Artificial Intelligence and Machine Learning Reshaping Today’s Pipeline Leak Detection Systems

By Jyoti Prakash

Category:
Industry Trends

While operating companies in upstream, midstream, and downstream oil & gas have been utilizing leak detection systems (LDS) for decades; new regulations, rising concerns about pipeline integrity and cybersecurity, and the availability of several powerful new technologies have spurred increased interest in these solutions among end users.  This bodes well for the overall market in the coming years.  Today’s increasingly software-based leak detection systems incorporate advanced automation plus new, disruptive technologies such as predictive analytics, artificial intelligence, and machine learning to improve pipeline safety, availability, and security.

Artificial Intelligence (AI) and Machine Learning (ML)

Today’s smart, connected devices push out more information than ever before, but this massive amount of data is often too much for control systems or humans to handle.  That’s why artificial intelligence (AI) will soon be embedded in equipment.  This might happen in three phases:

  1. Training the AI engines and creating the algorithms
  2. Teaching the machines what to do with the information, and
  3. Tiering decisions so they can look at information and provide guidance for operators 

Software that enables predictive maintenance and proactive monitoring through ML for safety will become the priorities, while SCADA software will become less important.  It’s one of the reasons pipeline managers are beginning to question why they need to spend millions of dollars every few years to upgrade their SCADA systems.

Many existing hardware-based leak detection systems rely heavily on existing instrumentation and SCADA systems to monitor pipelines and determine leaks.  Here, machine learning (ML) can also be employed to analyze patterns and help predict and prevent pipeline leaks and ruptures, improving both safety and profitability.  Pipeline operators have been using real-time data analysis (RTDA) for years, but this typically results in vast amount of clustered data over the length of the pipeline.  Advanced analytics and algorithms improve the ability to understand the “how, when, and what” about potential (and actual leaks) in any given condition.

leak detection systems leveraging machine learning
Typical Machine Learning Architecture for Leak Detection Systems

 

Predictive Analytics and Maintenance

By using software to look for trends in the data, operators can also predict breakdowns before they happen, often many months before if historical data is being analyzed.  Analytics are still in their infancy and it’s an area where algorithms need to be improved.  Algorithms have been used in software-based LDS for years, as they have the potential to help identify the patterns that could predict future leaks.  Instrumentation for pipelines is focused on production equipment and the flow of product because they impact revenue and top-line growth.  Analytics are being used primarily for condition-based maintenance, using things like failure models, time in service and time to failure.  Rules and models are being applied to sensor data for up-to-the-minute maintenance applications.  This contributes to safer operations and there’s also great potential in applying sensor-based analytics for condition-based leak detection.

Recent ARC research into the market for leak detection systems discusses several software-based leak detection technologies such as mass/volume balance, negative pressure wave, statistical analysis, RTTM, and E-RTTM, etc. for onshore, offshore, and subsea pipeline applications. 

As the industry is pondering about how to make the most out of RTDA and the vast information gathered, ARC invites suppliers and end users to share viewpoints on their operational challenges and advancements in AI and ML.  For further discussion or to provide any feedback on this blogpost, please contact the author, Jyoti Prakash, at jprakash@arcweb.com.

Engage with ARC Advisory Group