BRENNEN

Industries / Mining and materials / Steel and metals / Aluminum smelting

Smelting pots

Smelting pots, also called cells, pass a large direct current through alumina dissolved in a bath of molten salt, which splits it into liquid aluminum that pools at the bottom and oxygen that burns away the carbon anodes hanging in the bath.

6

Milestones achieved, from 1888 to 2014.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1888

    First commercial pots

    Alfred E. Hunt and a few associates raised $20,000, and the Pittsburgh Reduction Company was started on 18 September 1888 to work Charles Martin Hall's process. Its small plant on Smallman Street in Pittsburgh began operating on Thanksgiving Day 1888. Hall and the company's first employee, Arthur Vining Davis, built steel pots with a baked carbon lining to hold the molten salt, which they called the bath, and carbon anodes 3 inches across. On a good day the plant made about 30 pounds (14 kg) of aluminum, which sold for $8 a pound, by the history behind the Pennsylvania state marker at the site. Theodore Beck, writing in The Electrochemical Society's magazine in 2014, says today's prebaked-anode pots keep the same basic design except for size, hooding and fume recovery.

    electrochem.orgexplorepahistory.com

    Credit

    Charles Martin Hall; Alfred E. Hunt; Arthur Vining Davis

  2. 1940s

    Fifty-kiloampere prebake pots

    In the 1940s Alcoa built seven smelters for the US government using its N-40 pot, which ran at 50 kiloamperes (kA, thousands of amperes). A prebake pot uses anodes baked in a separate furnace before they are set in the pot. Pots are wired in series in long rows called potlines, so the same current passes through every pot in a line. Alan Tabereaux's review in the journal JOM counted more than 50 types of prebake pot in use in 2000, and found line current had risen 500% from these 1940-vintage 50 kA pots to 300 kA.

    tms.org

    Credit

    Alcoa

  3. 1969

    Feeding alumina at points

    A hard crust of frozen bath and alumina forms on top of the pot, and older pots were fed by breaking it along a whole side or down the middle. Alcoa's P-155 pot, introduced in 1969 and run at 170 kA, feeds alumina instead through a few points in the crust. Tabereaux lists point break-and-feed and computer process control among the main changes in prebake pots, and records side-worked lines such as Lynemouth in England and Kurri Kurri in Australia being converted to point feeding.

    tms.org

    Credit

    Alcoa

  4. 1987

    Söderberg line converted

    In a Söderberg pot the anode is made continually from briquettes of petroleum coke and pitch added at the top, and bakes in place inside the pot. It is named for Carl Wilhelm Söderberg, whose self-baking electrode Elkem patented in 1919. Halvor Kvande of the Norwegian University of Science and Technology writes that prebaked anodes, baked elsewhere where emissions are easier to capture, give much lower emissions of harmful polycyclic aromatic hydrocarbons (PAHs), and have almost universally replaced Söderberg pots. In one conversion Tabereaux records, a 60 kA Söderberg potline at Sundsvall, Sweden, became 56 Kaiser P-86 prebake pots at 150 kA in 1987.

    Lessonbaking the anode inside the pot saved a separate furnace but released its fumes where they were harder to capture.

    pmc.ncbi.nlm.nih.govelkem.comtms.org

    Credit

    Carl Wilhelm Söderberg and Elkem (electrode); Kubal and Kaiser Aluminum (Sundsvall conversion)

  5. 1999

    Magnetic compensation at Søral

    The current in a potline makes strong magnetic fields, which push on the liquid metal in the pots and can set it moving and waving, so they limit how much current a pot can take. Alusuisse designed asymmetric busbars, the heavy bars that carry current from pot to pot, to cancel part of the field. Using them, the Søral smelter at Husnes, Norway, raised ten test pots from 120 kA to 150 kA and decided to convert the busbars of all 320 of its pots by December 1999, Tabereaux reports.

    tms.org

    Credit

    Alusuisse; Søral

  6. 2014

    Pots at 570 kiloamperes

    Rio Tinto Alcan started the first 38 AP60 pots at Jonquière, Québec, at the end of 2013, and inaugurated them as the Arvida AP60 Technology Centre on 16 January 2014. Each pot is 18 meters long, twice the length of the company's AP18 pot, and each anode weighs 1,420 kg and is used up in 28 days. The first generation ran at 570 kA, with a goal of 600 kA. Kvande put the world's largest pots at 600 kA in 2014, against about 300 kA for a large pot. In May 2026 Rio Tinto began commissioning 96 more AP60 pots at Arvida, adding about 160,000 tonnes a year, and said older potrooms there would close in June.

    ap-technology.commagazine.cim.orgpmc.ncbi.nlm.nih.gov

    Credit

    Rio Tinto Alcan, building on Pechiney's AP series

Watch for

  1. Inert anode demonstration plant

    Rio Tinto is building a plant of ten ELYSIS pots at 100 kA beside its Arvida smelter in Québec. ELYSIS pots use inert anodes, which are not burned away and give off oxygen instead of carbon dioxide. When the plant was announced in June 2024, it was to make up to 2,500 tonnes a year of commercial-quality aluminum, with first production targeted by 2027 and Alcoa holding an option to buy up to 40% of the metal in the first four years. In June 2026 Rio Tinto said the plant was under construction.

    Achieved whenthe plant makes metal and a buyer or auditor reports the tonnes made, how long the anodes last and the metal's purity.

    magazine.cim.orgfastmarkets.com

    Claimed by

    Rio Tinto and ELYSIS, June 2026

Open

  1. 01

    Finish the open smelter registry

    Fill in the empty technology field of the open smelter registry published in October 2026: for each potline, its pot technology (prebake or Söderberg, and the named design, such as AP60), line current in kA, number of pots and start year. Add the Chinese smelters the registry leaves out. Link every field to a public source.

    The brief
    Credit · unclaimed
  2. 02

    Feed control under swinging current

    An open benchmark that scores alumina feed controllers on a public pot model while the line current is raised and lowered, as it is when a smelter sells flexibility to the grid. Alumina must be fed as fast as it is used. Too little sets off an anode effect, a sudden jump in resistance that gives off perfluorocarbon gases; too much leaves sludge on the pot floor.

    The brief
    Credit · unclaimed

Open 01 · the brief

Finish the open smelter registry

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

Fill in the empty technology field of the open smelter registry published in October 2026: for each potline, its pot technology (prebake or Söderberg, and the named design, such as AP60), line current in kA, number of pots and start year. Add the Chinese smelters the registry leaves out. Link every field to a public source.

Why it is still open

Zach Dissington's registry, released under the CC BY 4.0 license, lists 94 smelters operating in 2020 with name, country, location and capacity. Its technology field "is present in the schema but unpopulated", and China, home to roughly 120 smelters, is covered by 12 anchor plants. A 2025 inventory by Chang Tan and colleagues at Tsinghua University gives line current and start year for 249 smelters, but under coded IDs instead of names, drawn from a commercial survey, and marked "All rights reserved". The last public census of pot types found is Tabereaux's review from 2000.

essd.copernicus.orgnature.comtms.org

Built with what exists

The registry's CSV file and code; company annual and sustainability reports; papers from the TMS Light Metals and ICSOBA conferences; Chinese company filings and provincial announcements read with machine translation; a 2017 list of about 236 smelters on GitHub (withetu/Map-Alc-Report) as an index; an AI model to draft entries from long reports for a person to check (see the Note "Seven more flagged").

  1. First step

    Download the registry, fill in the technology field for its 27 operating European smelters from public sources, and send the additions to its author.

  2. Done looks like

    Pot technology, line current and pot count sourced for at least 150 smelters, at least 80 of them in China, each field checked by a second person, released under CC BY 4.0, with China's total capacity in the registry within 10% of the USGS figure.

Open 02 · the brief

Feed control under swinging current

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open benchmark that scores alumina feed controllers on a public pot model while the line current is raised and lowered, as it is when a smelter sells flexibility to the grid. Alumina must be fed as fast as it is used. Too little sets off an anode effect, a sudden jump in resistance that gives off perfluorocarbon gases; too much leaves sludge on the pot floor.

Why it is still open

Published controllers are tested on private plant pots or company models. A University of New South Wales team tested a model-based feed controller on one industrial pot (JOM, 2022), and Rio Tinto's research center and PhysicsX tuned feeder timing with the commercial Alucell software and a neural-network stand-in model (Light Metals, 2025). The pot simulators used to test controllers are commercial or in-house, such as KAN-NAK's MATLAB-Simulink model. Researchers at the Norwegian University of Science and Technology have published an open single-pot simulator, but for building models, with no rules for scoring controllers. A search of GitHub, Crossref and the web found no feed-control benchmark.

link.springer.comlink.springer.comicsoba.orggithub.com

Built with what exists

The open simulator (Python and CasADi, github.com/erlendlundby/Costa-aluminum), whose inputs include alumina feed, line current, aluminum fluoride feed, metal tapping and anode distance; current-swing scenarios taken from published trials, such as Trimet's 150 to 180 kA; standard control code for the usual overfeed-underfeed logic and for model predictive control.

  1. First step

    Run the simulator at steady current, then add one scenario that cuts current by 15% for 24 hours and record how a simple overfeed-underfeed controller copes.

  2. Done looks like

    A public repository with at least ten current-swing scenarios and at least three controllers, each scored on hours with alumina below a stated low limit, bath temperature range and alumina fed per tonne of metal, reproducible with one command.