouin

Rayleigh

Fig. 2 Illustration of a simplified BOTDA system

Raman

vB
Fig. 1 Scattering mechanisms

composition and interaction of the laser light with molecules of the fiber.
The Raman-based technique can achieve sensing up to
37 km, with a measurement time of \3 min and a temperature accuracy of 3 C, and can measure temperature
changes only [3, 5].
The Brillouin scattering-based technique has the ability
to sense temperature and strain changes along the fiber
optic cable. The wavelength of the reflected wave is closely
related to the changes of the surrounding temperature and
strain of the fiber optic cable [4]. This technique can
achieve less than 1 m spatial resolution, 1 min measuring
time, 2 C temperature resolution, and up to 50 km sensing
range [5]. The range can be extended using fiber optic
amplifiers in between. The reported strain accuracy is
approximately 10 lm strains [6]. As stated above, the
Brillouin scattering-based technique outperforms the
Raman-based technique, as it can achieve longer sensing
range, improved accuracy, and less measuring time and can
measure both temperature and strain. Therefore, the focus
of this research will be on the Brillouin-based sensing.
Brillouin Scattering
Brillouin scattering is caused by the fluctuations of the
refractive index of the fiber. These fluctuations take place
due to the variations of fiber composition, pressure, temperature, or density [7]. Along the way, a vibration of the
fiber molecules takes place traveling at the acoustic speed
which causes variations in density and as a result the
refractive index changes, causing the light to scatter; this
vibration is referred to as phonons.
The process is called inelastic because a transfer of
energy between the incident light, photons, and the molecules of the fiber takes place. If the energy is transferred
from the photons to the fiber material, then the backscattered light is downshifted in frequency. In this case,
photons lose energy. Conversely, if energy is transferred
from the fiber material (the silica glass) to the photons, then
the backscattered light is upshifted in frequency. Here,
photons gain energy and the frequency becomes higher.
The shift in frequency is called the Brillouin frequency
shift and is given by [7]

2nVa
;
k

Eq 1

where vB is the Brillouin frequency shift; Va is the acoustic

velocity of the phonons; n is the refractive index of the
fiber, and k is the wavelength of the incident light.
Distributed Brillouin Sensing Techniques
Several techniques are used for distributed Brillouin sensing. They include Brillouin optical time domain
reflectometry (BOTDR), Brillouin optical time domain
analysis (BOTDA), Brillouin optical frequency domain
analysis (BOFDA), Brillouin Optical correlation domain
analysis (BOCDA), and Brillouin Echo distributed sensing
(BEDS) [8]. The focus of this paper will be on the BOTDA
technique because it is one of the most commonly used as a
monitoring technique by the industry.
Brillouin Optical Time Domain Analysis (BOTDA)
BOTDA works by launching lasers in two opposite directions; one is pulsed and the other one is continuous. The
frequency difference of the two lasers can be used to
measure strain and temperature along the fiber [9].
v1 is the pulsed laser or the pump signal, and v2 is the
continuous wave (CW) and is referred to as the probe
signal.
In this configuration, the power of the probe signal is
transferred to the pump pulse, resulting in an increase in
the intensity of the pulse as it travels along the fiber
yielding a better signal-to-noise ratio (SNR) and hence
longer sensing range [1012]. This configuration is referred
to as Brillouin-stimulated scattering (BSS) (Fig. 2).
The Brillouin frequency shift is dependent on material
temperature and strain. The Brillouin scattering may lose
or gain energy; the energy loss is called a Stokes process
and energy gain is called an anti-Stokes process. To
enhance the interaction between the incident light (the
pump signal) and the Stokes, a probe laser is launched at
the opposite side of the fiber [13, 14]. For the first laser, a
square pulse is used, as this is usually used for timing
control, because the square pulses are of equal duration.
Every time a pulse is sent, it will have the same time
duration. From the pulse width, the spatial resolution can
be determined and this is defined in Eq 2. The spatial
resolution is the smallest length of the monitored object
whose temperature change can be determined.

123

876

J Fail. Anal. and Preven. (2015) 15:873882

Frequency Shift - Temperature Change Plot

Fig. 3 Configuration for temperature and strain detection

cs
Dz ;
2n

Eq 2

where c is the speed of light in vacuum, n is the fiber optic

cable refractive index, and s is the pulse width.
There exists a linear relationship between the frequency
shift and the changes of temperature and strain [6, 15, 17];
the frequency shift can be expressed as follows:
vB  vo ae De aT DT;

Eq 3

where vo is the reference Brillouin frequency at no strain

and at ambient temperature, expressed in MHz; ae is the
strain coefficient expressed in MHz/l; aT is the temperature coefficient expressed in MHz/C; DT is the
temperature change, which is the difference between the
measured temperature and the ambient or reference temperature, and De is the strain change.
The strain measurement is referred to as micro-strain
(le). If a fiber optic cable has an original length of 1 m and
because stress is stretched to 1.000007 m, then the strain
becomes seven micro-strains (7 lm).
From the frequency shift, sensing can be determined, but
the challenge here is that the shift is dependent on both
strain and temperature. From the frequency shift, it is
impossible to determine which change has occurred.
Temperature is the determining factor for the presence or
non-presence of a leak.
To address this challenge, the fiber optic cable is held
loose in close proximity to the pipe, as shown in Fig. 3.
The loose fiber is used to monitor the temperature
change only, assuming zero strain, since the fiber is held
loose and not attached to the pipeline. The Brillouin frequency shift can be expressed as
vB  vo aT DT:

Eq 4

Figure 4 illustrates the linear relationship between the

Brillouin frequency shift and the temperature for a system
with a reference frequency (vo ) of 14,650 MHz measured
at the ambient temperature with a rate of change of
1.4 MHz/C.
The Brillouin peak power can be expressed as [17]



Leff
aL
PB PCW e
1  exp gB PP
;
Eq 5
Aeff
where PCW is the input probe power; PP is the pulse power;
gB is the gain; and Leff and Aeff are the effective length and
effective area of the fiber, respectively. The Brillouin peak
power has a positive relationship to the temperature. As the

123

Frequency Shift - MHz

14800

14750

14700

14650
0

20

40

60

80

100

Temperature - C

Fig. 4 Brillouin frequency shift vs. temperature

temperature increases, the Brillouin peak power increases.

On the other hand, it is inversely related to strain. As strain
increases, the Brillouin peak power decreases [16, 17]. The
Brillouin peak power has a dependence on temperature and
strain [17, 18].
P B PO

dP
dP
DT
De;
dT
de

Eq 6

where PO is the reference power, ddPT is the temperature

coefficient (mW/C), and dP
de is the strain coefficient (mW/le).
Referring to Fig. 3, the fiber is laid near the pipeline, assuming
zero strain; the Brillouin peak power PB can be expressed as
PB PO

dP
DT:
dT

Eq 7

Either Eq 7 or Eq 4 can be used to determine the

temperature change. The location of the temperature
change can be determined using Eq 8:
d

CDt
;
2n

Eq 8

where d is the location of the temperature change, c is the

speed of light, n is the fiber optic cable refractive index,
and Dt is the time traveled.

Probability of False Alarm (PFA)

PFA is the probability of declaring a leak when in fact no
leak is present. Whenever noise power exceeds a predefined
noise power, a false alarm is declared. Stated differently,
the PFA is described as the likelihood that the LDS will
falsely detect a leak when none exists. The PFA is
expressed in terms of the threshold, which is the minimum

J Fail. Anal. and Preven. (2015) 15:873882

877

where PB(measured) is the measured Brillouin peak power

and NP is the noise power. Both the PB(measured) and NP are
measured at a given point in time. After obtaining the
values of the measured power, the temperature change can
be determined.

False Alarms Due to Noise

10
8
Noise

Threshold Power

Power

DT DTmeasured e;

where e is a random measurement error that corresponds to

noise in the signal with a mean of 0 and variance r2. The
Brillouin peak power PB in Eq 7 is the measured Brillouin
power. Substituting Eq 7 into 11, DT can be expressed as

0
-2
-4

The Mean
of the Noise

-6
-8

DT

-10
0

10

20

30

40

50

60

70

Time

DT
detectable temperature change. Any measured signal level
below the established threshold will be considered as a
noise signal and does not contain any true power.
Within the framework of signal detection theory, the
mean and standard deviation of the noise are random.
Noise is modeled as a normal distribution with zero mean
(l) and standard deviation (r): N (0, r), and a false alarm
occurs whenever the noise power exceeds the predefined
noise power [19]. Figure 5 illustrates the noise powers
varying levels over time and the threshold; any departure of
the noise power from the baseline (the threshold power)
will signal an alarm indicating false detection. In fact, the
signal is mainly a noise signal that is increased in amplitude and has exceeded the threshold value due to excessive
noise generated by the equipment, frequencies interfering
with the monitoring equipment, or other external factors.
Every signal has two elements, power and noise, that
correspond to measured temperature and error, respectively, as indicted in Eqs 914. It should be noted that
noise affects the final accuracy of the estimated DT.
Power Signal and Temperature Relationship
The signal-to-noise ratio is defined as
l2
:
r2

Eq 9

The variance corresponds to the noise power or

measurement error, and the square of the mean
corresponds to the power amplitude or measurement.
Referring to Eq 7, the power has a noise term that
results in error in the measurement of temperaturemeasured power signal = power ? noise:
PB PBmeasured NP;

Eq 10

PBmeasured  Po
e:
dP=
dT

Eq 12

Using Eqs 10 and 11, the temperature change (DT can

be expressed as

Fig. 5 Illustration of noise signal

SNR

Eq 11

PBmeasured  Po
NP
;

dP=
dP=
dT
dT

Eq 12:1

where Po is the reference power taken at the reference

temperature change (this is not to be confused with the
threshold temperature change) anddP=dT is the temperature
coefficient (mW/C). The first term in Eq 12.1 corresponds
to measured temperature and the second term corresponds
to measurement error (e). Mainly, the measured
temperature change is obtained from a set of n
measurements, DT1, DT2DTn, expressed as
DT1 DT2 . . .DTn
;
n
r
DT DTmeasured  p ;
n

Eq 13

DTmeasured

Eq 14

where DTmeasured is the measured temperature change, r is

the standard deviation of the measurement error, and n is
the number of samples or the number of measurements.
The system performs a number of scans, and each scan
measures the temperature of the monitored pipeline and
records the temperature at each point along the pipeline. At
the end of the scanning, the averages of measured temperatures as well as the variance of measurement errors are
computed.
Probability Density Function of the Noise Signal
Figure 6 depicts the probability density function for noise
power. As shown in the figure, the shaded area on the right
side represents the failure region. This is the probability
that the noise power (measurement error) exceeds the
predetermined threshold. The threshold represents the
minimum power level or the minimum detectable temperature change.

123

878

J Fail. Anal. and Preven. (2015) 15:873882

The PFA is expressed in terms of the threshold and the

noise power level (r2), and the threshold will be either the
lowest detectable power or the lowest detectable temperature change. The main task of the system is to detect if the
received signal power level has exceeded the threshold.
Therefore, to determine the PFA, the threshold and the
number of data samples need to be determined. Assuming
that the noise power level (r2) is known from previously
recorded data for a large population, then the noise level
for the sample of interest is r2/n, and the standard deviation
p
is (r/ n), where n is the number of data samples. There is
a PFA for every threshold value. The mean of the noise
signal is zero because the summation of the amplitudes of
the noise power is zero. This is because the noise power
signal levels vary and, as a result, the summation of the

different signal levels becomes zero. When the measured

signal level is greater than the threshold change, a false
alarm is declared.
Let us define Xth as the threshold power; then the
probability of false alarm becomes
X 2

2
1 1
PFA q r e2r =n dX:

Eq 15

2pr2 Xth
n

Integrating Eq 15 yields
2
13
0
16
B Xth C7
PFA 41  erf @qA5;
2
2r2

Eq 16

where erf is the error function. The threshold power is

calculated as
r
2r2 1
Xth
erf 1  2PFA:
Eq 17
n
Figure 7 illustrates the PFA for various threshold power
values for a system that has a bandwidth (DVB ) of 25 MHz
and a temperature coefficient (aT) of 1.52 MHz/C. The
figure reveals that as the threshold value increases, the PFA
decreases.
Figure 7 is re-plotted in Fig. 8 to show the temperature
change. This is for a system that has the same input data as
indicated above and a temperature coefficient (dP/dT) of
0.9 mW/C.
Minimum Detectable Temperature by Fiber OpticBased LDS
The minimum detectable Brillouin frequency shift dVB as a
function of SNR is expressed in Eq 18 [20]:

Fig. 6 Amplitude distribution of the noise power signal

10

10

10

10

False Detection

10

Probability of False Alaram

Probability of False Alaram

10

-5

-10

-15

-20

10

10

10

10
0

10

15

Brillouin Power Change - W

Fig. 7 PFA for various changing levels of power

123

20

False Detection

-5

-10

-15

-20

10

15

Temperature Change - C

Fig. 8 PFA for various temperature values

20

J Fail. Anal. and Preven. (2015) 15:873882

10

879

False Detection

10

10

Std. : 5
Std. : 10

0.3

Probability of False Alarm

Probability of False Alarm

Std. : 5
Std. : 10

-5

-10

0.25
0.2
0.15
0.1
0.05

10

-15

Temperature Change - C

10

12

Fig. 9 PFA for various temperatures using two different standard

deviation values

DVB
dVB p
;
2SNR0:25

Temperature Change - C

Fig. 10 PFA for various temperatures using two different standard

deviation values

Eq 18

PD versus PFA

where DVB is the Brillouin spectral width of the Brillouin

input signal and SNR is the signal-to-noise ratio. Using
Eqs 18 and 4 and assuming zero strain, the minimum
detectable temperature change dT can be expressed as [20]

Probability of Detection and Missed Detection

The PD is the likelihood that the LDS will correctly detect
an actual leak. The Probability of Missed Detection is the

Sample Size:24

0.1

0.2

0.4

0.3

0.5
PFA

0.6

0.7

0.8

0.9

1
PD

Eq 19

where aT is the temperature change coefficient expressed in

MHz/C and dVB is the difference between the reference
Brillouin frequency and the measured Brillouin frequency
shift (vB  vo ) as indicated in Eqs 3 and 4.
For a system with aT = 1.52 MHz/C, SNR =
21.18 dB, and DVB = 25 MHz, dT becomes 3.44 C; this
is the minimum temperature change that can be detected by
the system. From this, the minimum detectable peak power
can be estimated using Eq 7. Assuming that the Brillouin
reference power is 128 lW at a reference temperature of
5 C, with a power-temperature coefficient 0.9 lW/C, the
power threshold can be estimated to be 131.1 lW and the
threshold power change to be 3.1 lW.
The variance of measurement has an effect on the PFA
as indicated in Figs. 9 and 10; as the measurement variance
increases, the PFA increases.
An increase or decrease in the number of samples affects
the PFA and PD. This is illustrated in Fig. 11; as the size of
the samples increases, the PFA decreases.

X: 0.3914
Y: 1

0.5
0

X: 0.3031
Y: 1

0.5

Sample Size:100

0
0

0.1

0.2

0.3

0.4

0.5
PFA

0.6

0.7

0.8

0.9

1
PD

DVB
dT p
;
2aT SNR0:25

PD

X: 0.2133
Y: 1

0.5

Sample Size:1000

0
0

0.1

0.2

0.3

0.4

0.5
PFA

0.6

0.7

0.8

0.9

Fig. 11 PD vs. PFA using different sample sizes

probability that the system will not declare an actual leak.

The signal here is a combination of true signal power and
noise power that, respectively, correspond to measured
temperature change and measurement error of the temperature change. Figure 12 shows the probability density
function for power signal plus noise.
The shaded area in red is the failure region and it represents the probability of missed detection. The area to the
right side of the shaded area represents the probability of
detection. Detection occurs when the measured signal
power exceeds or equals a predetermined threshold and is
estimated by integrating the curve bounded by the lowest
detectable change and infinity. Let us define the lowest
detectable power change as Xth and power change as X; the

123

880

J Fail. Anal. and Preven. (2015) 15:873882

probability of detection will be the probability that X is

greater than or equal to the threshold (Xth) which can be
expressed as

13

16
BXth  lC7
PD 41  erf @ q A5:
2
2r2

Eq 22

n
1

Xl2

2
1
PD q r e2r =n dX;

Eq 20

2pr2 Xth
n

where Xth is the threshold power level change, n is the

number of measurements, l is the mean value taken for
every group of measurements, X is the amplitude that
represents the measured power level of the incoming
signal, and r2 is the noise power. Probability of missed
detection (PMD) can be expressed as
PMD 1  PD:

Four possible cases result from modeling the detection

and false detection:
PD

Leak exists and LDS indicates that a leak

exists

(hit)

PMD

Leak exists and LDS does not report the leak (miss)

PFA

Leak does not exist and LDS indicates it exists (false alarm)

POCR Leak does not exist and LDS indicates that it (correct
does not exist.
rejection)

Eq 21

Integrating Eq 20 yields
Figures 13 and 14, respectively, show the PD versus the
power level and temperature changes at a given threshold.
This is taken for a case with a threshold power change of
1.4 lW and temperature change of 1.5 C, SNR of 21 dB,
and temperature coefficient of 1.52 MHz/C. As the figures illustrates, PD is directly proportional to the power
level and temperature changes; the higher the power or
temperature changes, the higher the PD. According to
Eq 22, there are three parameters that need to be determined: the noise variance, the mean of the data samples,
which are obtained from the characteristics of the received
signal, and the threshold, which is determined as a function
of PFA using Eq 16.
PD can be expressed in terms of PFA by substituting
Eq 17 into Eq 22 as


p 
1
l n
PD erfc erfc1 2PFA  p ;
Eq 23
2
2r2
Fig. 12 Amplitude distribution of the power signal

Probability of Detection for various Power Levels - Threshold = 1.4 -W

Probability of Detection for Various Temperatures - Threshold = 1.5 C

10

Probability of Detection

Probability of Detection

10

-5

10

-10

10

-15

10

10

10

-5

10

Brillouin Power Change - W

Fig. 13 PD vs. signal power level change

123

10

15

20

-5

-10

-15

-5

10

15

Temperature Change - C

Fig. 14 PD vs. signal temperature change

20

25

J Fail. Anal. and Preven. (2015) 15:873882

881
PD versus PFA

Probability of Detection @ Different Values of SNR Values

1

0.8

Probability of Detection

Probability of Detection

0.9

0.7
0.6
0.5
0.4
0.3
0.2

SNR:30dB - Tth:1.8C - pfa:0.16

SNR:20dB - Tth:3.1C - pfa:0.037
SNR:10dB - Tth:5.60C - pfa0.00074
SNR:5dB - Tth:7.5C - pfa:0.000011

0.1
-2

10

12

14

0.8

0.6

0.4

0.2

16

Temperature Change - C

0.2

0.4

0.6

0.8

Probability of False Alarm

Fig. 15 PD for various thresholds with different PFA and SNR

values

Fig. 16 PD vs. PFA

Probability of Detection @ Different Numbers of Samples
1

1

and
p
p
nl
SNR
:
r

0.9

Probability of Detection

where erfc is the complementary error function and erfc

is the inverse of the complementary error function
(Fig. 15).
Referring to Eq 9, which is given below in Eq 24, the
2
power will be l2 and the noise power is r n.
l2
SNR 2
Eq 24
r=
n

0.8
0.7
0.6
0.5
0.4
Sample Size:50
Sample Size:100
Sample Size:200
Sample Size:400

0.3
0.2

Eq 25

Equation 23 can be expressed in terms of SNR as

"
r!#
1
SNR
1
PD
erfc erfc 2PFA 
:
Eq 26
2
2
The PD can be expressed in terms of the threshold by
substituting Eqs 19 into 22:

r

1
n
DVB
p
PD 1  erf
l
:
2
2r2
2SNR0:25 aT
Eq 27
Figure 16 illustrates that the PD increases with the
increase of the temperature change. This means that there
is higher probability of detection for higher temperature
change. Another observation that can be noted is that the
increase of the SNR leads to an increase of the PD and the
PFA together.
Figure 16 shows the relationship between PD and PFA;
for every PD value, there is a corresponding PFA value.

0.1
0

-4

-2

10

12

14

Temperature Change - C

Fig. 17 PD for different sample sizes

The figure also illustrates that as the PFA increases, PD

simultaneously increases.
Moreover, the larger the sample size or number of scans,
the better the SNR and PD values. Figure 17 illustrates that
as the sample size increases, the PD increases when the
detected T is higher than the threshold.

Total Probability of Missed Detection and Total

Probability of False Alarm
The system is considered a series system because the
failure of one LDS segment will lead to a complete failure
of the system. Assuming that the pipeline has n number of

123

882

J Fail. Anal. and Preven. (2015) 15:873882

independent optical repeaters located along the pipeline to

boost and regenerate the signal to achieve complete sensing
coverage, then the total probability of missed detection
(PMDT ) for the entire LDS can be expressed as
n
Y
PMDT 1  1  PMDn ; i 1; 2; . . .; n:
Eq 28
i1

Similarly, the total probability of false alarm (PFAT ) for

the entire LDS can be expressed as
n
Y
Eq 29
PFAT 1  1  PFAn ; i 1; 2; . . .; n:
i1

Conclusion
The PFA and the PD have been formulated and analyzed
for a fiber optic-based LDS. These two parameters form the
fundamental building blocks for assessing the performance
of the LDS and assist operators to implement the course of
action that might be required to enhance the reliability of
the system.
The analysis revealed that as the SNR increases, the PD
and PFA increase. The simultaneous increase of the PD and
the PFA presents a great challenge when designing or
specifying the system. Figure 16 illustrates a direct relationship between these two parameters; the PD increases
with the increase of PFA. To address this challenge, an
acceptable PFA should be selected first from which a satisfactory threshold can be determined. Once the threshold
is established, the PD can be immediately determined.
The missed detection and false detection are both
detrimental to the system performance, and the consequences of their occurrence cost time and money. The
question is which one is more costly; obviously, the consequences of missed detection will result in a greater
financial burden than the consequences of the false detection. Therefore, the initial step in designing such systems is
to determine the magnitude of risk that the operators can
accept and tolerate in the event a missed detection takes
place. Once this step has been accomplished, the performance parameters can easily be established. Future work
will extend this research to develop a probabilistic performance assessment framework for monitoring process
components.
Acknowledgments The authors gratefully acknowledge and
appreciate the partial financial support provided by the Research &
Development Corporation (RDC) of Newfoundland and Labrador,
Canada. Likewise, the authors greatly appreciate and thank ASMEs
permission to publish this paper in the Journal of Failure Analysis and
Prevention.

123

References
1. M. Nikles, Long-Distance Fiber Optic Sensing Solutions for
Pipeline Leakage, Intrusion and Ground Movement Detection.
Proc. SPIE 7316, 7316017316013 (2009)
2. H.F. Ulrich, E.P. Lehrmann, Telecommunications Research
Trends (Nova Science Publishers Inc., New York, 2008)
3. J. Park, G. Bolognini, D. Lee, P. Kim, P. Cho, F. Pasquale, N.
Park, Raman-based distributed temperature sensor with simplex
coding and link optimization. IEEE Photonics Technol. Lett. 18,
18791881 (2006)
4. Oil & Gas Journal. 2010, http://www.ilf.com/fileadmin/user_
upload/publikationen/OGJ_Sept-06-10_Walk_Frings
5. X. Bao, L. Chen, Recent progress in distributed fiber optic sensors.
Sensors 2012(12), 86018639 (2012). doi:10.3390/s120708601
6. X. Bao, M. DeMerchant, A. Brown, T. Bremner, Tensile and
compressive strain measurement in the lab and field with the
distributed Brillouin scattering sensor. J. Lightwave Technol.
19(11), 16981704 (2001)
7. G.P. Agrawal, Nonlinear Fiber Optics, 3rd edn. (Academic Press,
New York, 2001)
8. M. Soto, G. Bolognini, F. Pasquale, Long-range simplex-coded
BOTDA sensor over 120 Km distance employing optical preamplification. Opt. Lett. 36(2), 232234 (2011)
9. X. Bao, L. Chen, Recent progress in distributed fiber optic sensors. Sensors 12, 86018639 (2012). doi:10.3390/s120708601
10. X. Bao, D. Webb, D. Jackson, A 32-km distributed temperature
sensor based on Brillouin loss in optical fiber. Opt. Lett. 18(18),
15611563 (1993)
11. X. Bao, J. Dhliwayo, N. Heron, D.J. Webb, D.A. Jackson,
Experimental and theoretical studies on distributed temperature
sensor based on Brillouin scattering. J. Light Wave Technol.
13(7), 13401347 (1995)
12. M. Belal, Development of a High Spatial Resolution Temperature
Compensated Distributed Strain Sensor, PhD Thesis, Physical
and Applied Science Optoelectronics Research Center, University
of Southampton, Southampton, UK, 2011
13. T. Horiguchi, M. Tateda, Optical-fiber-attenuation investigation
using stimulated Brillouin scattering between a pulse and a
continuous wave. Opt. Lett. 14, 408410 (1989)
14. J. R. Smith, Characterization of the Brillouin Loss Spectrum for
Simultaneous Distributed Sensing of Strain and Temperature,
MSc. Thesis, Department of Physics, University of New Brunswick, NB, Canada, 1999A
15. K. Brown, Improvement of a Brillouin Scattering based Distributed Fiber Optic Sensor, PhD Thesis, University of New
Brunswick, 2006
16. T. Parker, M. Farhadiroushan, R. Feced, V.A. Handerek, A.
Rogers, Simultaneous distributed measurement of strain and
temperature from noise-initiated Brillouin scattering in optical
fibers. IEEE J. Quantum Electron. 34, 645659 (1998)
17. J. Smith, A. Brown, M. DeMerchant, X. Bao, Simultaneous
distributed and temperature measurement. Appl. Opt. 38, 5382
5388 (1999B)
18. P.C. Wait, T.P. Newson, Landau Placzek ratio applied to distributed fiber sensing. Opt. Commun. 122, 141146 (1996)
19. T.D. Wickens, Elementary Signal Detection Theory (Oxford
University Press Inc., New York, 2002)
20. T. Horiguchi, T. Kurashima, Y. Koyamada, Measurement of
Temperature and Strain Distribution by Brillouin Frequency Shift
in Silica Optical Fibers, in SPIE Distributed and Multiplexed
Fiber Optic Sensors, vol. 1797 (11) (1992)