Industries / Mining and materials / Glass and ceramics
Melting and furnaces
Melting turns the batch into molten glass inside a furnace lined with refractories, bricks and blocks made to survive heat and molten glass, and keeps that glass hot, clean and steady for the forming machines, often for years without a stop.
5
Milestones achieved, from c. 1250 BCE to 2002.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
c. 1250 BCE
Glass made from raw materials
At Qantir-Piramesses in Egypt's Nile Delta, workshops dated to about 1250 BCE made glass from its raw materials in two stages. Raw materials were first melted into a rough glass at 900 to 950°C, in reused beer jars, then crushed and washed, colored, and melted again into ingots in ceramic crucibles at 1,000 to 1,100°C. Most of the glass was red, colored with copper, and the ingots went to other workshops to be made into objects.
Credit
Thilo Rehren (University College London) and Edgar Pusch (Pelizaeus-Museum, Hildesheim), Science, June 2005
1856
Furnaces recover their exhaust heat
Friedrich Siemens, working with his brother William in England, was granted British patent No. 2861 for the regenerative furnace on 2 December 1856. Two chambers of ceramic bricks take turns soaking up heat from the exhaust and warming the incoming air and fuel gas, with the flow reversed periodically. The first was built in a Rotherham glassworks about four years later, by The Chemical Engineer's account, and with regenerators glass came to be melted in large rectangular tanks instead of clay pots that lasted about 20 days each; the early tanks themselves were rebuilt every year. Friedrich later ran the family glassworks in Dresden, and Grace's Guide says his glass furnace "forms the basis of the modern system of glass manufacture."
Credit
Friedrich Siemens and William Siemens
1928
Cast blocks line the tank
Fused-cast refractories are made by melting oxides such as alumina, zirconia and silica (AZS) and casting them into blocks, which come out dense, with about 1% porosity, and resist molten glass. M. Dunkl and colleagues at Didier-Werke wrote in 1997 that fused-cast AZS had been used in glass tanks since 1928, with oxidized grades from the 1960s. Kevin Selkregg, writing in the American Ceramic Society Bulletin in 2018, credits Monofrax with high-alumina fused-cast blocks in the late 1930s and says AZS became the lining of choice after the Second World War, raising glass-contact life from 8 to 12 months for silica brick to 3 to 5 years.
Credit
Monofrax, by Selkregg's account; Dunkl and colleagues name no maker
1991
Oxygen firing eats the crown
From 1991, glassmakers began burning fuel with pure oxygen instead of air, which shrinks the exhaust by as much as 75% and, with 100% oxygen, cuts nitrogen oxides, by the US Department of Energy's 2002 count. Silica crowns, the arched furnace roofs, which last about ten years in air-fired furnaces, then corroded two to three times faster, in some cases losing 20 cm in six months. A study led by PPG Industries and Sandia National Laboratories from 1998, funded by the Department of Energy, traced the attack to sodium hydroxide vapor. The exhaust of oxygen-fired furnaces is about 65% water vapor against about 18% with air, by the study's calculations, and the team built a portable instrument to measure the vapor in working furnaces.
LessonChanging the flame changes the furnace's atmosphere, and refractories have to be chosen for the gases as well as the heat.
Credit
PPG Industries and Sandia National Laboratories (study, report SAND2005-1196)
2002
Electricity heats the melt
Electric boosting passes a current through the molten glass between molybdenum electrodes to add heat to a gas-fired furnace. By the US Department of Energy's 2002 profile of the US industry, about 60% of regenerative container-glass furnaces had electric boost, which typically supplied 10 to 15% of a furnace's energy. About 10% of container furnaces and the glass-wool insulation plants melted with electricity alone, under a "cold top" of unmelted batch that keeps the heat in.
LessonAll-electric melting was the most efficient but stayed in smaller furnaces because of the price of electricity, and worn electrodes can put too much molybdenum into some glasses.
Credit
counted by the US Department of Energy's Office of Industrial Technologies
Watch for
Defects forecast two days ahead
CelSian says its Celfos system combines furnace data, machine learning and its furnace flow simulations (GTM-X) to predict defects such as cords, seeds and bubbles up to 48 hours before they appear, and that it runs at several container, float and glass fiber plants. The announcement gives no accuracy figures and names no sites.
Achieved whena glassmaker or an outside party publishes predicted against actual defect rates over several months of production.
Claimed by
CelSian, May 2026
Cameras track the batch line
AMETEK Land says its ImagePro Glass AI software, released on 9 June 2026, uses neural networks on images from up to 16 thermal cameras to track the batch line, where the unmelted batch floating on the glass ends, and to detect flames without manual setup. It gives no accuracy figures.
Achieved whena plant or tester outside AMETEK publishes how closely it locates the batch line against reference measurements.
Claimed by
AMETEK Land, June 2026
Radar reads worn furnace bottoms
In conference slides posted in October 2025, PaneraTech says radar scans from its SmartMelter service detect refractory wear and glass seeping into furnace bottoms 2 to 3 years before outside temperatures rise. It says cooling and repairs guided by the scans extend furnace life by 5 to 6 years, drawing on assessments at 22 float glass makers.
Achieved whena glassmaker or outside party publishes radar readings checked against measured block thickness at a rebuild, or campaign lengths of monitored and unmonitored furnaces.
Claimed by
PaneraTech (presenter Mohit Gohel), October 2025
Open
01
Open furnace data benchmark
A public, documented time series from working glass furnaces, with temperatures, fuel and electricity use, pull (glass drawn per hour), batch charging and glass level, plus a forecasting task and baseline scores, so that control and AI methods can be compared on the same data. Work with the furnace's owner and export what their controller already records; change nothing on the burners or controls.
The briefCredit · unclaimed02
Furnace campaign ledger
An open table of glass furnace campaigns worldwide, a campaign being the years a furnace runs between rebuilds. For each: plant, owner, glass type, furnace type and fuel, start date, rebuild date, tonnage if published, and why it ended, compiled from company announcements and trade press, with summaries of campaign length by furnace type.
The briefCredit · unclaimed
Open 01 · the brief
Open furnace data benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, documented time series from working glass furnaces, with temperatures, fuel and electricity use, pull (glass drawn per hour), batch charging and glass level, plus a forecasting task and baseline scores, so that control and AI methods can be compared on the same data. Work with the furnace's owner and export what their controller already records; change nothing on the burners or controls.
Why it is still open
Furnace models shared on GitHub train on data they do not publish (a 2024 crown-temperature regression) or on controller outputs from simulation (a 2026 melter-level model), and Argonne National Laboratory's open Glass Furnace Model (released 2015) simulates furnaces rather than recording them. Commercial tools such as Celfos report no accuracy on shared data. Searches of GitHub and Hugging Face found no public glass furnace dataset.
Built with what exists
Exported controller logs, or a data logger such as an ESP32 board running ESPHome (an open project already automates a studio glass furnace this way), a gas meter with a pulse output, and Python forecasting libraries such as statsmodels or Darts.
First step
Ask a university or public glass studio whose furnace controller already logs temperatures to share three months of exported logs.
Done looks like
At least six months of one-minute data from one or more furnaces, released under an open license with a data sheet, and a benchmark of three methods forecasting glass temperature one hour ahead, with errors in °C.
Open 02 · the brief
Furnace campaign ledger
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table of glass furnace campaigns worldwide, a campaign being the years a furnace runs between rebuilds. For each: plant, owner, glass type, furnace type and fuel, start date, rebuild date, tonnage if published, and why it ended, compiled from company announcements and trade press, with summaries of campaign length by furnace type.
Why it is still open
Campaign lengths reach the public one furnace at a time, often from the companies that sell the furnaces or services: a builder reports a 17-year oxy-fuel campaign, and a radar service says its guidance adds 5 to 6 years. The National Glass Association's World of Glass database, free to its members and sold to others, lists float plants by company and location, and its page mentions no furnace start or rebuild dates. Searches found no public statistics across many furnaces.
Built with what exists
Company press releases and annual reports, trade press archives (Glass International, Glass Worldwide, Glassonweb), a spreadsheet or SQLite database, and a large language model to pull dates from articles, with every date checked by hand.
First step
List every float glass furnace rebuild announced in 2024 to 2026 and trace each back to the start of its previous campaign.
Done looks like
300 campaigns, each with a start and an end date backed by a cited source, released as an open CSV, with median campaign length by glass type and fuel.