Accurately alarming a pipeline leak


For years now we’ve been able to monitor tank levels, compressors, pump operations and flow meters. Most of that is pretty cut-and-dried. The tank is at whatever level it’s at; the pump is either running or it’s not. An alarm tells us what we need to know the minute we need to know it.

But detecting a pipeline leak involves combining a number of data points and deciding which ones, or which combination of points, means there really is a leak. I’m stressing “really” because the industry first believed any anomaly needed to be alarmed because at that point an anomaly equaled leak, and we defiantly know the faster that leak could be stopped, and repaired; the more profits from your production ended up staying in your pocket.


Here are two examples of how that wasn’t as easy as it sounded:

You might assume that a pressure variation anywhere in the pipeline would indicate a leak. Or that the absence of any liquid means it all leaked out at that spot. So the system should send an alarm to appropriate people to travel to the site, check it out, and get it fixed.


The trouble is, pipelines are long and pressure fluctuations around bends and on hills are common and normal. Sometimes, when a pump stops, the liquid is siphoned off by suction downstream. Some injection wells goes to vacuum, and the biggest problem is inaccurate metering, from plug or wash turbines to bad calibrations. You can imagine how many false alarms that could create. Anything more than a few in a month is a huge problem.

That’s when alarms start to get ignored.

Just like those robo-calls that flood your phone, or annoying spam emails, or mass text messages that we all get, and that we all ignore. Which means a real leak may not get noticed for a long time, eliminating any benefit from the alarm system all together.

Lots of companies who experience leaks said that the system alarmed but its looks just like another false alarm. We’ve come a long way toward fixing that—and making alarm systems react accurately.

We recently tested our new pipeline leak detection system, that's part of the new Zedi Smart-Field™ solution that is our autonomous wellsite production suite, on an SWD pipeline. Here’s what we learned, and how we’ve eliminated almost all false alarms.

It’s important to note that SWD pipeline leaks can do tens of thousands of dollars in damage to nearby land—ranches, government lands, and roadways. It can lead to fines, mandated shutdowns for cleanup and repairs, which can stop production, and instantly diminish your time and profits. There are strong economic and plenty of safety reasons to find and fix leaks like this as quickly as possible.

Our client had been experiencing more than one false alarm per day, leading to some actual true leak alarms being ignored.

The company let us monitor a test section of their SWD pipe, where they had eyes on site and knew whether a leak was actually happening. When our sensors detected an anomaly in pressure or flow rate the operators would “instruct” the machine-learning algorithm as to whether or not this involved a true leak.

With every instance the algorithm would get “smarter” (that was directed by the human operator). To test our system, a Zedi customer simulated a “leak” of 5%, 2,5% and 1% of the flow—and all three “leaks” Zedi Smart-Field™ Pipeline Leak Detection made the right call.

Time to celebrate, right?


That is indeed a huge step. But we’re not stopping there. We’re developing more algorithms and more software to reduce even further the false alarms. We’re making the system easily navigable by all end users. That’s where it goes from helpful, accurate and effective—to when it’s completely accurate, and so easy to use, and so “Wow” that the interface becomes transparent. It’s just a normal part of your day as; “This is how we get our pipeline information.” That’s our goal, and we’re getting closer every day.



Topics: Safe, Pipeline Leak Detection, Alarms, AI, ML

Subscribe to get new blog posts >

Contact us >

Recent Posts