Industries / Mining and materials / Mining and quarrying / Coal mining
Roof support and ventilation
This work keeps the rock above and beside miners from falling and moves fresh air through the workings to carry away gas, dust and heat.
7
Milestones achieved, from 1858 to 2007.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1858
The Guibal fan
Mines were long ventilated by furnaces: a fire at the foot of one shaft heated the air so that it rose, drawing fresh air down another. Théophile Guibal, a French engineer working in Belgium, experimented with a large centrifugal fan from 1855 and patented it in 1858. One 1875 estimate counted 180 Guibal fans in England, 85 in Belgium, 60 in France and 30 in Germany, and it was the most common fan in the Ruhr region from 1872 to 1890, by the account of economic historians John Murray and Javier Silvestre.
LessonFurnaces stayed in use for decades after fans arrived, even in gassy mines, because they were cheap, reliable and had no moving parts.
Credit
Théophile Guibal
1950s
Roof bolts replace timber
Roof bolts are steel rods set into holes drilled up into the roof to hold the rock layers together. US coal mines took them up in the late 1940s and 1950s faster than any earlier mining technology, with the US Bureau of Mines as their strongest advocate. Yet by the end of the 1950s the rate of deaths from roof falls had risen, and it took another decade before bolts clearly saved lives, by the account of NIOSH researcher Christopher Mark.
LessonFast adoption of a better support did not by itself make mines safer.
Credit
US Bureau of Mines (promotion); US coal operators
1960
The Coalbrook collapse
On 21 January 1960 the roof caved in at the Clydesdale Coalbrook North Colliery near Sasolburg, South Africa, and the central part of that section of the mine collapsed. The collapse damaged the main fan and cut off the fresh air supply, and 435 men died, by the Minerals Council South Africa's count. In 1963 the Chamber of Mines set up a research body, the Chamber of Mines Research Organisation (COMRO), as a direct result.
LessonPillar design in the collieries of that era rested largely on trial and error, and Coalbrook showed the cost.
Credit
Chamber of Mines of South Africa (research response)
1967
A formula for pillars
Pillars are blocks of coal left standing to hold up the roof. After Coalbrook, M.D.G. Salamon and A.H. Munro worked out a formula for pillar strength statistically from 27 cases of failed pillars and from pillars that stood, which led South African coal mining to design pillars with a safety factor, the ratio of a pillar's strength to the load on it. Pillars kept failing after the formula came in, and by 2011 the database held 86 failed cases and 337 stable ones.
LessonA design rule drawn from 27 failures has to be re-checked as new failures come in.
Credit
M.D.G. Salamon and A.H. Munro
1969
Roof plans for every mine
From 1900 to 1968 the US rate of deaths from falls of roof stayed between about 0.6 and 1.0 per million employee hours. In the eight years after the Federal Coal Mine Health and Safety Act of 1969 it fell to about 0.15, a 75% cut. NIOSH researchers attribute most of the drop not to new technology but to the requirement that essentially all roofs be supported according to a mandated roof control plan.
LessonApplying known methods everywhere did more than new technology, by NIOSH's analysis.
Credit
US Congress; federal mine inspectors
1990s
Pillar design from case histories
NIOSH developed ARMPS (Analysis of Retreat Mining Pillar Stability) in the mid-1990s. In retreat mining, miners cut out the pillars as they pull back, and ARMPS suggests safe pillar sizes from a database of real mining cases. NIOSH researchers report that after it came into wide use, pillar squeezes fell and massive pillar collapses were largely eliminated, and MSHA used it in its own investigations.
Credit
NIOSH (Christopher Mark and Frank Chase)
2007
Crandall Canyon
On 6 August 2007 pillars collapsed across a wide area of the Crandall Canyon mine in Utah, trapping six miners, who died. On 16 August a coal outburst, a violent ejection of coal from a stressed pillar, killed two miners and an MSHA inspector working on the rescue. MSHA found the collapse "was not a 'natural' earthquake" but came from a flawed mine design, that the consultant's ARMPS analysis was "inappropriately applied", and that its results for the section being mined were left out of the reports given to MSHA. An independent review found that MSHA's own failure to adequately evaluate the roof control plans contributed.
LessonA design tool protects miners only if it is applied to the right ground and someone checks the result.
Credit
MSHA (investigation); MSHA Independent Review Team
Watch for
AI adviser for coal bursts
A coal burst, or rockburst, is a sudden, violent failure of highly stressed coal or rock. Researchers fine-tuned a large language model on burst knowledge and regulations, linked it to microseismic monitoring (records of the small tremors rock gives off as it cracks), and had it write daily burst-risk reports. They report 93.01% accuracy on their own four-part test set and a field deployment at a burst-prone face in Inner Mongolia.
Achieved whensomeone outside the team scores its daily warnings against recorded bursts and large tremors at more than one mine over several months.
Claimed by
Cao Anye and colleagues, Journal of China Coal Society, April 2026
Open
01
Public scores for burst warnings
A public test set for coal burst warnings: years of seismic records from coalfields where bursts happen, with each recorded burst marked by date and place, so that any warning method can be scored on years it never saw.
The briefCredit · unclaimed02
Second reader for pillar plans
Software that reads a mine's pillar design submission, reruns the stability check with NIOSH's ARMPS program, and lists anything missing or below NIOSH's guidelines, for a human plan reviewer to look at.
The briefCredit · unclaimed
Open 01 · the brief
Public scores for burst warnings
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test set for coal burst warnings: years of seismic records from coalfields where bursts happen, with each recorded burst marked by date and place, so that any warning method can be scored on years it never saw.
Why it is still open
Published warning methods are scored on their authors' own data. The 2026 burst adviser above reports accuracy on its own test set, and a 2026 paper on predicting rockbursts from microseismic records at a mine in Xinjiang has since been retracted. A 2026 review of AI burst warning by Cao Anye and colleagues names "the lack of standardized benchmark databases" as a common challenge. Searches for this brief found case studies of single mines but no shared test set for coal mines.
Built with what exists
Public earthquake catalogs, such as the US Geological Survey's ComCat and the regional catalog of the University of Utah Seismograph Stations, whose records MSHA used at Crandall Canyon; MSHA accident investigation reports for dated bursts; Python with the ObsPy seismology library.
First step
Download the earthquake catalog for the central Utah coalfields for 2000 to 2025 and mark the dates of the 2007 Crandall Canyon collapse and outburst.
Done looks like
Ten or more years of seismic records for at least one coalfield, with every MSHA-reported burst in that period marked, and scores for at least two warning methods on held-out years, given as the share of bursts warned of in advance and false alarms per month.
Open 02 · the brief
Second reader for pillar plans
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Software that reads a mine's pillar design submission, reruns the stability check with NIOSH's ARMPS program, and lists anything missing or below NIOSH's guidelines, for a human plan reviewer to look at.
Why it is still open
At Crandall Canyon, MSHA's independent review found that the district's ARMPS analysis "did not examine barrier pillar stability in any way", that no independent analysis was ever done for the section where the miners died, and that reviewers were not trained in LaModel, another pillar program used to justify the plan.
Built with what exists
NIOSH's ARMPS pillar design program; Claude or Gemini to read plan documents and pull out pillar sizes, depths and stability factors (see the Note "Seven more flagged", on a Claude-based reviewer of drug-making records); Python.
First step
From the Crandall Canyon report and its appendices, list every pillar design input and stability factor the investigators checked, to serve as the answer key for the first test case.
Done looks like
Run on the design documents in at least five MSHA investigation reports into pillar failures, it flags every design shortfall the investigators named, with no more than two false flags per report.