The use of pipeline is considered as a major means of conveying petroleum products such as fossil fuels, gases, chemicals and other essential hydrocarbon fluids that serve as assets to the economy of the nation [1
]. It has been shown that oil and gas pipeline networks are the most economical and safest mean of transporting crude oils and they fulfill a high demand for efficiency and reliability [2
]. For example, the estimated deaths due to accidents per ton-mile of shipped petroleum products are 87%, 4% and 2.7% higher using truck, ship and rail, respectively, compared to using pipelines [4
]. However, as transporting hazardous substances using miles-long pipelines has become popular across the globe in recent decades, the chance of the critical accidents due to pipeline failures increases [5
]. The causes of the failures are either intentional (like vandalism) or unintentional (like device/material failure and corrosion) damages [6
], leading to pipeline failure and thus resulting in irreversible damages which include financial losses and extreme environmental pollution, particularly when the leakage is not detected in a timely way [8
The average economic loss due to incidents of pipeline leakages is enormous [10
]. Over the past three decades, pipeline accidents in USA damaged property which costed nearly $
7 billion, killed over 500 people and injured thousands [11
]. For example, the incident of pipeline explosion in the community of San Bruno, California, USA on September 6, 2010 killed eight people, and injured more than fifty [11
]. In a similar incident of pipeline defect that occurred in Michigan, USA on July 26, 2010, more than 840,000 gallons of crude oil spilled into Kalamazoo River with estimated cost of $
800 million [11
]. The causes of pipeline damage vary. Figure 1
shows a pie chart that illustrates statistics of the major causes of pipelines failure which include pipeline corrosion, human negligence, defects during the process of installation and erection work, and flaws occurring during the manufacturing process and external factors [12
Based on these statistics, incidents of pipeline leakage are hard to entirely avoid as the sources of failures are diverse. However, in order to reduce the impacts of oil spillage on society it is very important to monitor pipelines for the timely detection of leakage or even leak prediction, as early detection of leaks will allow quick responses to stop oil discharge and proper pipeline maintenance. Hence, it is possible to reduce the loss rate, injuries and other serious societal and environmental consequences due to the pipeline failures.
Several pipeline leak detection methods have been proposed during the last decades using different working principles and approaches. Existing leakage detection methods are: acoustic emission [13
], fibre optic sensor [16
], ground penetration radar [19
], negative pressure wave [21
], pressure point analysis [24
], dynamic modelling [27
], vapour sampling, infrared thermography, digital signal processing and mass-volume balance [29
]. These methods have been classified using various frameworks. Some authors have classified them into two categories: hardware and software-based methods [34
]. In an attempt to group these methods based on technical nature further research efforts have been made [36
] which has led to the classification of available leakage detection systems into three major groups, namely internal, non-technical or non-continuous and external methods [2
]. In this study, we will classify different methods into the following categories: exterior, visual or biological, and interior or computational methods. A detailed classification of these methods is shown in Figure 2
. The exterior approach utilises various man-made sensing systems to achieve the detection task outside pipelines. Moreover, the biological approach utilises visual, auditory and/or olfactory senses of trained dogs or experienced personnel to detect leakage. In addition, the interior approach consists of software based methods that make use of smart computational algorithms with the help of sensors monitoring the internal pipeline environment for detection task. Remote monitoring can be achieved by carrying camera or sensing systems to designated locations by smart pigging, helicopter or Autonomous Underwater Vehicles (AUVs)/drones or using sensor networks [2
This paper aims to examine the state-of-the-art achievements in pipeline leakage detection technologies and to discuss research gaps and open issues that required attention in the field of pipelines leakage detection technology. The rest of the paper is organised as follows: Section 2
presents the exterior-based leak detection methods and compares their strengths and weaknesses; Section 3
presents the visual/biological-based leak detection methods; Section 4
presents the interior- based leak detection methods and features their strengths and weaknesses. The comparative performance analysis of the reviewed methods is given in Section 5
. Section 6
gives the guideline for selecting an appropriate leak detection method for various operating environments. The research gaps and open issues on pipeline leakage detection and characterisation are discussed in Section 7
. Finally, a summary of this paper, and possible future directions are presented in Section 8
3. Visual/ Biological Leak Detection Methods
Visual/biological methods of detecting leakages refer to the traditional process of detecting oil spillage in pipeline surroundings using trained dogs, experienced personnel, smart pigging or helicopters/drones [2
]. This method usually utilises trained personnel who walk along the pipelines and search for anomalous conditions in the pipelines environment. Trained observers can recognise the leaks through visual observation or smelling the odour coming out from crack point. Similarly, the noise or vibrations generated as oil escapes from rupture point also applicable in this method to detect and locate pipeline failures. Both dogs and smart pigging function in a similar way to the experienced personnel. The pig is sometimes equipped with sensors and data recording devices such as fluorescent, optical camera or video sensors with great sensing range if the visibility level is high. A trained dog is more sensitive to the odour of certain gases than human beings or pigging in some cases [113
]. Conversely, dogs are not effective for prolonged operation for more than 30–120 min of continuous searching due to fatigue [115
]. These on-site inspection methods can only be applied to onshore or shallow offshore pipeline networks. Besides, the detection time is also based on the frequency of inspections which normally takes place in some countries such as the USA for at least once every three weeks [35
]. The recent development of remotely operated vehicles (ROVs) has transformed the operation style of offshore oil transportation operators. It has been shown that ROVs are durable for performing subsea pipeline inspection tasks and functioning in deep water that cannot be accessible by dog, pigging or human divers [116
]. The operation principle of ROVs is based on teleoperation that involves a master-slave system. The slave is a ROV which is designed to interact with the extremely hazardous subsea environment while the master human operator is located in a safe place to remotely control the slave robot’s motions using input devices, like joysticks or haptic devices [117
]. All robot commands, sensory feedback and power are sent through an umbilical cable connecting the ROV and the deployment vessel.
The emergence of autonomous underwater vehicles (AUVs) in subsea pipeline inspection and monitoring has reduced the extent of human operator involvement in unmanned vehicles through the implementation of intelligent control machinery and thus drastically lower the chance of human casualties. Though, the operation principle of AUVs is similar to the teleoperation of ROVs, only limited skilled operators are required in supervisory control of AUVs [118
]. There are numerous types of AUVs and ROVs available for oil and gas infrastructural monitoring. Examples of commercially available ROVs and AUVs primarily deployed in the oil and gas industry are shown in Figure 7
. The use of unmanned vehicles for pipeline inspection has the advantage of being a remote operating system; making it suitable for inspection in a remote and hazardous environment. Lower cost of maintenance and higher operation safety are also some of the advantages of unmanned vehicles. Unfortunately, these systems also have drawbacks. For example, the cost of purchase or hiring an AUV/ROV is extremely high. Additionally, bad weather conditions such as clouds, winds or other climatological agents can restrict the performance of these vehicles. There are also legal constraints for the use of the unmanned system in some certain areas due to safety concerns because unmanned vehicles usually lack onboard capacity to sense and avoid other AUVs in advance [119
]. However, great effort has been spent on underwater robot sensing and navigation research to realise fully autonomous AUVs for pipeline inspection and monitoring tasks with minimal human intervention. [120
As bolt connections are widely utilised in the assembly of different sections of petroleum pipeline systems, effective technique for monitoring bolted flange connections is essential. Several vision-based assessment methods for real-time bolt looseness detection have been proposed [122
]. Nguyen et al. [125
] proposed a vision-based algorithm to identify bolt-looseness in steel structure bolted flange connections. A similar vision-based monitoring technique for detection of bolted joints looseness in wind turbine tower structures was proposed by Park et al. [126
] which can be adopted to pipeline monitoring in a fairly straightforward fashion. Wang et al. [127
] proposed a new vision-based bolts looseness detection method to address the issues of difficulty in detecting the status of bolt image acquired from any arbitrary perspective and high performance bolt looseness recognition model. The algorithm developed shows high capability to identify the mark on the bolt and bolt position on the flange connection in offline mode. In order to enhance the robustness of this method further online training is required. Similarly, the method should be improved to be able to recognise bolts looseness in a pool of large flag bolts.
5. Performance Comparison of Leak Detection Technologies
This section presents a qualitative performance analysis of various pipeline leak detection approaches based on the literature cited above and American Petroleum Institute (API) performance requirement guidelines [4
]. Various performance criteria are considered for comparison such as system operational cost, sensitivity, accuracy, leak localisation, system mode of operation, ease of usage, leak size estimation, ease of retrofitting and false alarm rate. The analysis is performed using two and three-level performance comparison. In the three-level analysis comparison, the operational cost, sensitivity and false alarm rate are compared in the range of low, medium and high. Figure 10
shows abar chart representing the three-level analysis of the reviewed methods based on their unique strengths and weaknesses. As shown in Figure 10
, most of the techniques have high operational cost except NPW and vapour sampling. However, the high rate of false alarms is the major weakness of these two methods. In general, all methods perform well in terms of sensitivity, except IRT, GPR and NPW. The rate of false alarms in most of the techniques such as acoustic emission, NPW, vapour sampling, dynamic modelling and DSP are high. Though many researchers have been working on alleviating these drawbacks, reducing false alarms in acoustic emission and DSP appears to be a challenging task as acoustic emissions are highly sensitive to random ambient noise and the DSP approach mainly depends on instrument calibration accuracy. Besides, different circumstances such as pipeline corrosion, bending and blockage can easily lead to false alarms in DSP. Among all the reviewed methods, the dynamic modelling method shows high sensitivity in detecting the presence of pipeline leakages. However, the high complexity of the mathematical models involved and strict experienced personnel requirements are the key challenges of this method. With the help of recent advances in high performance computing and cloud computing technologies, the dynamic modelling approach will become more and more popular in the oil and gas industry.
The performances of various pipeline leakage detection methods are next compared using two-level performance analysis. System accuracy, system mode of operation, leak localisation, leak size estimation, ease of usage and ease of retrofitting are the criteria employed to evaluate the performance of the reviewed methods using a yes or no, high or low, and steady or transient state or not applicable (indicated by “—”) scale. Table 3
shows a summary of the comparison. The study shows that none of the methods satisfies all attributes as they all vary in merits and critical shortcomings. For example, the systems based on infrared thermography are proved to be better in terms of system accuracy, leak localisation, easy usage and easy retrofitting, however, estimation of the leakage rate is difficult with this method. Similarly, almost all methods satisfy the ease of retrofitting or upgrading criterion except the fibre optic sensing method, where a point of breakage can lead to total system failure thereby requiring total sensor network replacement. System accuracy is also an important criterion to evaluate the performance of a pipeline leak detection system. Although some of the methods perform better in regards to this criterion, system detection capability also depends on other factors such as instrument calibration, and the quality and quantity of the instruments used.
7. Research Gaps and Open Issues
Based on the various reviewed pipeline leak detection methods, research gaps and future research directions are identified in this section. The performance of pipeline leakage detection methods generally varies depending on the approaches, operational conditions and pipeline networks. However, guidelines set by American Petroleum Institute (API 1555) such as sensitivity, accuracy, reliability and adaptability [92
] must be met before we can consider any leak detection system suitable for production solutions. Moreover, leak localisation and estimation of the leakage rate are also important as they will facilitate spillage containment and maintenance at an early stage to avoid serious damage to the environment. The simplest way to achieve this goal is through deployment of a vast number of leak detection sensors in a sensor network between the upstream and downstream of the pipeline. By doing so, it will easy to isolate the leak position and thus improve the ability to track when a sensor acquires anomalous information at the expense of high implementation cost.
Remote monitoring of oil and gas pipeline networks using wireless communications technology provide benefits of low cost, fast response and the ability to track the locations where leakages occur. However, to attain benchmark performance in monitoring pipelines remotely some of the design issues that require research attention include sensing modality, sensing coverage and leak localisation. As mentioned in the previous sections, several sensors are designed for monitoring pipeline leakages using different sensing modalities. Usually, sensors are deployed for monitoring steady-state conditions where the physical pipeline context is expected to remain stable over time. Variations in physical parameters of the pipeline operation such as vibration, temperature, pressure etc. are expected to be detectable and communicated to reveal the incidences of anomalies. Leaks can only be accurately detected if the incident is within the vicinity of the monitoring sensor and thus the accuracy of leak detection systems becomes questionable if the leaks are not within the receptive fields of the sensors. Sensors deployed for remote monitoring of pipelines are employed to perform both sensing and communication functions, however, the challenge of how to cover a monitoring region efficiently and relay the obtained measurements to their neighbouring nodes is also challenging in wireless sensor networks (WSNs), which impact on the network performance can be severe. There are many issues in designing optimal WSNs, particularly for pipeline monitoring. These issues include: (i) self-organisation, (ii) fault-tolerance, (iii) optimal sensor node placement, (iv) sensor coverage, (v) energy saving routing, (vi) energy harvesting and so on.
During the lifetime of the sensor network some of the deployed sensor nodes are expected to experience hardware failure and the network may not be able to cope with this failure. This will limit the effectiveness of the whole network. The operation and performance of WSNs is largely dependent on optimal node placement as communication among the sensor node is required to transmit the acquired data. Besides, sensor placement also influences the resources management such as energy consumption in WSNs [184
], while the energy consumption influences the network lifetime [185
]. In that case, sensor placement in pipeline monitoring requires further research attention. The development of self-organisation strategies has become an important research issue in WSNs. Sensor nodes are smart enough to autonomously reorganise themselves to share sensing and data transmission tasks when some nodes fail. The issue of coverage problems has been addressed in the literature [186
]. Some of these studies have proposed methods for achieving high sensor coverage [189
], while development of analytical model and optimisation approaches for WSN coverage was proposed in some studies [192
]. However, the development of simple but realistic models for analysis and optimization still remains as a challenging research questions. Since a high percentage of pipeline systems are made up of underground and underwater pipelines networks and the power required for real-time sensing and data communications in such environments is demanding, better replacement of sensor nodes in these settings is expensive or infeasible for large sensor networks. In order to achieve long-lived networks in these energy constrained environments, different energy consumption minimisation methods such as low energy adaptive clustering hierarchy [195
], in-network processing [196
], and sleep mode configuration [197
] have been applied. Energy can also be harvested from the resources in the pipeline surroundings such as fluid flow, pipe vibration, pressure and water kinetics using piezoelectric transducers. Although great improvements have been observed in research and development of wireless sensor network technology, efficient and reliable energy storage and generic plug and play energy harvesters from multiple sources remain open research challenges.
Leak localisation is very essential in pipeline monitoring as it will speed up the repair process. Different methods of defect localisation in pipelines have been proposed [198
]. The performance of these techniques, however, varies in terms of accuracy, degree of complexity and operation environments. Mobile sensor nodes with built-in Global Positioning System (GPS) have been successfully deployed to determine and report the geographical location of pipeline leakages. The use of mobile sensor nodes in pipeline environments is essential as it can enhance coverage and recover the network from any failure which partitions the whole network into multiple disconnected subnetworks. However, the cost of implementation of these sensor nodes with GPS capability is extremely high. Besides, it may be difficult for the GPS signals to penetrate the metal or concrete walls which protect pipelines. If all sensor nodes are static, their locations are marked using GPS and stored permanently in a map in the deployment phase. Leaks can then be localised based on the known locations of reporting sensor nodes. On the contrary, scalability of the pipeline leakage detection sensor network is another research challenge when the coverage of the pipeline network is huge. In this regard, localization techniques with satisfactory performance will be a welcome addition to the leak detection mechanism toolbox. The effect of temperature variation which is a common type of environmental uncertainty, affects the accuracy of flow monitoring systems significantly. Environmental uncertainties can affect the properties of fluids in pipelines such as fluid density, viscosity, friction factor, etc. Although, some studies have provided insights for the development of temperature-dependent flow models [182
], these investigations are only limited to short flow models in which spatial changes of the temperature can be neglected. A robust temperature variation compensation approach will provide additional advantages for fluid flow modeling. It is important to detect the valid leaks and reduce the number of false positive alarms so that pipeline leak detectors can attain acceptable accuracy. All leakage detectors are based on inference based on evidence acquired from sensors [201
]. The input evidence signature is usually noisy or error prone. The noise is in general random in nature and its underlying probability distribution is unknown. The source of the noise comes from inaccurate system measurements, instruments calibration, system modelling, data processing, feature extraction as well as communications. For example, in an acoustic emission leak detection method data acquired using acoustic sensors noise disruption as well as signal attenuation phenomena are usually inherent. In order to reduce the effect of this noise, certain design requirements for signal filtering must be met. Effectiveness of some of the signal filtering algorithms such as Savitzky-Golay, Ensemble, Applet [202
] can lessen the degree of signal distortion to acceptable level. An autonomous system which can detect, locate and quantify the rate of leakage with the capability to manage a large amount of acquired data is essential for planned and unplanned leak incidents. Advanced data visualisation tools will definitely help in showing the state of flow activities for decision making in leak detection, localisation and characterisation, and pipeline maintenance. In addition, data driven self-testing incidents analysis and other offline performance validation methods will also enhance the system flexibility.
The subsea industry activity has been continuously growing, which has made the sector a truly global industry with the industry operations amounting to billions of NOK in turnover [203
]. However, pipeline leakages remain one of the major challenges in this sector [204
] although various efforts have been made to guarantee early detection of leakages in subsea pipelines. In [163
], computational fluid dynamics modelling was devised to describe underwater gas release and dispersion trajectories. The challenges of this approach are that seawaves can easily alter the gas dispersion movements and in the event of large leakages, the gas release rate and dispersion trajectory could be arbitrary. The mechanistic modelling of detection pipeline leaks at a fixed inlet rate presented in [24
] provides insights for monitoring hydrocarbon parameters. However, the algorithm is limited because the external conditions that can easily lead to subsea pipeline instability in a subsea environment were not taken into consideration. Updated information about the internal flow conditions as well as pipeline integrity that are independent of the weather and sea conditions is needed for further innovation in this area. Moreover, experiment leak scenarios as a function of leak opening size in the laboratory and data processing in a way suitable to establish signals indicating hydrocarbon spillage will provide benefits in designing a functional basis for leak detection.
In general, the aim of future pipeline monitoring is to design a real-time intelligent pipeline leak detection and localisation system for subsea pipeline networks. The effect of environmental factors, in particular, hydrodynamic forces due to oblique wave and current loading on subsea pipelines still require further research study. Extensive simulation and laboratory experiments are being conducted to study the effects of leakage parameters, like size and shape, on the flow mechanism and validate different models. Numerical simulations of fluid flow in pipeline using computational fluid dynamics (CFD) have been proved to provide a better understanding of pipeline internal flow and the conditions of pipeline leaks on various scales, thereby reducing the cost in experimental studies. However, high computational complexity remains one of the major drawbacks of CFD. Further research efforts are still required to optimise and/or parallelise CFD solution algorithms in terms of computation and memory resource constraints.