Industries / Energy, water and waste / Oil and gas extraction / Oilfield services
Casing and cement
Casing is the steel pipe set inside a drilled hole, and cementing fills the gap between that pipe and the rock so that oil, gas and water stay in their own layers.
4
Milestones achieved, from 1919 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1919
Cement shuts out water
Erle P. Halliburton had learned oil-well cementing at the Perkins Oil Well Cementing Company in California. In 1919 he started his own company, which became the Halliburton Oil Well Cementing Company in 1920, based in Wilson, Oklahoma, according to the Oklahoma Historical Society. Cementing pumps cement down the casing and up the outside of it, where it sets into a seal. In his 1920 patent application Halliburton wrote that "water has caused the abandonment of many wells which would have developed a profitable output." The patent, for a "Method and Means for Cementing an Oil Well," was granted in 1921.
Credit
Erle P. Halliburton
2010
Unstable foam cement
For the final casing of BP's Macondo well, BP and Halliburton chose "nitrogen foam cement," slurry filled with tiny nitrogen bubbles that lightened it from about 16.7 to 14.5 pounds per gallon so it would not break the fragile rock. Halliburton's two February lab tests and its first April test all showed the foam was unstable, meaning bubbles could merge and leave the set cement porous. The result of a final test was sent to BP on 26 April 2010, six days after the blowout, and may not have been in hand when the cement was pumped. Tests run by Chevron for the presidential commission "strongly suggest" the slurry was unstable. BP also skipped a cement evaluation log, relying on full mud returns.
Lessona failed lab test must stop a cement job, and fluid returning at surface does not show where the cement went.
Credit
National Commission on the BP Deepwater Horizon Oil Spill and Offshore Drilling; Chevron (independent tests)
2014
Counting cracked well seals
Anthony Ingraffea and colleagues read 75,505 state inspection reports covering 41,381 oil and gas wells drilled in Pennsylvania from 2000 to 2012. Statewide, shale gas wells showed cement or casing problems six times as often as conventional wells. For wells drilled after 2009, the risk was 1.57 times higher for shale gas wells than for conventional wells of the same age, and wells in the state's northeast carried 8.5 times the cumulative risk of wells elsewhere.
Credit
Anthony Ingraffea, Martin Wells, Renee Santoro, Seth Shonkoff
2021
Leak tests made public
Greg Lackey and colleagues, working with the US Department of Energy's National Energy Technology Laboratory, searched 33 state regulators' databases and compiled 474,621 well test records from 105,031 wells in Colorado, New Mexico and Pennsylvania. In Pennsylvania 14.1% of wells tested before 2018 showed sustained casing pressure or casing vent flow, signs that gas is leaking into the space between casing strings. Regions of Colorado and New Mexico ranged from 0.3% to 26.5%, and directional wells were more likely to leak than vertical ones in Colorado and Pennsylvania. The laboratory published the dataset in 2022.
Credit
Greg Lackey and co-authors; National Energy Technology Laboratory
Watch for
Cements that withstand CO2
Engineers at ProPetro Services and Texas Tech University soaked samples of a standard class H cement, a resin-cement blend and a fly-ash geopolymer (a binder made without portland cement) in supercritical CO2 at 1,500 psi and 120 °F for 56 days. They report that porosity changed little and peak strength changed by less than 10% in all three, and say the materials could suit wells that store CO2.
Achieved whena laboratory outside the team repeats the result over longer exposures, or wells cemented with these materials pass cement logs after years of CO2 injection.
Claimed by
Athar Hussain and co-authors (ProPetro Services and Texas Tech University), April 2026
Gas wells reused for CO2
The Porthos project converted four former gas wells at the P18-A platform in the Dutch North Sea into CO2 injection wells, with TAQA as main contractor, and sealed and removed three others. Ultrasonic tools lowered on a cable checked the steel and cement, and pipe worn by about 30 years of use was replaced. Porthos says "the steel and cement of all wells are in excellent condition." First injection was planned for autumn 2026; Porthos now says the system is not expected to operate before the second half of 2027.
Achieved whenCO2 has been injected and the Dutch regulator or another outside party reports that the wells hold it without integrity failures.
Claimed by
Porthos, as published by Gasunie, February 2026
Open
01
An open cement test set
A public table of laboratory results for well cement, drawn from published papers and theses: the recipe (cement class, water ratio, additives), the test temperature and pressure, the thickening time (how long the slurry stays pumpable) and the strength gained over time. Add a baseline model that predicts the results, scored only on papers it was not trained on.
The briefCredit · unclaimed02
Recount Pennsylvania's well seals
A rerunnable count of casing and cement problems in Pennsylvania oil and gas wells drilled from 2000 to 2025, built from the state environment department's public well inventory and inspection records, with a language model sorting each inspector's comment and people checking a sample. It would show whether failure rates changed after 2012.
The briefCredit · unclaimed
Open 01 · the brief
An open cement test set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of laboratory results for well cement, drawn from published papers and theses: the recipe (cement class, water ratio, additives), the test temperature and pressure, the thickening time (how long the slurry stays pumpable) and the strength gained over time. Add a baseline model that predicts the results, scored only on papers it was not trained on.
Why it is still open
Machine-learning work on well cement runs on data nobody can check. A software firm's January 2026 case study predicted thickening time and strength for an unnamed oilfield services company from that company's own lab records. A July 2026 Texas A&M University at Qatar study modeled 1,119 cement samples from its own lab and does not say where the data can be found. A June 2026 cement-additive predictor on GitHub anonymized its data "for confidentiality."
Built with what exists
Vision-language models that read tables and charts in papers, including open-weight models that run on your own computer (see the Note "A trillion open parameters"); the test methods in API Recommended Practice 10B-2 to define each field; Python with scikit-learn or XGBoost for the baseline; two people to check the extracted values.
First step
Take 20 open-access papers on class G or H cement that report thickening time, extract their tables both by hand and with a model, and count the disagreements.
Done looks like
At least 1,500 test results from at least 100 published sources, a hand check of a random 10% showing at least 95% of values extracted correctly, and baseline thickening-time errors reported on held-out papers, all published under an open license.
Open 02 · the brief
Recount Pennsylvania's well seals
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A rerunnable count of casing and cement problems in Pennsylvania oil and gas wells drilled from 2000 to 2025, built from the state environment department's public well inventory and inspection records, with a language model sorting each inspector's comment and people checking a sample. It would show whether failure rates changed after 2012.
Why it is still open
Ingraffea and colleagues' count, built from the same department's spud and compliance reports, stops at wells drilled in 2012. The three-state dataset of Lackey and colleagues covers Pennsylvania wells tested before 2018 and uses pressure tests, not inspectors' notes.
Built with what exists
Pennsylvania Department of Environmental Protection spud and compliance reports; Ingraffea and colleagues' published method as the reference; a low-cost language model for sorting many short comments (see the Note "The jobs agents repeat"); two human coders; Python with a survival-analysis library such as lifelines.
First step
Download the compliance records for one northeastern county's wells drilled from 2010 to 2012, code 100 inspector comments by hand, and compare the rate with the published one for that region.
Done looks like
Run on wells drilled from 2000 to 2012, the tool reproduces the 2014 study's rates within its confidence intervals; it then reports yearly problem rates for wells drilled from 2013 to 2025, with the coded comments, the code and the two coders' agreement (Cohen's kappa) published.