Q1) How have monitoring technologies like SCADA based systems and acoustic sensors been improved with the emergence of AI?
No significant improvement has been seen in SCADA based systems and acoustic sensors in their ability to detect and locate theft events with the emergency of AI. Fundamentally it is caused by their own limitations. While SCADA data can support leak detection systems, it often samples at low rates and relies on widely spaced pressure and flow meters, so it struggles to pick up the subtle or intermittent signatures of theft. Even with faster sampling, SCADA data lacks the accuracy and resolution for identifying the thefts that are kept at such a low rate that they are below the flow meter accuracy level (eg 0.5%).
Acoustic methods offer the identification of sudden events, but the signal dissipates quickly along the line. This means they require a high density of instrumentation to provide reliable coverage, which is not always practical. Unlike leaks, a controlled theft operation may not generate the acoustic signature that the acoustic leak detection method is designed to detect.
To apply AI for effective theft detection, it is necessary to have the quality data available so advanced data analytics can be applied.
The most effective theft detection system starts with high quality data, combining advanced data analysis with different detection algorithms. For example, by fusing negative pressure wave detection with statistical volume balance, operators gain both redundancy and precision. This allows tapping points to be pinpointed to within meters and thefts to be recognized even when they fall below normal flowmeter accuracy or are masked by process noise. Portable, GPS-synchronized units and non-intrusive sensors further extend monitoring into remote areas where conventional infrastructure is limited, ensuring continuous visibility along the right-of-way.
Q2) With evolving measures to prevent illegal tapping, how have the methods of the thieves changed?
Oil Theft
Fuel thieves now operate with two distinct approaches. The first is the “try and remain hidden” method, which involves slow valve operations, long hoses of up to 1,500 meters and theft volumes kept below flowmeter accuracy levels. These taps are sometimes left in place for years, occasionally with double tapping points and are often used at night or operated remotely. Such methods are common among organized criminal groups, supported by corrupt officials or insiders, and are deliberately designed to avoid detection by modern monitoring systems. Recent “try and remain hidden” tactics we’ve seen deployed by thieves include the digging of tunnels that lead to the pipeline and remotely operated theft valves.
In contrast, the “slash and grab” approach is more opportunistic, characterized by crude tampering, larger taps, poor installations, or angle grinding pipelines. This technique, while less sophisticated, carries a higher risk of rupture and catastrophic failure. Although less subtle, it continues to emerge in regions where theft is driven by basic needs rather than organized crime.
Water Theft
In water systems, theft has also evolved, driven by scarcity and rising costs. At one end of the spectrum is the “slash and grab” approach, where thieves break into pipelines to extract water quickly, often leaving the line damaged and sometimes causing catastrophic failure. We’ve seen this across parts of Asia and the US.
At the other end there are more organized and hidden operations. These include carefully installed illegal connections that mimic normal operations, or large-scale schemes involving construction equipment, counterfeit uniforms and vehicles. Cases include organized crime groups in Latin America selling stolen water and in Oceania in response to drought. These examples show how theft is moving from opportunistic damage to carefully planned, professionalized activity, requiring equally advanced monitoring and detection responses.
Q3) How is satellite imagery, specifically Synthetic Aperture Radar (SAR) and high-resolution optical imagery, being used to identify suspicious ground activity near a pipeline right-of- way? What are the technical limitations and advantages of using satellite technology for theft detection compared to on-the-ground sensors?
Technology is key to the success of fuel theft detection to provide pipeline operators with the data they need and timely alerts when thefts occur. Satellite imagery offers an effective external layer of defense. Synthetic aperture radar (SAR) can detect subtle ground disturbances, vehicle tracks or excavation near a right-of-way, while high-resolution optical images provide visual confirmation of changes in terrain, vegetation cover or new structures that may suggest tampering. This capability is particularly valuable in remote or politically unstable regions where patrols are limited.
However, satellite data has limitations. Cloud cover, vegetation and irregular orbital passes can obscure or delay visibility. Updates are periodic rather than continuous, which means short-lived theft events may not be captured. For this reason, satellite imagery should be viewed as complementary rather than standalone. It is most effective as part of a multilayered approach that includes hardware attached to the pipeline (such as Atmos Eclipse and Atmos Odin), real-time monitoring software (Atmos Wave Flow), and offline expert analysis through services like Atmos Theft Net. Together, these layers provide both continuous internal monitoring and broad external oversight.
Q4) What can be done to prevent cyber-attacks on SCADA systems and how big is the aspect of cybersecurity in the development of new systems?
A lot can be done to prevent cyber-attacks on SCADA systems, there are government and industry guidelines on SCADA cybersecurity. Cyber security risks increase as pipeline companies adopt more cloud based solutions due to internet connectivity.
To move forward with technology while staying cybersecure, Atmos has integrated ISO/IEC standards into our cybersecurity framework. We have developed a service that facilitates secure data transfer in one direction only to prevent any data input to the SCADA environment. Additionally, the leak and theft detection data are encrypted during transmission using secure sockets layer (SSL) and transport layer security (TLS) to provide secure communication across the network.
Q5) What are the challenges in synchronizing and interpreting data from different detection systems with varying sample rates and measurement principles?
One of the key challenges in pipeline theft detection is that no single method works effectively under all operating conditions. Different systems operate at different sampling speeds and use different measurement principles, making synchronization and interpretation complex. For example, mass balance methods rely on flow and pressure meter data at SCADA sampling rates, usually less than 1 Hz, while negative pressure wave methods depend on high-resolution pressure data at 60 Hz. Aligning these datasets so they can be compared in real time requires advanced algorithms to handle timing differences, data gaps and varying sensitivities.
Atmos Wave Flow addresses this by merging multiple detection technologies. Its flow balance element uses real-time transient modeling to correct for instrumentation errors and meter drifts, while its negative pressure wave element applies three data analysis algorithms to separate genuine leak or theft events from operational transients. By fusing these signals into a single system, operators gain faster detection, more accurate location and fewer false alarms across a range of pipeline conditions.
Ultimately, the challenge is not just about synchronizing different sample rates, but also about interpreting different “languages” of data. Combining methods in one system ensures that subtle anomalies aren’t dismissed as noise, while still avoiding the false positives that can occur when methods are used in isolation.
Q6) What are the technical considerations for strategically placing sensors along a pipeline to maximize the probability of detecting a tap? Is there an advantage using point sensors compared to continuous sensors and vice versa?
Effective theft detection relies on more than just sensor placement. Sensors must capture high-frequency data to detect subtle anomalies in real time. Non-intrusive devices like Atmos Eclipse provide flow, pressure, and temperature monitoring in high-consequence areas, while portable units like Atmos Odin log 60 Hz pressure data with GPS synchronization, giving accurate coverage in remote terrain where SCADA is limited.
Strategic spacing is critical for methods such as negative pressure wave detection, which use triangulation to pinpoint tapping points to within meters. If it is difficult to find a tapping point, the placing of an additional portable unit would enable the localization of the tapping point. Combining this with flow balance analysis ensures both sudden and slow withdrawals are captured with precise location estimates.
Point sensors deliver precise, localized data and are cost-effective at hot spots, but the coverage is limited. Continuous systems using pressure sensors offer wide-area monitoring and are effective against third-party interference. A layered approach, using both, provides the best balance of precision and coverage.
Q7) What are the primary sources of false positives in AI-based pipeline detection systems and how can machine learning help with reducing them?
The main sources of false positives in AI-based pipeline leak/theft detection systems include the inability to distinguish operational changes from leaks or thefts, lack of high-quality data, data faults due to loss of sensors or communications and poor algorithms used by AI.
It is impossible for AI or machine learning to improve the performance of theft detection if no quality data is available. Since most thieves are very skilled in manipulating their theft operations to keep the theft rate low to avoid detection, high frequency and high resolution data are essential. In addition, every pipeline has its own characteristics with varying responses to thefts, so it’s essential to develop the machine learning algorithms to cover pipeline operations eg pump start/stop, valve open/close, fluid property changes, instrument faults, pigging and pipeline shutdown. Finally, the algorithms must be able to identify the small changes caused by a theft.
Got more questions about Fuel Theft & Illegal Tapping? Want to speak to an expert at Atmos International or know more about our solutions? Contact us here.
The Expert
Harry Smith, Sales and Senior Research Engineer, Atmos International
Harry holds a dual position within Atmos of Sales Engineer, specialising in the UK and Europe, and Senior Research Engineer for some of Atmos’ key services. Within his role Harry has a unique position of being customer facing to better understand the needs and challenges of pipeline operators and EPCs and being able to interpret them within the research & development team. Since joining in 2014 Harry has played an essential role in the project management, scoping and delivery of multiple pipeline leak detection systems.
