Industries / Energy, water and waste / Pipelines and midstream
Inspecting from inside the pipe
In-line inspection sends a tool called a pig, a plug pushed along by the flow, through the pipe to measure the wall for corrosion, dents and cracks.
5
Milestones achieved, from 1964 to 2026.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1964
The first smart pig
The first commercial smart in-line inspection tool, one carrying sensors rather than just scraping, was built in 1964, according to a 2017 history by DNV GL engineer Matt Ellinger. It used magnetic flux leakage: the tool magnetizes the steel, and where metal is lost, the field leaks out and sensors pick it up. High-resolution tools followed in the 1970s.
Credit
the history names no builder (see Checks)
1988
Sound measures the wall
The first ultrasonic pig run in Elf Aquitaine's 20-year record (the tool times sound echoes to measure wall thickness) was a Pipetronix Ultra Scan on a 17-km water line at the Lacq gas field in 1988. Elf's engineers wrote that the tool "was not fully operational" and needed three runs to size the pits. They also found oil lines sometimes needed 10 to 15 cleaning runs before inspection.
Lessona new tool needs several runs and a clean line before its numbers can be trusted.
Credit
Pipetronix; Elf Aquitaine
2010
Cracks found, left in
Enbridge's Line 6B ruptured at Marshall, Michigan, on 25 July 2010 and released about 843,444 gallons of crude. The NTSB found that Enbridge relied on one crack-inspection tool and did not allow for its measurement error. Six crack-like defects 9.3 to 51.6 inches long, assessed in 2005, were left unrepaired until the rupture.
Lessonevery measurement has an error band, and repair rules must allow for it.
Credit
Enbridge (operator); NTSB (investigation)
2012
Crack tools checked
For a PHMSA-funded program, Kiefner & Associates and DNV compared 13 crack-tool runs over 741 miles of electric resistance welded pipe (pipe whose seam was welded with electric current) against digs, pressure tests and burst tests. In none of the 13 cases would the team say the inspection gave full confidence. In three, tools missed defects larger than their stated detection limits, and in others depth estimates differed by up to 2 to 1.
Lessontools must be checked against cut-out pipe, not only field measurements.
Credit
J. F. Kiefner and colleagues, for Battelle
2026
A public inspection dataset
PipeMFL-240K holds 249,320 magnetic flux leakage images from 12 pipelines, about 1,530 km in all, with 200,020 labelled boxes in 12 categories. Its authors call it the first public dataset and benchmark of this scale for pipeline inspection, and it was accepted by the KDD 2026 datasets track after review. It labels where features are, not how deep.
Credit
Tianyi Qu and co-authors
Watch for
Inspection without a pig
ROSEN says its NIPA service judges the condition of pipes that cannot take a pig by combining surface surveys of the coating and corrosion protection, magnetic readings taken from a distance, and AI predictions drawn from data on more than 26,000 past inspections. It says NIPA covered more than 700 km in 2025 and found serious metal loss that needed repair, and it won a Gas Industry Award. No accuracy figures are given.
Achieved whenan operator or outside party publishes predictions compared against digs or later pig runs, with error figures.
Claimed by
ROSEN, May 2026
Open
01
Size the defects too
A public benchmark of simulated magnetic flux leakage signals from metal-loss defects of known depth, length and width, and a baseline model that estimates depth from the signal.
The briefCredit · unclaimed02
Update the 2012 review
A public table of pipeline failures since 2012 where an in-line inspection ran before the failure, saying which tool ran, what it reported, and what was found afterward.
The briefCredit · unclaimed
Open 01 · the brief
Size the defects too
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public benchmark of simulated magnetic flux leakage signals from metal-loss defects of known depth, length and width, and a baseline model that estimates depth from the signal.
Why it is still open
The largest public dataset, PipeMFL-240K, labels where features are but not their size, and its authors say the field relied on private data. The 2012 PHMSA-funded review found tool and field depth estimates differing by up to 2 to 1.
Built with what exists
Open-source finite-element software, which solves physics equations over a mesh, such as FEniCSx or Elmer; PyTorch; PipeMFL-240K images as a check on realism; one graphics card.
First step
Simulate one defect in a steel plate and compare the signal shape with a published simulation.
Done looks like
At least 10,000 simulated defects with true sizes; a baseline whose depth error is reported as the share within 10% of wall thickness; a second group reproduces the score to within one percentage point.
Open 02 · the brief
Update the 2012 review
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of pipeline failures since 2012 where an in-line inspection ran before the failure, saying which tool ran, what it reported, and what was found afterward.
Why it is still open
The 2012 review covered 13 crack-tool runs on one kind of pipe. No later public compilation across tool types and threats was found.
Built with what exists
NTSB pipeline reports, PHMSA failure investigation reports and Canada's Transportation Safety Board pipeline reports; an AI model that cites the page behind each fact it extracts (Note: "Findings that cite notes"); a spreadsheet.
First step
List all NTSB pipeline accident reports since 2012 and mark those that mention an in-line inspection.
Done looks like
Every qualifying report from the three sources since 2012 in one table, each field linked to a page; a second person checks 20 reports with at most one disagreement per report.