Industries / Mining and materials / Steel and metals
Ironmaking
Ironmaking removes the oxygen from iron ore to leave metallic iron, either in a blast furnace burning coke or in a shaft furnace fed with gas.
4
Milestones achieved, from 1709 to 2021.
1
To watch for: announced, not yet confirmed outside.
1
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1709
Coke replaces charcoal
Abraham Darby used coke instead of charcoal to smelt iron in the Old Furnace at Coalbrookdale. Others had tried coke before him; the Institution of Mechanical Engineers credits Darby with making the change economically viable.
Credit
Abraham Darby I
1828
Hot air blast
James Beaumont Neilson patented blowing a furnace with heated air instead of cold air. It was first used commercially at Colin Dunlop's Clyde Iron Works, where teething problems lasted until 1834, and it then spread through the west of Scotland on the strength of its fuel savings.
Lessona patented process can take years of plant work before it runs reliably.
Credit
James Beaumont Neilson
1969
Iron reduced by gas
The first MIDREX plant began production on 17 May 1969 at Oregon Steel Mills in Portland, Oregon. It strips oxygen from iron ore pellets with a gas made from natural gas, giving solid direct reduced iron (DRI, also called sponge iron) for electric furnaces, with no coke. Worldsteel counted 153.0 million tonnes of DRI made worldwide in 2025.
Credit
Donald Beggs and Jack Scarlett of Surface Combustion, a division of Midland-Ross
2021
Hydrogen-made steel delivered
In June 2021 the HYBRIT pilot plant in Luleå, Sweden, made sponge iron using hydrogen produced without fossil fuels, in place of coal and coke. In July SSAB rolled steel from it at Oxelösund, and the first delivery went to the Volvo Group. This was pilot-scale output, not volume production.
Credit
HYBRIT, a venture of SSAB, LKAB and Vattenfall founded in 2016
Watch for
Hybrit demonstration plant
LKAB plans a plant at Gällivare, Sweden, to make up to 1.5 million tons a year of sponge iron with hydrogen. Sweden's Land and Environment Court in Umeå approved it in June 2026, subject to appeal, and the investment decision is still pending.
Achieved whenthe plant is built and an outside party reports it making sponge iron at industrial scale.
Claimed by
LKAB, June 2026
Open
01
Test a reduction model elsewhere
A test set of published hydrogen reduction curves from laboratories other than the Max Planck group. Each curve shows how much of a pellet's oxygen is removed over time, with its temperature, hydrogen pressure, pellet size and ore type recorded. Run the group's open model on every curve and publish where it holds and where it fails.
The briefCredit · unclaimed
Open 01 · the brief
Test a reduction model elsewhere
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test set of published hydrogen reduction curves from laboratories other than the Max Planck group. Each curve shows how much of a pellet's oxygen is removed over time, with its temperature, hydrogen pressure, pellet size and ore type recorded. Run the group's open model on every curve and publish where it holds and where it fails.
Why it is still open
The Max Planck group's machine-learning model for hydrogen reduction is fitted to 179 of its own experiments at 700 to 1,000 °C and 0.5 to 1.0 bar of hydrogen. The group says it is validated only within that sampled range and disclaims use beyond it.
Built with what exists
The group's code and data on GitHub (github.com/bajpaianurag/Hydrogen_Reduction_Reaction_Transition_ML); a plot digitizer to read curves from published figures; Python. An AI model can draft the table of test conditions from each paper for a person to check; the Note "Seven more flagged" covers an AI reviewer of long documents.
First step
Clone the repository, reproduce one of the group's own fits, then list 20 papers that report reduction curves between 700 and 1,000 °C.
Done looks like
At least 100 curves from at least 10 independent laboratories, published under an open licence, each with the model's error in reduction degree (percentage points) and a plain list of the conditions where error is largest.