Industries / Mining and materials / Glass and ceramics / Refractories
Wear monitoring
Wear monitoring measures how much lining is left in a working vessel, from outside or between batches, and predicts when it must be repaired before molten metal or hot gas reaches the steel shell.
5
Milestones achieved, from 1963 to 2022.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1963
Scanners watch kiln shells
A rotary kiln is a long, slightly tilted steel tube, lined with brick, that turns slowly while cement or lime raw material passes through it; a hot spot on its shell shows where the lining inside has thinned or fallen out. In October 1963 J. H. Meyer and J. P. Puckett of the Leeds & Northrup Company in Philadelphia described a kiln-shell temperature scanner that detects hot spots "when they start to form," so that operators can act "before the fault becomes serious." In 1990 J. P. Kemmerer told the cement industry's IEEE conference that a high-speed shell scanner tied to a database of each refractory installation lets shell temperatures be read against which brick was put where.
Credit
J. H. Meyer and J. P. Puckett (Leeds & Northrup); J. P. Kemmerer
1974
A laser ranges the lining
AGA AB's patent, first filed in November 1974, describes how converter linings were judged before: by eye, which was quick but inaccurate; by stereophotogrammetry, accurate to perhaps 30 to 40 mm but with results only after a day or more; or by radioactive wires set into the lining, whose loss was read from samples of the charge. AGA's answer was a modulated laser rangefinder aimed from outside the hot vessel, with a target accuracy of about 5 mm. In 2008 China Steel Corporation reported measuring blast furnace lining thickness with a 3D laser scanner through an opening in the upper furnace, and in 2025 RHI Magnesita called laser scanning the "benchmarking methodology" for runner wear.
Credit
Ragnar Scholdström, Holger Marcus and Lennart Nordström (AGA AB); Shih-Kang Kuo, Wen-Chieh Lee and Shan-Wen Du (China Steel)
1994
Breakout below a taphole
The hearth is the bottom of a blast furnace, where molten iron and slag collect before they are drained through tapholes. On 5 May 1994, after making 9.5 million tonnes of iron, Dofasco's No. 4 blast furnace in Hamilton, Ontario, broke out 250 mm below its west taphole; the spilled iron started a fire and the furnace lost 33 days of production. During the repair Dofasco installed 203 new thermocouples in the hearth, concentrated around the tapholes and the "elephant foot" where wall meets floor, at several depths so that heat flow could be calculated.
LessonDofasco answered the breakout with many more sensors where wear concentrates, read as heat flow through the wall, not as single temperatures.
Credit
R. J. Donaldson, A. J. Fischer, R. M. Sharp and D. W. Stothart (Dofasco)
2000
Thermocouples map hearth erosion
In 2000 Jan Torrkulla and Henrik Saxén of Åbo Akademi University in Finland published a model that estimates, from thermocouples in a hearth's floor and walls, the most severe erosion of the lining and the thickness of the skull, the layer of frozen iron and slag on its hot face, illustrated on data from two Finnish furnaces. Such estimates can be checked only when a furnace is blown out: Nippon Steel core-drilled the sidewall of Tobata No. 1 after its 1998 blow-out and found five distinct layers in the carbon brick, and a model by Germany's steel research institute BFI was calibrated against wall thickness measured at the 2016 relinings at Dillingen and Eisenhüttenstadt.
Credit
Jan Torrkulla and Henrik Saxén; Akihiko Shinotake and colleagues (Nippon Steel); Yalcin Kaymak and Thorsten Hauck (BFI) with ArcelorMittal Eisenhüttenstadt and Dillinger
2022
Echoes overstate the lining
Echo methods strike the furnace shell and time the sound waves that bounce back from boundaries inside the lining. Hatch says it developed its acousto ultrasonic-echo method (AU-E) in 1998, and in 2020 Hatch and the JinXi steel group compared AU-E readings with thickness measured during the June 2019 reline of JinXi's No. 9 blast furnace in Tangshan, China, which they called a "rare opportunity". In 2022 Colin Commandeur of Tata Steel IJmuiden and University of Amsterdam researchers showed that the usual reading of impact-echo spectra should be "used with care" on linings built in layers, and that in simulated cases both it and their improved method "overestimate the refractory thickness."
LessonAn echo reading is only as good as its model of every layer behind the shell, and an overestimate errs toward the unsafe side.
Credit
Afshin Sadri and colleagues (Hatch); Colin Commandeur, Rudolf Sprik, Chris Stolk and Gerard Louwerse (Tata Steel and University of Amsterdam)
Watch for
Runner wear predicted without draining
A blast furnace main runner is the refractory-lined channel that carries iron and slag away from the taphole; to measure its wall, operators normally drain it. RHI Magnesita says a machine learning model trained on two years of laser scans, castable batch records and tapping data from two blast furnaces predicted the remaining runner wall with a maximum relative error under 20%, about 40 mm, and within 10 mm in one 2025 campaign at CSN's Blast Furnace 3 in Brazil, and that Ternium Brazil used its predictions to run Blast Furnace 2's runner longer without draining it for inspection.
Achieved whenCSN, Ternium or another user publishes predicted against measured thickness for every campaign over a period, including those the model got wrong.
Claimed by
RHI Magnesita, November 2025
Kiln shells read through smoke
Researchers at Shanghai Jiao Tong University and Chongqing University of Science and Technology report a neural network that picks out a rotary kiln's shell in infrared images despite smoke, dust and other hot objects, scoring 97.03% mean intersection-over-union (a measure of how closely an outlined area matches the true one) on their own calibrated kiln dataset, as a basis for overheating warnings and judging lining wear.
Achieved whenthe dataset and code are released and the method is tested on a kiln it was not trained on, against hot spots confirmed at a shutdown.
Claimed by
Jie Li, Jianxin He and colleagues, May 2026
Open
01
Lining scans made public
A public dataset of repeated laser scans of the working lining of steel ladles or converters over whole campaigns (the run of heats between relines), each scan paired with the process record of the heats since the last one, plus a forecasting task, the thinnest remaining lining after the next 20 heats, with baseline scores.
The briefCredit · unclaimed02
Hearth models scored at reline
An open benchmark that sets estimates of hearth erosion against what was measured when furnaces were blown out: for each furnace, the thermocouple positions and last readings, the estimated erosion line, and the remaining thickness measured from cores or during dismantling; plus an open-source inverse heat-conduction model scored on all of them.
The briefCredit · unclaimed03
Kiln hot spots, labeled
An open set of rotary kiln shell scans covering at least a year from at least two kilns, with each hot spot labeled by what was found at the next shutdown (lining lost, coating thin, or nothing), the brick installation records, and a task: predict which zones will need new brick.
The briefCredit · unclaimed
Open 01 · the brief
Lining scans made public
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public dataset of repeated laser scans of the working lining of steel ladles or converters over whole campaigns (the run of heats between relines), each scan paired with the process record of the heats since the last one, plus a forecasting task, the thinnest remaining lining after the next 20 heats, with baseline scores.
Why it is still open
Laser scans are routine, but they stay with plants and suppliers; RHI Magnesita's runner model trains on customers' scans and process data that are not published. The open ladle wear model that SINTEF released for the steelmaker Sidenor in 2023 ships with the process record of a single heat and no measured lining thickness, and a September 2026 analysis on GitHub that builds on it reports the same single heat. Searches of Zenodo, Hugging Face and GitHub in November 2026 found no public set of lining scans.
Built with what exists
Point clouds exported from a plant's existing laser scanner; CloudCompare and Open3D to align each scan to the empty shell; Python forecasting libraries such as scikit-learn or Darts; the SINTEF and Sidenor model as a physics baseline; an open-weight model run on the plant's own computers if the plant will not let raw data leave (see the Note "A trillion open parameters").
First step
Ask the teams behind published scan-based studies (CSN and Ternium in Brazil, China Steel, SINTEF and Sidenor) whether one anonymized campaign of scans can be released, and draft the data sheet.
Done looks like
At least three full ladle or converter campaigns and at least 50 scans, released under an open license with a data sheet, and three forecasting methods scored in millimeters of error on a campaign held out from training.
Open 02 · the brief
Hearth models scored at reline
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open benchmark that sets estimates of hearth erosion against what was measured when furnaces were blown out: for each furnace, the thermocouple positions and last readings, the estimated erosion line, and the remaining thickness measured from cores or during dismantling; plus an open-source inverse heat-conduction model scored on all of them.
Why it is still open
Each check of a hearth estimate against a reline stays inside one paper, with its own furnace and no shared data. BFI's model was validated at the 2016 relinings at Dillingen and Eisenhüttenstadt and runs on commercial COMSOL software, and its authors note that a direct comparison "is only possible during a relining"; Hatch compared AU-E readings with JinXi's No. 9 furnace at its 2019 reline. Searches of GitHub for "blast furnace hearth" and "hearth erosion", and of Zenodo, found no open hearth wear code or dataset in November 2026.
Built with what exists
Thermocouple layouts and measured and estimated profiles digitized from published papers with WebPlotDigitizer, plus any data plants agree to share; an inverse heat-conduction model written in Python with SciPy or FEniCSx; an AI model to find and tabulate published blow-out comparisons, every figure checked by hand.
First step
Collect the blow-out comparisons already in print (Tobata No. 1, JinXi No. 9, Dillingen, Eisenhüttenstadt) and note which give numbers that can be digitized.
Done looks like
At least five blown-out furnaces in one open file, and an open model whose estimated remaining thickness is published beside the measured value at every comparison point, with the mean and largest errors in millimeters.
Open 03 · the brief
Kiln hot spots, labeled
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of rotary kiln shell scans covering at least a year from at least two kilns, with each hot spot labeled by what was found at the next shutdown (lining lost, coating thin, or nothing), the brick installation records, and a task: predict which zones will need new brick.
Why it is still open
Shell scanners have watched kilns since the 1960s, and in 1990 Kemmerer argued for tying their readings to a refractory database, but the scans stay with plants and scanner vendors. The 2026 study by Li and colleagues built a calibrated infrared kiln dataset for picking out the shell; its abstract does not say it is public or link hot spots to shutdown findings. Searches of Zenodo, Hugging Face, GitHub and the web in November 2026 found no public kiln shell scan dataset.
Built with what exists
Exports from a plant's existing shell scanner; shutdown inspection reports and brick replacement logs; Python with pandas; an image segmentation model such as a U-Net trained in PyTorch for thermal images.
First step
Ask a cement or lime plant with a shell scanner to export one year of scans and its last two shutdown reports.
Done looks like
At least 12 months of scans from at least two kilns with at least 30 hot spots labeled against shutdown findings, released openly, and a baseline whose hit rate and false-alarm rate for predicting replaced zones are published.