LEAK DETECTION ALGORITH FOR PIPELINE NETWORKS

SCHOOL OF PHYSICS AUTH – MSc in Electronic Physics 1

LEAK DETECTION ALGORITH FOR

PIPELINE NETWORKS

PAPASTAVROU GEORGIOS-NAPOLEON

Identification Number: 11261

Supervisor

Nikolaidis Spyridon, Professor

Thessaloniki 16/03/2021

ARISTOTLE UNIVERSITY OF THESSALONIKI

SCHOOL OF PHYSICS

MSc in Electronic Physics

(Radioelectrology)



ABSTRACT

Pipe networks constitute the means of transporting fluids widely used nowadays.

Considering the importance of monitoring pipeline systems, to increase the operational

reliability and minimize the losses, this work focuses on the development of a technique to

detect leakages (either gas or liquid) in pipelines, based on the acoustic method. This method

is a non-destructive method which has presented reliable results. The methodology consisted

of capturing the experimental data through a sensor (e.g., piezoelectric, accelerometer) installed

at the surface of the pipeline without penetrating it and coupled to an analog to digital converter

to store and analyze the data.

Acoustic emission signals of pipeline carry information about structure integrity. This

work aims to study the characteristics of signals and develop an efficient algorithm for leakage

detection. According to the literature, there are a lot of features that varies from normal

operation to leakage cases. Time domain and frequency domain features are the key to our

study and have been examined in depth. Especially low frequencies (up to 11kHz) are mainly

studied because of the increase in the range of sensors and thus the reduction of the cost. The

proposed detection algorithm considers the frequency contents of the noise and the signals of

the leaks as well as some statistical characteristics of the noise.

Another important benefit of this method is the real time monitoring of the pipe network

which leads to an almost instant response. Minimizing the losses is crucial for domains like oil

refineries and power plants, as they can cause financial break down, impact the environment

negatively and affect the functioning of domestic household.

However, applying such a method in noisy environments is still a challenge. Because

of the level of noise and the complexity of acoustic signal in low frequencies, the identification

of an event becomes very difficult. In addition to that, accidental and periodic events like pump

decompression make the leak detection even harder. All these obstacles have been verified by

experimental work and the challenge of this thesis was to overcome that kind of difficulties

and make the final decision trustworthy.

An experimental set up along with two pipeline structures in a petroleum plant were the

application domains of the presented research. The algorithm of this research has been tested

for a wide range of parameters in real conditions with high accuracy.

Regardless the difficulties, the results of this work have shown reliable detection of

leaks in sufficient distances while the ability of the discrimination of accidental events been

achieved. The implementation of this algorithm to a single board computer was the last part of

this work, as the processing and the storage requirements are limited.

Keywords: leakage detection, pipelines, noise statistics, algorithm development, industry

applications


ACKNOWLEDGEMENTS

First and foremost, I would like to express my sincere gratitude to my advisor Professor

Nikolaidis Spyridon for the continuous support from my beginning of studies. His motivation,

patience and enthusiasm were extremely helpful. Furthermore, I want to thank him for giving

me the opportunity to take part in this research and write this thesis.

Besides Professor Nikolaidis, my sincere thanks go to my fellow labmates

Kousiopoulos Georgios-Panagiotis, Kampelopoulos Dimitrios and Karagiorgos Nikolaos for

stimulating discussions and encouragement during this research.

Last but not least, I would like to thank the Hellenic Petroleum plant in Thessaloniki,

Greece and the department of mechanical engineering (AUTH) for giving me the chance to

contact the experiments on pipe networks during real conditions.


Chapters Introduction……………………………………………………………………………………………6

CHAPTER 1 PIPELINE NETWORKS .............................................................................................. 7

1.1 Fundamentals of pipeline networks ..................................................................................... 7

1.2 Statement of the problem ..................................................................................................... 8

1.3 Experimental set up .............................................................................................................. 9

1.4 Hellenic Petroleum plant pipeline networks ..................................................................... 10

1.4.1 E-1404 (Water pipeline network) .............................................................................. 10

1.4.2 Dr-1452 (Atmospheric air pipeline network) ........................................................... 11

CHAPTER 2 LEAK DETECTION METHODOLOGIES ............................................................. 12

2.1 Background ......................................................................................................................... 12

2.2 Acoustic Emission method.................................................................................................. 13

2.3.1 Hardware ..................................................................................................................... 15

2.3.1 Software ....................................................................................................................... 18

CHAPTER 3 FEATURE ANALYSIS ............................................................................................... 19

3.1 Noise analysis and statistics variation ............................................................................... 19

3.2 Leakage cases analysis ....................................................................................................... 22

3.2.1 Experimental structure leakages ............................................................................... 24

3.2.2 Leakages in E-1404, Dr-1452 ..................................................................................... 26

3.3 Periodic/Accidental events analysis .................................................................................. 28

3.4 Correlation function analysis ............................................................................................ 31

CHAPTER 4 DEVELOPMENT OF THE ALGORITHM ............................................................. 32

4.1 Algorithm definition ........................................................................................................... 32

4.2 Leakage detection algorithm and results .......................................................................... 36

4.3 Conclusions .......................................................................................................................... 38

4.4 Future work ......................................................................................................................... 39

4.5 References ............................................................................................................................ 40


Introduction

In this master thesis, a leakage detection algorithm is developed for pipeline networks

in noisy environments. In addition to this, a verification of the above-mentioned algorithm

takes place in metallic pipes with real conditions not only at an experimental set up but also at

the Hellenic Petroleum plant. The aim of this research is to minimize the losses from leakages.

The concept based on the instant response. For instance, when an orifice is created, from a bare

minimum to a huge amount of waste takes place, so the response has to be immediate.

Pipeline networks remain the current and most efficient means for gas and liquid

transportation. It is already known that corrosion, aging, third-party damage are some of the

events that occur occasionally, which induces many serious problems. Some of the most

concrete examples are: waste of limited and valuable resources like petroleum or water, loss of

life and property, environmental pollution or in extreme cases leakages could be proved fatal

to human life. During the period of 2009-2010, in UK alone, approximately 3200 MEGA liters

of water were wasted due to failure or leaks in pipeline network.

As a result, leakage detection becomes a necessity in today’s pipeline networks. But

how constant monitoring takes place in pipelines whose length tends to be hundreds of

kilometers? Many methods have been proposed in the past [1-4]. In this thesis a non-destructive

method is proposed which uses acoustic signals that are created by the leaks [5-10]. In the

presence of leak in a pipe, vibro-acoustic waves are transmitted along the pipe walls. These

waves could be detected with the use of an acoustic sensor or accelerometers that are installed

at the outer surface of the pipe.

So, apart from the sensor, which is an accelerometer to this case, an analog to digital

converter along with a computer are required for the processing and storage of the data. At

first, a data acquisition card and a laptop have been used for this purpose. Last but not least,

simple batteries have been used for the power supply of our system.

On the grounds of how the leak is detected, regardless the increase in the existing power

of sound, some specific characteristics in frequency domain are presented which in general

differ from the ones of the ambient noise. Even though using acoustic signals for leakage

detection is a well-studied problem, detecting such signals in extremely noisy environments

like petroleum plants is still a challenge.

The development of this method started with the extensive analysis of the noise and its

statistics variation. Then, leak signals are studied also with the usage of mathematical tools

such as FFT and wavelet. Furthermore, an extensive comparison of the two has been conducted

for feature extraction. These features would be responsible for the output of our system, which

is no more than a logic output, “have a leak” or “don’t have a leak”. As for the last part or this

research, the implementation of this algorithm has been conducted as the process and the

storage requirements were limited.


CHAPTER 1 PIPELINE NETWORKS

1.1 Fundamentals of pipeline networks

Pipeline networks are the lifeline structures of the world for the transportation of liquid

or gas through a system of pipes, compressors, pumps, valves, regulators, tanks, reservoir or

even heaters. Data from 2014 reveal a total of 3.500.000Km of pipeline in 120 countries of the

world. They mainly carry chemical and petrochemical products over long distances. These

products could be valuable such as water but at the same time could be proved environmentally

hazardous (e.g., oil, natural gas).

In regard with the product, which is carried, the type of the pipe could differ in many

ways. Not only the element of the pipe itself (e.g., plastic pipe or metallic pipe) but also the

diameter and the thickness of it. Furthermore, characteristics such as pressure, temperature and

flow play extremely important role for the study of the network. This means that before a leak

detection system is set up, an extensive research of the pipeline network should first be

conducted.

Figure 1: A typical pipeline network from a petroleum plant

In addition to this, as it is observed from the above figure, the complexity of a typical

pipeline network is huge. This means that in case of a faulty operation, the root of the problem

may take too long to be found, and this delay could be proved critical. But despite the

complexity of the network, another think to be mentioned is the level of the noise in such

systems. One typical example is the noise cancellation headset which is required close to those

networks. Finally, the environment of network to network may differ. More specifically, there

are above the ground pipeline networks (as it can be seen on the above figure), buried pipeline

networks or even under water pipeline networks.

To conclude, there are numerous parameters that characterize a pipeline network. The

aim of this research is to study most of these parameters and develop an efficient leak detection

system which corresponds better to most of them.


1.2 Statement of the problem

Pipelines have to fulfill high demands of not only efficiency but also safety and

reliability. Most leaks are caused by corrosion and equipment failure or even incorrect

operation. Insufficient maintenance of the network may result in corrosion at the construction

joints, low points where moisture collects and harm the pipe. Of course, there are exterior

factors that may be responsible for leakage occurrence. For instance, nearby excavation or

natural force are some concrete examples.

According to the Pipeline and Hazardous Materials Safety Administration of U.S. the

64% of U.S. energy commodities transported by pipeline networks. 1.6 BILION tons of

hazardous materials shipped annually by all modes. In 2012 it is reported damages of $450

million, 16 fatalities and 61 injuries as the annual average for the last 10 years. As for the

extensive usage of natural gas nowadays, the total length of gas pipelines has increased

significantly. This makes pipeline leakage occurring an even more possible scenario.

Apart from the energy sources, water leakages are also extremely important. A

significant amount of water is lost in the water supply system. Water leakages have been a

major problem for many regions around the world. In some countries water loss due to water

leakages in the supply network exceeds 40% of the water in the supply system. For instance,

municipalities in Norway are seeing an average water loss of more than 30% due to leakages

in their water distribution network. Traditional field survey methods are costly and time-

consuming, as the networks are becoming bigger and bigger.

Another important reason for leakage occurring is the aging of the pipes. Typical

examples of water distribution systems in many cities worldwide contain a large percentage of

old pipes. For example, in German systems over 54% of the pipes are older than 25 years and

24% are older than 50 years. So, if the life span of the pipe expired, it is reasonable to be more

vulnerable to leakage cases.

As a result, it is important to find an efficient way to detect the leakage quickly and

locate the leak position accurately. If the damage is diagnosed early, it could be repaired, and

the network could be functional in a short period of time without any adverse impact on the

society. Many methods have been proposed in the past for leakage detection. Some of them

require the pause of the operation of the whole network or even a small segment of it. Other

methods require days or in some cases even weeks in order to detect the leak. Acoustic method

has been proven to be the most effective while real time monitoring of the network is taken

place, which leads in quick detection with high sensitivity. But the problem with this method

is the noise dependance and the false alarm rate.


1.3 Experimental set up

For the purpose of this research an experimental structure has been created. It includes

a 20-meter steel pipe with a diameter of 8cm, filled with water and equipped with a proper

pump for flow creation and an expansion vessel for pressure control. The pipe carries

approximately 100lt of water and its thickness is 3.6mm while the capacity of the expansion

vessel is 100lt. The pressure level is regulated from 1 bar to 10 bars which is the max pressure

that the pipe could withstand while the frequency of the flow fluctuates from 0-50Hz. Last but

not least, an electrical resistance in the pipe is responsible for the temperature increase of the

water.

In 5 different places on the pipe, 4 groups of orifices of 0.5mm, 1mm, 2mm and 4mm

have been created and stay closed using taps. In order to emulate the leak, the corresponding

tap should be opened. With this structure, a wide range of parameters is controlled.

Additionally, leaks in different distances and with 4 different sizes are checked. The

experimental set up where the measurements have been performed, is shown in Figure 2.

Figure 2: Experimental structure

Furthermore, noise of about 30 minutes duration from the pipes in the petroleum plant

has been recorded and has been applied in the experimental structure with the help of an

amplifier and a loudspeaker. The RMS value of the applied noise in the experimental structure

was set to be similar to the corresponding RMS value of the measured noise in the plant.

Real time conditions tried to be emulated in this experimental structure. Although

conditions in a petroleum plant are far more complicated, this set up was the perfect place for

testing the algorithm and fixing any discrepancies. Finally, one thing that should be mentioned

is that not only for safety reasons but also for economic reasons this structure is placed inside

a dry cargo container and carries only water instead of any other oily product.


1.4 Hellenic Petroleum plant pipeline networks

Apart from the experimental structure, this research used as application domain the

facilities of the Hellenic Petroleum plant in Thessaloniki, Greece. Some specific rules had to

be followed in order to have access at the pipeline network. First and foremost, special suit

which protects the worker against fire and specific shoes with outdoor shell had to be

purchased. Moreover, helmet, special glasses and noise cancellation headset were provided by

the industry as it was another rule in order to have access at some specific networks. After the

access to the refinery was granted, a research for the selection of the proper pipeline network

had been started. More specifically, two pipeline structures with no oily products were the aim

of the research. One structure should have carried liquid and the other some kind of gas in order

to check both states of matter. Another obstacle that emerged was the temperature of the

pipeline structure. Because of the mounting of the sensors on the surface of the pipe, the

temperature should have been between the range of the temperature that the sensor could

withstand. Last but far from least, leaks should have been emulated with the usage of valves.

After all these requirements two different steel pipeline structures have been selected for the

development of the leak detection method.

1.4.1 E-1404 (Water pipeline network)

The first pipeline is called with the code E-1404. This structure carries water while its

diameter is approximately 8 inches. The pressure of the liquid inside the pipe is 4kg/cm2 and

the temperature is around 20-35°C. Additionally, the wall thickness of the pipe is 8.18mm

while the flow rate of the water is around 78m3/h. The length of the whole structure (E-1404)

is huge and complicated. So, a small part of this structure has been selected for the purpose of

the research. The length of the selected part is around 100m which was more than enough. The

selected part of the pipeline network E-1404 is pictured at the following figure.

Figure 3: E-1404 Pipeline Structure

With red circle are pictured the emulated leaks, while ch0, ch1, ch2 are the most common

spots for the sensors.


1.4.2 Dr-1452 (Atmospheric air pipeline network)

The second pipeline structure is called with the code Dr-1452. This structure carries

atmospheric air while its diameter is approximately 2 inches. The pressure of the gas inside the

pipe is 8.3kg/cm2 and the temperature is around 40°C. Furthermore, the wall thickness of the

pipe is 6.67mm while the flow rate of the gas is around 240sm3 /h. The length of the whole

structure (Dr-1452) is huge and complicated. So, a small part of this structure has been selected

for the purpose of the research. The length of the selected pipeline structure is approximately

150m. In figure 4 the design of the network is provided.

Figure 4: Dr-1452 Pipeline Structure

Similar to previous figure, with red circle are pictured the emulated leaks while ch0,

ch1, ch2 are the most common spots for the sensors. In contrast with the E-1404, a pneumatic

valve is included in this structure. This valve is used for the expansion of the gas, and the relief

takes place approximately every 5minutes.

Another important difference in these two structures is the pressure inside the pipe.

Because of the gaseous state of matter in Dr-1452 the pressure is 8.3kg/cm2, while in E-1404

the pressure is only 4kg/cm2. This difference will be proved extremely important in next

chapters.


CHAPTER 2 LEAK DETECTION METHODOLOGIES

2.1 Background

As it is already mentioned leaks occur in pipeline networks due to corrosion, faulty

operation, equipment failure, excavation, natural forces, or even third-party damage. There are

a lot of methods that have been proposed for the detection of the leaks and inspection of the

pipe. Most of the times the selection of the method depends on the undergoing circumstances.

For example, is the state of matter gas or liquid? Or is the product oily or not? Or what is the

temperature of the fluid inside the pipe?[11-14]

Typical examples require the pipeline to be shut down for damage detection. More

specifically, a hydrostatic test is taken place so as to test the axial flaws on pipelines. This is a

strength test. The fluid which is carried by the pipeline is removed, the network with the

potential leak is isolated and the pipe is pressured with the usage of a gas. On the one hand it

is acceptable that this method has performed reliable results but on the other hand this is a time-

consuming method which leads to economic losses.

One completely different method is the ground penetrating radar which is a geophysical

method that survey the landscape. This device gives images of the subsurface and features of

the soil. Neighboring differences can mask the detection of the leak and with the usage of

diffraction tomography the leak stands out clearly.

In addition to this kind of approaches, there are numerous methodologies in the

literature for real time leak detection. These methods could be categorized as internal methods

and externals methods, depending on the utilization of the internal or external parameters of

the pipeline. Pressure drop of the fluid is one concrete example of internal method. There are

two pressure sensors at the beginning and at the end of the network and they compare the level

of the pressure. If the pressure at the end of the network is lower than the beginning, then

something faulty is happening.

Vapor sensing is one typical example of exterior method. It works with different

approaches. One approach suggests the sampling of soil around the pipeline and then these

samples are analyzed for contamination by oily products. Other approach suggests the usage

of humidity sensors close to the pipe, when the humidity level is raised, a potential leak may

occur. But this method works only on buried pipeline networks.

Among monitoring methods, the Acoustic based method has performed better results

than the others. Not only detecting the leak but also pinpointing the location in real time testing.

The main drawback is because of the attenuation in pipeline structures the proximity of the

sensor is close. A summarization table shows the main advantages and disadvantages of

methods that have been studied.


Table I. Leak detection methods

Method Advantages Disadvantages

Hydrostatic test Accurate leak detection Expensive method with

delay of detecting the leak

Ground penetrating radar Effective for buried

pipelines

Expensive and works ONLY

for buried pipelines

Pressure drop method Inexpensive and easy to

implement

Leaks have to be great

enough

Vapor sensing Independent from operator Limited to only buried

pipelines

Acoustic Emission Detecting small leaks even

in buried pipelines and

pinpointing the location

Depended on noise method,

with close proximity of the

sensors

2.2 Acoustic Emission method

Acoustic emission method is a passive non-destructive method that based on

propagating elastic waves released from active sources. Passive means that only data

acquisition with some calculations are required in order to detect the leak, and non-destructive

means that the pipe isn’t penetrated. Acoustic emission method has been studied since 1980 by

several researchers. The most common approaches for detecting and localizing the leakage

involve the feature extraction and pattern recognition. Those features could be time-domain or

frequency domain features.

Acoustic emission is the transient elastic waves within a material, caused by the rapid

release of localized stress energy. An event source is the phenomenon which releases elastic

energy into the material, which then propagates as an elastic wave. Acoustic emissions can be

detected in frequency ranges up to 100MHz, but most of the released energy is within 1kHz to

1MHz range. The three major applications of AE techniques are:1) source location – determine

the locations where an event source occurred; 2) material mechanical performance – evaluate

and characterize materials/structures; and 3) health monitoring – monitor the safe operation of

a structure, for example, bridges, pressure containers, and pipelines, etc.

When a leak occurs, it produces an acoustic noise (energy) around the place of leakage.

More specifically, transient waves are generated by the rapid release of energy from localized

source within the material. These waves travel along the pipe and could be easily detected with

the usage of a sensor. If some features differ from the baseline, an alarm will be activated. The

following figure provide the topology of the method.


Figure 5: Acoustic emission method topology

It is easily understood from the topology that due to attenuation the closer the sensors

are to the leak, the better. In general, there are a lot of factors that affect the result of the method.

Geometry of the pipe, material, minimum leak rate, internal pressure are some concrete

examples.

In this research, an algorithm based on the acoustic emission method is developed with

very short response time. The proposed approach has been developed for protecting pipeline

structures in the process area of an oil refinery.

2.3 Equipment specifications

The selection criteria of equipment were one of the most critical issues. At first,

durability was a constant problem since every piece of this equipment would be used at the oil

refinery of Hellenic Petroleum. Secondly, the budget was restricted so the options were

narrowed down.

Αt first what kind of sensors would be utilized? According to the literature several

sensor configurations are deployed like microphones, hydrophones, piezoelectric sensors,

accelerometers, etc. Due to the combination of relatively flat frequency response along with

the low cost, accelerometers were the most appropriate choice. After the signal has been

received, digitalization of the data was the next step. As a result, an efficient analog to digital

converter should have been purchased. Last but not least, a laptop had been used for storing

and processing the data.


2.3.1 Hardware

The deployed accelerometer came from PCB PIEZOTRONICS and named after PCB

MODEL 352C33 [15]. Its specifications are shown below:

• Sensitivity: (±10%)100 mV/g (10.2 mV/(m/s²))

• Measurement Range: ±50 g pk (±490 m/s² pk)

• Broadband Resolution: 0.00015 g rms (0.0015 m/s²

rms)

• Frequency Range: (±5%)0.5 to 10000 Hz

• Sensing Element: Ceramic

• Weight: 0.20 oz (5.8 gm)

These sensors are easy to operate and interface with signal analysis, data acquisition

and recording instruments, powered by simple, inexpensive, constant-current signal

conditioners. For this reason, the constant-current source accompanies the sensor. A typical

schematic of the source is shown in Figure 6.

Figure 6: Typical System Schematic

The power supply consists of a current-regulated ,18 to 30 VDC source. This power is

regulated by a current-limiting circuit, which provides the constant-current excitation required

for proper operation of the sensor. In general, batter-powered devices offer versatility for

portable, low noise measurements, whereas line-powered units provide the capability for

continuous monitoring.

When choosing a mount method, a lot of parameters should be considered closely.

Characteristics like location, ruggedness, amplitude range, accessibility, temperature, and

portability play extremely important role. However, the most important and often overlooked

consideration is the mounting technique. According to the datasheet of the accelerometer, there

are six possible mounting techniques, and they affect the frequency response of the sensor.


Figure 7: Mounting Configurations and their effects on high frequency

On the above figure it is shown the six possible mount techniques and their effects on

the performance of the accelerometer. As it easily observed, the high-frequency response of

the accelerometer is compromised as mass is added to the system or the mounting stiffness is

reduced. In this research the mounting pad technique was chosen. This means that the effective

range of the frequency domain reaches up to 10kHz according to the specifications of the

sensor.

Apart from the accelerometer, the NI-9234 was part of the system [16]. It is a four-

channel dynamic range signal acquisition module for making high accuracy measurements

from IEPE sensors. The specifications of this data acquisition card are shown below:

• 4-Channel dynamic signal acquisition

• 102dB dynamic range

• 24-bit resolution

• 51.2 kS/s/ch

• -40°C to 70°C operating, 5g vibration, 50g shock

• Software-selectable AC/DC coupling

• Anti-aliasing filters

• Smart TEDS sensor compatibility

The input signal of each channel is buffered, conditioned, and then sampled by a 24-bit

Delta-Sigma ADC. The circuitry of the card is shown below.


Figure 8: Circuitry of NI-9234

This data acquisition card (DAQ) is connected to the sensor via BNC cable. The main

advantage of this module is the regulation of almost every parameter for the data acquisition

through an already known software, Labview. In addition to this, NI-9234 is attached to a

portable device NI-9174 CompactDAQ. All analog channels of data acquisition card are

referenced to chassis ground through a 50Ω resistor. This is a rugged chassis that integrates

connectivity, data acquisition and signal conditioning into modular I/O for directly interfacing

to any sensor or signal. It uses FIFO logic with a maximum of 127 samples per slot. A pictured

of this chassis is shown in Figure 8.

Figure 9: CompactDAQ

For the last part of connectivity, USB interface is used for the connection of

CompactDAQ with the laptop.

To sum up, the topology of this research includes one accelerometer which is connected

via BNC cable to the NI-9234 data acquisition card. The NI-9234 card which is attached to the

chassis (NI-9174), and finally the chassis which is connected to a laptop via USB cable.


2.3.1 Software

National Instruments which is the manufacturer of data acquisition card suggest the

usage of Labview software. Labview is an acronym and stands for Laboratory Virtual

Instrument Engineering Workbench (Labview) and it is a system develop platform for visual

programming language. Using CompactDaq with Labview provides not only easier

customization of data acquisition but also visualization of the measurement data.

LabVIEW integrates the creation of user interfaces (termed front panels) into the

development cycle. LabVIEW programs-subroutines are termed virtual instruments (VIs).

Each VI has three components: a block diagram, a front panel, and a connector pane. The last

is used to represent the VI in the block diagrams of other, calling VIs. The front panel is built

using controls and indicators. Controls are inputs: they allow a user to supply information to

the VI. Indicators are outputs: they indicate, or display, the results based on the inputs given to

the VI. The back panel, which is a block diagram, contains the graphical source code. All of

the objects placed on the front panel will appear on the back panel as terminals. The back panel

also contains structures and functions which perform operations on controls and supply data to

indicators. The structures and functions are found on the Functions palette and can be placed

on the back panel. Collectively controls, indicators, structures, and functions are referred to as

nodes. Nodes are connected to one another using wires, e.g., two controls and an indicator can

be wired to the addition function so that the indicator displays the sum of the two controls.

Thus, a virtual instrument can be run as either a program, with the front panel serving as a user

interface, or, when dropped as a node onto the block diagram, the front panel defines the inputs

and outputs for the node through the connector pane. This implies each VI can be easily tested

before being embedded as a subroutine into a larger program.

LabVIEW provides a powerful platform for undertaking a wide variety of different

applications. It started as an environment for managing test programming, but since its

inception, the applications for which it can be used have considerably expanded. It has

expanded from being a graphical test management language to become a graphical system

design environment.

This means that it can be used for an enormous variety of interesting and diverse

applications. Not only can it be used for equipment control (including the control of the large

Hadron Collider at CERN) and a variety of data acquisition applications (including car

development simulation where Big Data monitoring is undertaken) to the system design arena

where it has been used for development of projects from RF circuitry to biomedical equipment,

green technology and much more.


CHAPTER 3 FEATURE ANALYSIS

3.1 Noise analysis and statistics variation

To begin with, an extensive study of the noise was the first task of this thesis. As it

already mentioned, the presented research uses as application domain the environment of the

Hellenic Petroleum plant in Thessaloniki along with an experimental structure which was

presented in Chapter 1.3. The noise in different points in these structures has been studied. The

corresponding frequency spectrums as well as the statistics have also been analyzed. A number

of measurements in different time periods has been taken for this purpose.

The Fast Fourier Transform (FFT) was used as the mathematical tool to evaluate the

dominant peak in frequency spectrum during a measurement. Furthermore, the sampling

frequency was f=25.6kHz. According to the Nyquist theory the sufficient sampling rate has to

be double of the highest frequency of a signal (bandwidth). This means that using f=25.6kHz,

the bandwidth that is studied reaches up to 12800Hz. Actually, because of the anti-aliasing

filter which is included in the data acquisition card the maximum frequency that is measured

accurately is approximately 11000Hz.

In Figures 10 and 11 typical transient and spectrum diagrams of noise for the

experimental structure are provided.

Figure 10: Transient diagram for noise in experimental structure


Figure 11: FFT diagram for noise in experimental structure

As it is observed from the above Figures, most of the noise energy in the experimental

structure is concentrated at the lower frequencies, and more specifically at frequencies less than

1kHz. However, in case of pipeline structures in Hellenic petroleum plant, the noise energy

extends in frequencies up to 5 and 11-12kHz. This is reasonable, since such kind of noise is

mainly due to engines, pumps, and compressors, involved in the pipeline structures.

Additionally, it has to mentioned that these pipeline structures are placed in process areas, so

the noise magnitude differs.

In the following figures 12 and 13, spectrograms of noise for the E-1404 and Dr-1452

respectively are shown.

Figure 12: FFT diagmar for noise in E-1404


Figure 13: FFT diagmar for noise in DR-1452

The differentiation of the noise from network to network was the first obstacle of this

research. To overcome this difficulty, an adaptive filtering is taken place at the beginning of

the proposed algorithm and will be discussed thoroughly at the next chapter. The critical issue

that is noticed is the stationarity of the noise behavior after filtering. To become more specific,

the 4 first moments of the filtered noise for data streams of 20s measured every 5 minutes in a

time period of one hour have been estimated for the pipeline structures. In mathematics, the

moments of a function are quantitative measures related to the shape of the function’s graph.

The concept is used in both mechanics and statistics. For example, if the function is a

probability distribution, then the zeroth moment is the total probability (i.e., one), the first

moment is the expected value, the second central moment is the variance, the third standardized

moment is the skewness, and the fourth standardized moment is the kurtosis. The mechanical

concept is closely related to the concept of moment in physics. The mathematical types of these

moments are shown below:

𝑀𝑒𝑎𝑛: 𝑥 =1

𝑛(∑ 𝑥𝑖) =

𝑥1 + 𝑥2 + ⋯ + 𝑥𝑛

𝑛

𝑛

𝑖=1

𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒: 𝜎2 =∑ (𝑋𝑖 − 𝑋)2𝑛

𝑖=1

𝑁

𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠: 𝑆𝑘𝑒𝑤 =1

𝑁∑ [

𝑋𝑖 − 𝑥

𝜎]

3𝑛

𝑖=1

𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠: 𝐾𝑢𝑟 =1

𝑁∑ [

𝑋𝑖 − 𝑥

𝜎]

4𝑛

𝑖=1


The results of the statistics are presented in Tables II and III:

Table II. Noise statistics after filtering for pipeline structure E-1404

Table III. Noise statistics after filtering for pipeline structure Dr-1452

It is shown that the dispersion is quite small. Only the skewness presents slightly higher

dispersion. For this reason, this metric is not used in the leakage detection method. Also, the

mean value is filtered as the DC component.

3.2 Leakage cases analysis

Apart from the noise study, leakage experiments have also taken place. It is noticed that

there are two characteristic waveform signatures: burst type and continuous type. Burst types

are the waveforms that the occurrence of the leak takes place between the time window of the

measurement, while continuous types are the waveforms that the occurrence of the leak has

been lost. Both types have been studied thoroughly during this research. In figures below,

concrete examples of both types are provided.

mean variance skewness kurtosis rms

1st 5′ 8.70E-08 1.04E-04 -0.00371 2.99872 0.01019

2nd 5′ -2.08E-08 1.15E-04 0.00140 2.99792 0.01074

3rd 5′ -3.72E-08 1.11E-04 -0.00362 2.99415 0.01055

4th 5′ -9.27E-08 1.19E-04 0.00042 2.99861 0.01089

5th 5′ 1.72E-09 1.07E-04 0.00080 3.00043 0.01035

6th 5′ 7.63E-08 1.10E-04 0.00407 3.01162 0.01049


1st 5′ 1.36E-09 2.64E-06 -0.00047 3.06219 0.00162

2nd 5′ 3.98E-09 2.62E-06 0.00274 3.58407 0.00162

3rd 5′ 3.58E-09 2.51E-06 0.00072 3.06312 0.00158

4th 5′ -3.7E-10 2.12E-06 -0.00015 3.05670 0.00146

5th 5′ 3.10E-09 2.66E-06 0.00282 3.06256 0.00163

6th 5′ 1.17E-09 2.51E-06 0.00847 3.03662 0.00158


Figure 14: Transient diagram of burst type leakage

Figure 15: Transient diagram of continuous type leakage

On figure 14, it is easy to be decided that a leak took place to the pipeline network, only

by looking the transient diagram. But on the figure 15, it is difficult to be called if there is a

leak or not. Most of the times leakage occurring is an almost instant event. This means that it

is extremely possible the outburst of the event NOT to take place between the time window of

the measurement.


3.2.1 Experimental structure leakages

Once the experimental structure was constructed, a series of measurements had been

started. The characteristics of orifices with different diameter had been tested. In this

experiment the sensor placed 4.30m away from the orifices, the pressure of the liquid inside

the pipe was around 7bars and the pump’s frequency was 30Hz. The FFT diagrams are shown

below.

Figure 16: FFT diagram for leak 1mm

Figure 17: FFT diagram for leak 2mm

With these measurements, it is noticed that the bigger the diameter of the orifice the

more the energy of the signal. In addition, the energy distribution depends on the diameter of

the orifice. This is extremely important because studying the FFT diagram may result in

guessing the diameter of the orifice.


Apart from the experiments with different diameter of orifices, even more tests had

been conducted with different parameters. Temperature, pressure, frequency of the pump were

some concrete examples.


3.2.2 Leakages in E-1404, Dr-1452

Except for the leakage tests in experimental structure, leakage tests have been also

conducted at the oil refinery of Hellenic Petroleum Plant. The specifications of each pipeline

structure have been performed at the chapter 1.4. In figure 18 and 19 typical spectrum diagrams

of leak signals in the two pipeline structures are provided. It is observed that, most of the energy

of the leak signal remains in higher frequencies. This leads to easy removal of the most part of

noise energy by simply using the appropriate low-pass filters without significant degradation

of the useful signal (due to leak). However, in some cases, depending on the noise energy

distribution, band stop filters may be also used. This strengthens the scenario of adaptive

filtering which mentioned in chapter 3.1.

Figure 18: FFT of leakage signal in the pipeline structure E-1404

Figure 19: FFT of leakage signal in the pipeline structure Dr-1452

It is also noticed, that due to the material at Dr-1452 which is atmospheric air under

8bars pressure, most of the energy concentrates in higher frequencies. In contrast with the water


under 4bars pressure which is the material of the E-1404 structure whose signal energy is

concentrated all over the bandwidth. So, it is observed that the higher pressure of the network

results in higher frequency bands.

Similar to noise signal experiments which were conducted in order to calculate the first

4 moments of the distribution, leakage experiments were also conducted to study the

differences.

Table IV. Leak statistics after filtering for pipeline structure E-1404

Table V. Leak statistics after filtering for pipeline structure Dr-1452

The first thing that is observed is that even in leaks the dispersion of these moments is

quite small. Furthermore, in comparison with noise signal features (Tables II and III), a huge

increase in variance and in rms is noticed. So, it is decided that variance and rms to be the first

two features of the algorithm. But would happen if an accidental or a periodic event took place?

That would result in an increase in signal energy, which in turn leads to an increase in variance

and rms. Thus, this problem pinpointed the next series of experiments.


1st 5′ 0,007865 0,001715 0,005659 3,004472 0,042153

2nd 5′ 0,005645 0,001758 0,015351 2,997891 0,042302

3rd 5′ 0,007707 0,001805 0,043565 3,018119 0,043177

4th 5′ 0,010541 0,001779 0,01638 3,043419 0,04347

5th 5′ 0,009953 0,001843 -0,06434 2,836491 0,044027

6th 5′ 0,008512 0,001847 0,01562 3,043419 0,043524


1st 5′ -0,00104 0,41955 -0,01434 3,073111 0,647725

2nd 5′ 0,002321 0,420666 -0,02125 3,051516 0,648586

3rd 5′ 0,004041 0,423518 -0,01783 3,068948 0,650782

4th 5′ 0,005307 0,447073 -0,01708 2,874512 0,668591

5th 5′ 0,012284 0,410724 0,270095 3,117749 0,640371

6th 5′ 0,010597 0,410885 -0,03237 2,86236 0,6456248


3.3 Periodic/Accidental events analysis

It is extremely possible during a measurement an accidental or periodic event to take

place. For example, an impulse from a rock which collides with the pipe, or a hit on the pipe

with an impact hammer are quite possible scenarios. Moreover, in complicated networks, such

as these at Hellenic Petroleum Plant, automatic valves that change their state are included. This

change would have an impact on the energy of the signal. Therefore, experiments with such

events had been conducted in order to study the phenomenon.

With the usage of an impact hammer, impulses had been created on the experimental

structure.

Figure 20: Transient Diagram of in impusle on the pipe

Figure 21: FFT diagram of an impulse on the pipe


A certain increase on the power of the signal is noticed. Furthermore, the energy

distribution of FFT diagram is similar to leakage cases. The key here, was the kurtosis metric.

Kurtosis is a statistical measure that describes the sharpness of the measured signal’s peak[17].

According to the literature in case of normal distribution, the kurtosis value is around 3. This

can be seen by Tables II-V. Higher Kurtosis levels means that occasional extreme values (either

positive or negative) are included in the signal. For instance, a hit on the pipe with an impact

hammer will cause an increase on the kurtosis value of the signal.

So, using an impact hammer again, impulses were created on the Dr-1452 pipeline

network. The results can be seen in next figure. This figure is a 3-axis transient diagram. Left

axis represents the acceleration, horizontal axis stands for time and right axis is the Kurtosis

value. The sharp spikes of the signal are the created impulses. Red dots represent the kurtosis

value of the signal every second. It is easily observed that the kurtosis value is around 3 for

every second that doesn’t include an impulse. When impulses are created, the kurtosis value

increases significantly.

Figure 22: Kurtosis value (dots) for several hits

Apart from accidental events, periodic events had been studied also. One comes from

the pneumatic valve relief (periodic) existing in the structure Dr-1452(compressed air). On the

following figure the transient signal is shown in a triple axis diagram. Similar to previous

Figure the right axis corresponds to Kurtosis value. The start of the event is quite obvious and

the problem which emerged is that the phenomenon needs approximately 20 seconds before

the stabilization of the system.


So, it is decided to include this feature to the leakage detection algorithm. Kurtosis

would be mainly used to recognize accidental and periodic events which otherwise would be

considered as leakages. In this way false alarms are avoided.


3.4 Correlation function analysis

The last feature that used in the leak detection algorithm was the cross-correlation

integral. In signal processing, cross-correlation is a measure of similarity of two series as a

function of the displacement of one relative to the other. It is similar in nature to the convolution

of two functions. If X and Y are two independent random variables with probability density

functions f and g, respectively, then the probability density of the difference Y-X is formally

given by the cross-correlation (in the signal-processing sense) f*g. In contrast, the convolution

f*g (equivalent to the cross-correlation of f(t) and g(-t)) gives the probability density function

of the sum X+Y [18].

For continuous functions f and g, the cross correlation defined as:

(𝑓 ∗ 𝑔)(𝜏) ≜ ∫ 𝑓(𝑡)𝑔(𝑡 + 𝜏)𝑑𝑡𝑡0+𝑇

𝑡0) (1)

For example, consider two real valued functions f and g, differing only by an unknown

shift along the x-axis. One can use the cross-correlation to find how much g must be shifted

along the x-axis to make it identical to f. The formula essentially slides the g function along

the x-axis, calculating the integral of their product at each position. When the functions match,

the value of f*g is maximized. This is because when peaks are aligned, they make a large

contribution to the integral. Similarly, when troughs align, they also make a positive

contribution to the integral because the product of two negative numbers is positive.

So, apart from the above-mentioned features of the filtered noise (rms, variance,

kurtosis), the correlation function is also used for leakage detection. It has been observed that

the noise of the pipes is correlated to a large extent. Filtered noise signals from different time

intervals, when correlated, the integral (summation) of their cross-correlation function is

relatively high. However, correlating a noise signal with a leak signal, results in a lower integral

value. That difference in the integral results allows the exploitation of the correlation function

as another metric for leakage detection.


CHAPTER 4 DEVELOPMENT OF THE ALGORITHM

4.1 Algorithm definition

To sum up, based mainly on the spectral differentiation between the leakage and the

noise signals, as well as the statistics of the noise after filtering, an algorithm has been

developed for leakage detection in noisy pipelines which includes the following steps:

• Appropriate filtering of noise.

• Definition of the used signal features for decision making regarding the

existence of leakage.

• Definition of threshold based on the statistics dispersion of the filtered noise.

• Determination of the correlated relation between noise and leak signal

• Recognition of accidental and periodic events which are differentiated from the

appearance of leakage.

• Leak detection as threshold exceedances.

It has to be mentioned that the proposed approach aims to monitor critical pipeline

structures of specific sectors in the plant which are placed over the ground and so finding the

location of the leakage is not a request.

Figure 23: Flowchart of the algorithm

The flowchart of the algorithm is given in Figure 23. Because of the used sampling

frequency of 25.6kHz, the effective spectrum for the measured signals in limited at about of

11kHz. At first, the filter bands for noise elimination have to be defined. Since most of the

noise energy is concentrated in frequencies below 250Hz, these frequencies are removed. The

remaining frequencies, from 250Hz to 11000Hz, are divided in discrete intervals with a step of

0.25kHz. The energy distribution of the noise in these frequency bands is extracted. A number

of 10 sets of measurements of predefined duration of 1s, are used for the extraction of the

average energy values of noise. The noise in bands where its average energy value exceeds the

expected one, when a uniform distribution is considered, is removed with the usage of band-

stop filters. Then the statistics of the filtered noise are calculated. In this algorithm the metrics

that are used are: the signals energy estimated in time, the power spectral density (PSD)

calculated in frequency, the kurtosis metric and finally the correlation integral. The extreme


values of the above features obtained from 10 sets of measurements (on the filtered noise) of

the same duration are found. Then appropriate thresholds as a percentage deviation from these

maximum values are defined. This means that any signal whose feature values overcomes the

defined threshold values, is a candidate for being a leakage signal. Thresholds have been

defined empirically, studying the noise behavior after many tests. Concrete examples are

shown on the below figures.

Figure 24: Probability density of signal energy


Figure 25: Probability density of Spectral Power

On the above histograms (Figures 24, 25), the probability density of noise and leak

signals for each metric are shown. For the normal distribution, the values less than one standard

deviation away from the mean account for 68.27% of the set; while two standard deviations

from the mean account for 95.45%; and three standard deviations account for 99.73%.

Thresholds have been defined adding three standard deviations at signal energy and spectral

power metrics distribution.

Apart from signal energy and spectral power, the correlation function is also used for

leakage detection. During the threshold estimation process, each one of the 10 sets of

measurements is correlated to each other and the integral of the correlation is calculated for

every iteration. The signal waveform with the lowest integral value is used as a reference and

the 85% of this value is used as threshold. During the detection procedure, every acquired

signal is correlated to that reference. As a result, a lower correlation value is an indication for

a potential leak.


Figure 26: Probability density of Correlation

As it is already mentioned, two kinds of accidental and periodic events have been

considered. One comes from the pneumatic valve relief (periodic) existing in pipeline structure

Dr-1452 (compressed air) and the other is possible hits on the pipes with metallic objects

(accidental). Kurtosis increases sufficiently in the case of such events and can detect such

phenomena eliminating possible false alarms.

The Boolean function model for leakage detection is given by expression (2):

T=X1 AND X2 AND X'3 AND X'4

Where Xi is a binary variable of the metric i and turns into 1 when the corresponding metric

violates its threshold value, otherwise it is 0. More specifically, X1 corresponds to the Energy

estimated in time, X2 corresponds to power (from PSD), X3 corresponds to Kurtosis and X4 to

the Correlation integral. T=1 is an indication of leak. Moreover, for the described algorithm, a

Virtual Instrument (VI) in LABVIEW has been developed, leading to an automatic leakage

detection system.

The simplicity of the algorithm and the limited requirements in terms of processing and

storage resources, makes it appropriate for implementation as an embedded system. A data

logger with the appropriate vibration sensor (accelerometer) located at the middle of a pipe and

incorporated with some storage and processing capability for real time calculations for the four

metrics, is enough for monitoring sufficient length (e.g. 150m) of the pipe. It has to be

mentioned that the configuration phase (filter bands and thresholds definition) is done once at

the beginning or periodically, e.g. once per hour, depending of the noise behavior. How often

the configuration of the system has to be determined, in the environment of the petroleum pant,

is under further study.


4.2 Leakage detection algorithm and results

The algorithm is as follows. The first 10s (10 periods of 1s) are used for the automatic

configuration of the filter. The next 10s are used for defining of the thresholds and the reference

waveform for the correlation. In this time interval, the maximum values for each feature are

detected. To define the thresholds, it is used 3% and 5% difference over the maximum value

for the Energy and power, respectively and 30% over the maximum value of Kurtosis. The

thresholds have been defined ad hoc. However, after many measurements, it has been observed

that thresholds defined in this way result to be stricter than the case of 3σ over the mean value

(regular choice in case of normal distributions). So far, no false alarm has been observed in

experiments of half an hour.

After the algorithm configuration, the system checking for leaks every second. In order

to verify the proposed method, the whole parametric range of the experimental structure has

been applied. Various leaks were created in different distances for different orifice diameters,

pressure values, flows, and temperatures. The pressure range for the tests was in range 2.5-

7.5bars, the temperature was in the range 25-50°C and the flow corresponding to 10, 20 and 30

Hz operation of the pump. On the other hand, all the leaks are detected for the whole parametric

range. Only the most remote (at 12 meters) orifice of 0.5mm at a pressure lowers than 4bars

could not be detected reliably. The results of the above experiment are presented in the

following table.

The notation k/N in the contents of the table means for k the number of hits and for N

the number of experimental samples (measurements in a period of one second). For example,

in case of leaks, 300/300 means 100% success in the detection of the existence of leak in a

number of 300sets of measurements. In case of NO leak, 0/812 means no false alarm in a

number of 812 sets of measurements. N=812 corresponds to 13.52 minutes (812 seconds), and

N=300 corresponds to 5 minutes (300 seconds). Similar results were observed in many such

experiments verifying the method.

Table VI. Experimental Results


As illustrated in Table VI, there were zero alarms in case of NO leakage, which was a

primary target. It has to be mentioned that particular metrics may present some false alarms,

but not simultaneously and thus the expression (2) remains at 0. As for the leakages, 100%

success rate was achieved for leaks at a pressure level greater than 4bars, regardless the distance

and the diameter of the orifices. For the 0.5mm orifice and pressure less than 4bars, the success

was at 75% (225/30).

Accidental and periodic events were created in the experimental structure. All of them

had been detected without the creation of any false alarm. Moreover, different tests had been

carried out to check the effectiveness of the algorithm. For example, an increment to the

measurement duration has been tested to 5s. A similar effectiveness was observed. Also, the

measured signal mas multiplied by a scalar factor trying to emulate the use of Automatic Gain

Control. The situation remained unchanged.

The proposed algorithm was applied in the two pipelines of the oil refinery, E-1404 and

Dr-1452, with similar effectiveness. The automatic configuration was employed in a

measurement window of 10s+10s for the determination of the appropriate filter and thresholds.

In E-1404 with a pressure of 4bars, a leakage of 8mm orifice was detected in a distance of 40m.

In the Dr-1452 pipe with a pressure of 8bars, a leakage of 4mm orifice was detected in a

distance of 75m.


4.3 Conclusions

Leak detection in pipeline networks is an important problem to be solved for preventing

catastrophic failures. Acoustic based method is a highly sensitive, nondestructive method that

relies on the propagating elastic waves created by leak turbulences [19-26].

In this research, a new algorithm for automatic detection of leakages in noisy pipeline

structures is introduced. The algorithm is based on the detection of the acoustic signals created

by the leakage. It exploits their spectral differentiation from the noise as well as the stationarity

of the statistics of the noise after filtering. Only four metrics are used for the detection of the

leaks, which leads to lower response time and fewer computational requirements. In a

measurement window of 10s+10s, the configuration of the appropriate filter for noise

elimination and the definition of the appropriate thresholds for leak signal identification are

firstly performed and then the detection procedure begins. The proposed leakage detection

algorithm is based on measurements in window of 1s, defining the response time of the method.

The method was verified in an experimental structure for a wide range of the four

parameters, orifice diameter, fluid flow, pressure and temperature and with the application of

a corresponding noise (from a petroleum plant) to create realistic conditions. No false alarms

were observed. All artificial leakages were detected except for the furthest (at 12m) with an

orifice diameter of 0.5mm and pressure below 4bars (worst case scenario). Also, the method

was applied in real conditions with the same success in two pipeline structures in a petroleum

plant. Leakages detected up to 75 meters[27-28] in pipes carrying air at 8 bar pressure and for

an orifice diameter of 4mm. In general, leak signals at a level of 12dB below the level of the

noise were detected reliably.


4.4 Future work

It is undeniable that there is still much work that can be done in this research. The next

steps include the optimization of the algorithm and the improvement of the method with many

ways:

• Statistical decision making. To become more specific, the output of the algorithm will

not be decided by only one measurement. Numerous measurements will be conducted,

and the output of the algorithm will be a statistical result from these measurements.

This will lead to better accuracy and in less false alarms.

• Implementation of the algorithm in a single-board computer (e.g., Raspberry) will result

in a more compact method and more attractive idea.

• Expand the method to more than one pipeline structures. The idea here, is to expand

the method to a whole pipeline network with hundreds of pipes. Numerous sensors will

be used in order to monitor the whole network 24 hours. There will be one central node

that will check the sensors and will set the alarm in leakage cases.

• Localization of the leakage is a future plan of this work. Pinpointing the exact place of

the orifice in large pipeline networks will be a savior for industrial businesses.


4.5 References

[1] Mutiu Adesina Adegboye, Wai-Keung Fung and Aditya Karnik, "Recent Advances in

Pipeline Monitoring and Oil Leakage Detection Technologies: |Principles and Approaches,"

Sensors 2019, 19, 2548, MDPI; doi:10.3390/s19112548

[2] Lei Shu, Mithun Mukherjee, Xiaoling Xu, Kun Wang, Xialing Wu, “A survey on Gas

Leakage Source Detection and Boundary Tracking with Wireless Sensor Networks,” IEEE

Access, Vol. 4, pp. 1700-1715, Apr. 2016. DOI: 10.1109/ACCESS.2016.2550033

[3] Lawrence Boaz, Shubi Kaijage and Ramadhani Sinde, “An overview of pipeline leak

detection and location systems”, Pan African International Conference on Science, Computing

and Telecommunications (PACT), July 2014. DOI: 10.1109/SCAT.2014.7055147.

[4] Pal-Stefan Murvay, Ioan Silea, “A survey on gas leak detection and localization

techniques,” Journal of Loss Prevention in the Process Industries, Vol. 25, Is. 6, pp. 966-973,

Nov. 2012

[5] Lingya Meng, Li Yuxing, Wang Wuchang, Fu Juntao, "Experimental study on leak

detection and location for gas pipeline based on acoustic method," Journal of Loss Prevention

in the Process Industries, 25, Is. 1, pp. 90-102, Jan. 2012.

[6] Hazim Yalcinkaya, "Reliable Monitoring of Leak in Gas Pipeline Using Acoustic Emission

Method," Master Thesis in Civil Engineering, University of Illinois, Chicago, 2013.

[7] Thang Bui Quy, Sohaid Muhammad, Jong-Myon Kim, "A Reliable Acoustic Emission

Based Technique for the Detection of a Small Leak in a Pipeline System," Energies MDPI,

2019, 12, 1472; doi:10.3390/en12081472

[8] Brunner, A. J., Barbezat, M., "Acoustic Emission Monitoring of Leaks in Pipes for

Transport of Liquid and Gaseous Media: A Model Experiment," Advanced Materials Research,

Vols. 13-14, pp. 351-356, Feb. 2006. https://doi.org/10.4028/www.scientific.net/AMR.13-

14.351

[9] Shuaiyong Li, Yumei Wen, Ping Li, Jin Yang, Lili Yang, “Leak Detection and Location for

Gas Pipelines Using Acoustic Emission Sensors”, IEEE International Ultrasonics Symposium,

pp. 957-960, Oct. 2012. DOI: 10.1109/ULTSYM.2012.0239

[10] Pugalenthi Karkulali, Himanshu Mishra, Abhisek Ukil, Justin Dauwels, “Leak Detection

in Gas Distribution Pipelines using Acoustic Impact Monitoring”, 42nd Annual Conference of

the IEEE Industrial Electronics Society, Oct. 2016. DOI: 10.1109/IECON.2016.7793352

[11] Hao Jin, Laibin Zhang, Wei Liang, Qikum Ding, "Integrated leakage detection and

localization model for gas pipelines based on the acoustic wave method," Journal of Loss

Prevention in the Process Industries, Vol. 27, pp. 74-88, 2014.

[12] Xu Qingqing, Zhang Laibin, Liang Wei, "Acoustic detection technology for gas pipeline

leakage," Process Safety and Environmental Protection, Vol. 91, pp. 253-261, 2013.

[13] W.Liang , L. Zhang, Q. Xu and C. Yan, “Gas pipeline leakage detection based on acoustic

technology”, Engineering Failure Analysis, vol. 31, pp. 1-7, 2013.

https://doi.org/10.1109/ACCESS.2016.2550033

https://doi.org/10.1109/SCAT.2014.7055147

https://doi.org/10.4028/www.scientific.net/AMR.13-14.351

https://doi.org/10.4028/www.scientific.net/AMR.13-14.351

https://doi.org/10.1109/ULTSYM.2012.0239

https://doi.org/10.1109/IECON.2016.7793352


[14] L. Weiguo, W. Xiaodong, W.Haiyan, M. Changli and W. Fenwei, “A dual-sensor-based

method to recognize pipeline leakage and interference signals”, Journal of Loss Prevention in

the Process Industries, vol. 38, pp.39-85, 2015.

[15] PCB Piezotronics, “ICP Accelerometer Installation and Operation Manual”, Model

352C33, 2002

[16] National Instruments, “Datasheet NI 9234 4 AI, ±5 V, 24 Bit, 51.2 kS/s/ch Simultaneous,

AC/DC Coupling, IEPEAC Coupling”, 2015.

[17] S. Kokoska and D. Zwillinger, “CRC Standard Probability and Statistics Tables and

Formulae, Student Edition,” 2000.

[18] Dimitrios Nikolaidis, “High performance Implementation of the cross-correlation

algorithm on FPGA for telecommunication applications” Diploma Thesis, NTUA, Greece, Oct.

2019.

[19] A. Martini, M. Troncossi and A. Rivola, “Leak detection in Water-Filled Small-

Diameter Polyethylene Pipes by Means of Acoustic Emission Measurements”, Applied

Sciences, vol. 7, no.1, p. 2, 2016.

[20] C. Liu, Y. Li, L. Meng, W. Wang and F. Zhang, “Study on leak-acoustics generation

mechanism for natural gas pipelines”, Journal of Loss Prevention in the Process Industries, vol.

32, pp. 174-181, 2014

[21] Y. Gao, M. Brennan, P. Joseph, J. Muggleton and O. Hunaidi, “On the selection of

acoustic/vibration sensors for leak detection in plastic water pipes”, Journal of Sound and

Vibration, vol. 283, no. 3-5, pp.927-941, 2005.

[22] Q. Xu, L. Zhang, and W. Liang, “Acoustic detection technology for gas pipeline leakage”,

Process Safety and Environmental Protection, vol. 91, no. 4, pp. 253-261,2013.

[23] F. Wang, W. Lin, Z. Liu, S. Wu and X. Qiu, “Pipeline leak detection by Using Time-

Domain Statistical Features”, IEEE Sensors Journal, vol. 17, no. 19, pp. 6431-6442, 2017.

[24] A. Lay-Ekuakille, C Pariset and A. Trotta, “Leak detection of complex pipelines based on

the filter diagonalization method: robust technique for eigenvalue assessment”, Measurement

Science and Technology, vol. 21, no. 11, p. 115403, 2010.

[25] W. Caesarendra and T. Tjahjowidodo, “A Review of Feature Extraction Methods in

Vibration-Based Condition Monitoring and Its Application for Degradation Trend Estimation

of Low-Speed Slew Bearing,” Machines, vol. 5, no. 4, p. 21, 2017.

[26] J. D. Butterfield, G. Meyers, V. Meruane, R. P. Collins, and S. B. M. Beck, “Experimental

investigation into techniques to predict leak shapes in water distribution systems using

vibration measurements,” Journal of Hydroinformatics, vol. 20, no. 4, pp. 815–828, Feb. 2018.

[27] Y.Zhang, S. Chen, J. Li and S. Jin, “Leak detection monitoring system of long distance

oil pipeline based on dynamic pressure transmitter”, Measurement, vol. 49, pp. 382-289, 2014

[28] G. Papastavrou, G. Kousiopoulos, D. Kampelopoulos, N. Karagiorgos, D. Porlidas and S.

Nikolaidis, "Leak Detection Algorithm for Pipelines in Noisy Environments," 2020 IEEE 20th


Mediterranean Electrotechnical Conference ( MELECON), Palermo, Italy, 2020, pp. 159-164,

doi: 10.1109/MELECON48756.2020.9140558.

LEAK DETECTION ALGORITH FOR PIPELINE NETWORKS

Documents