Industries / Mining and materials / Steel and metals / Aluminum smelting
Casting
Casting is the casthouse work of cleaning molten aluminum and freezing it into ingots, slabs for rolling, billets for extrusion, or continuous strip.
7
Milestones achieved, from 1938 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1938
Water sprayed on the ingot
Early aluminum was poured into 45 kg steel tilt molds, and the ingots cracked, held pockets of uneven chemistry and had to be heavily machined. William T. Ennor of Alcoa's Massena works filed a patent on 2 July 1938, granted in November 1942, for direct chill (DC) casting: metal freezes into a shell in a short mold, and water sprayed straight onto the shell finishes freezing the ingot as it is lowered. Alcoa's Robert Sanders writes that the method spread to virtually all Alcoa plants during the 1930s, and that the largest ingot grew from about 3.1 tonnes in 1951 to 15.5 tonnes for sheet by 2001.
Credit
William T. Ennor, Alcoa
1948
Casting without stopping
A continuous caster freezes metal into an endless bar, strip or slab instead of one ingot at a time. Sanders credits the first commercial use to Properzi, whose wheel-and-belt caster made electrical conductor wire from 1948. The Hunter twin-roll caster, which freezes strip between two water-cooled steel rolls, was commercialized in the early 1950s, and Alcan installed one of the first successful Hazelett twin-belt casters in 1959. By 2001 more than 260 twin-roll casters were running worldwide, and roll and slab casters made nearly 25% of US sheet and foil.
Lessoncontinuous casting skips costly hot rolling, but it has stayed mostly limited to low-alloy grades, typically 2.5% magnesium or less.
Credit
Properzi; Hunter; Hazelett; Alcan
1957
Water under molten metal
Alcoa scientist George Long published "Explosions of molten aluminum and water: cause and prevention" in Metal Progress in 1957, which a 2025 conference paper calls one of the earliest public scientific papers on the hazard and still relevant. When molten aluminum meets water trapped on a wet, rusty or uncoated surface, the water can flash to steam or react with the metal and explode. From 1968 the Aluminum Association led industry research that identified coatings for casting pits and tools that stop such explosions, and its handling guidelines list the coatings that passed testing.
Lessonmost casthouse explosions start with water where molten metal can reach it, so the defenses are dry, preheated tools and tested coatings.
Credit
George Long, Alcoa; the Aluminum Association
1969
A magnetic field as mold
Zinovy Getselev and five co-inventors received US patent 3,467,166 in September 1969, filed in March 1967, for casting in an alternating electromagnetic field that holds the liquid top of the ingot in place, so the metal freezes without rubbing on a mold wall. A 1995 review by J. W. Evans of the University of California, Berkeley, with co-authors from Reynolds Metals and other firms, says electromagnetic casting (EMC) "is used extensively in the aluminum industry" to cast ingots with a good surface for rolling.
Credit
Zinovy Getselev and colleagues
1975
Ceramic foam filters
Before inline treatment, plants bubbled chlorine through the furnace to remove hydrogen and let oxide particles settle, which Sanders calls inefficient, ineffective and environmentally unsound. Michael Pryor and Thomas Gray of Swiss Aluminium patented a filter made by soaking a flexible foam in ceramic slurry and firing it, which leaves a rigid ceramic sponge that molten aluminum flows through on its way to the mold. The patent was filed in January 1974 and granted in July 1975. In the patent's own test, about 5,400 pounds (2,450 kg) of alloy passed through the filter, and it removed more residue than a commercial bed filter of tabular alumina. Sanders describes such filters as an inexpensive, smaller alternative to large bed filters.
Credit
Michael J. Pryor and Thomas J. Gray, Swiss Aluminium
1981
Explosions reported in common
The Aluminum Association's Molten Metal Incident Reporting program collects anonymous reports of explosions from plants worldwide, so companies can learn from each other's. Its 2025 report counts 4,834 incidents from 1981 through 2024, graded Force 1, the least severe, to Force 3, the most severe. It received 141 reports for 2024, down from a record 198 in 2022. Two were Force 3, both during billet casting outside the United States, and together they killed 7 people; their details came from government reports, not company reports. Wet or rusty drain pans were the leading cause of explosions while moving metal from 2008 to 2024 (154 incidents), followed by wet tools (144).
Credit
The Aluminum Association
2024
Four billet pits explode
Four explosions at vertical DC billet casting stations in China, on 18 February, 9 July, 26 July and 9 August 2024, killed 15 people and injured 38, by the count of a 2025 conference paper that draws on city government investigation reports. China Daily reported that one of them, at Henan Zhongrui Nonferrous Metal Materials in Yongcheng, killed 5 and injured 14. The paper's authors, who work for makers of pit coatings, found four things in common: weak operating procedures, long shifts, casting tables run on hydraulics or steel cables, and no approved pit coatings. The Aluminum Association's summary for 2024 lists two Force 3 explosions, not four.
Lessonprotections known for decades do not help where they are not taught or used.
Credit
Alex Lowery of Wise Chem and Liu Jin of Pyrotek China, from investigation reports by the Fuqing and Shangqiu city governments
Watch for
AI grades billet structure
RUSAL says its automatic microstructure analysis system (AMAS) measures eight features of 6xxx extrusion billet, such as grain size and the iron-bearing particles, from microscope images in about 15 minutes, where a technician working by hand takes four to five hours. Its results matched operators using image software "in most scenarios"; for one feature it was right in about 70% of cases, inclusion analysis was not yet working, and RUSAL says performance on plant laboratory equipment still needs validation.
Achieved whenan outside laboratory or customer publishes a blind comparison with qualified metallographers on the same samples, or RUSAL reports routine use in its plant laboratories with measured agreement.
Claimed by
RUSAL, October 2025
Open
01
Billet microstructure answer key
A public set of microscope images of 6xxx extrusion billet samples, each graded by at least three independent laboratories for grain size, the depth of the alloy-rich layer near the surface, porosity, inclusions and the state of the iron-bearing and magnesium-silicon particles. Add a script that scores any person or program against the laboratories' median.
The briefCredit · unclaimed02
Public casthouse explosion register
A public, case-by-case list of molten aluminum explosions taken from official sources such as government accident investigation reports, OSHA inspection records and court filings. Each case gets the date, country, plant type, operation (melting, moving metal or casting), severity, cause in the Aluminum Association's categories, deaths and injuries, and a link to its source.
The briefCredit · unclaimed
Open 01 · the brief
Billet microstructure answer key
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of microscope images of 6xxx extrusion billet samples, each graded by at least three independent laboratories for grain size, the depth of the alloy-rich layer near the surface, porosity, inclusions and the state of the iron-bearing and magnesium-silicon particles. Add a script that scores any person or program against the laboratories' median.
Why it is still open
In a RUSAL round robin, five metallography laboratories measuring identical images got results that differed by up to a factor of two. RUSAL has not published its own training images, and its authors call small training sets their biggest problem. The public micrograph benchmarks gathered in a 2026 Zenodo deposit, such as ultrahigh-carbon steel (UHCS) and additively made steel (MetalDAM), are not labeled for the features used to grade billet. Open aluminum micrographs, such as the University of Cambridge's DoITPoMS teaching library, carry descriptions but no grain-size labels.
Built with what exists
Archived billet micrographs from university metallography laboratories and extruders; ImageJ or Fiji for manual measurement; open segmentation models such as Meta's Segment Anything and a U-Net as baselines; Zenodo or Hugging Face for hosting.
First step
Write a grading guide from RUSAL's eight parameters and ask two university laboratories and one extruder for archived billet micrographs they can share.
Done looks like
At least 500 fields of view from at least 30 billets from at least three casthouses, each graded by three or more laboratories, published under an open license with the spread between laboratories for each feature and one baseline model's error against the laboratory median.
Open 02 · the brief
Public casthouse explosion register
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, case-by-case list of molten aluminum explosions taken from official sources such as government accident investigation reports, OSHA inspection records and court filings. Each case gets the date, country, plant type, operation (melting, moving metal or casting), severity, cause in the Aluminum Association's categories, deaths and injuries, and a link to its source.
Why it is still open
The Aluminum Association's program is anonymous and is published only as yearly totals; for the two deadliest 2024 explosions it had to rely on government reports instead of company reports. With no shared case list, counts differ: the Association lists two Force 3 explosions with 7 deaths for 2024, while a 2025 conference paper built from Chinese city government reports counts four explosions at Chinese billet pits with 15 deaths. Searches of GitHub, Zenodo and the web found no such register.
Built with what exists
Accident investigation reports from city and provincial emergency management bureaus in China; OSHA's public inspection search; machine translation; an AI model to draft each entry for a person to check (see the Note "Seven more flagged"); a spreadsheet or SQLite database published on GitHub.
First step
List every fatal explosion reported at aluminum casthouses from 2020 to 2025 and code the first 20 cases.
Done looks like
At least 100 incidents since 2000 from at least five countries, each with at least one official source, coded by two people working apart, with their agreement on cause measured and reported.