Industries / Mining and materials / Cement and concrete
The kiln
The kiln is where raw meal is heated to around 1,500 °C so that it partly melts and forms clinker, the hard grey nodules that are ground into cement, and it burns most of a cement plant's fuel.
7
Milestones achieved, from c. 1842 to 2023.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
c. 1842
Burned hotter, ground harder
Joseph Aspdin's product was one of several made to copy Roman cement, a fast-setting cement used for stucco. His son William left the family works in Wakefield in 1841, set up at Rotherhithe on the Thames, and within months sold a cement made with more limestone in the mix and burned hotter, to a hard clinker. The historian Dylan Moore calls William the inventor of Portland cement "in the modern sense of the term". Isaac Charles Johnson, who worked out how to copy the product for J. B. White at Swanscombe around 1845, later claimed to have invented "true" Portland cement; Moore calls that claim "specious".
LessonThe better cement cost more to make: more limestone, more fuel, and clinker so hard that millstones needed re-dressing after about 36 hours instead of a month.
Credit
William Aspdin
1890
The rotary kiln works
A rotary kiln is a long brick-lined steel tube that turns slowly on a slight slope, so the material tumbles down toward a flame at the lower end. Frederick Ransome patented one in Britain in 1885, but his trial kilns there, fired with producer gas (a fuel gas made from coal), could not get hot enough to make clinker and were given up. The Keystone Portland Cement Company built a kiln to Ransome's design at Coplay, Pennsylvania, fired it with cheap Ohio oil from November 1889, and had it working well by spring 1890. Coal-fired kilns followed by 1895, and by Moore's account the US industry grew from almost nothing to half of world cement output in 1907.
LessonThe British trials failed for want of heat, not because the design was wrong.
Credit
Frederick Ransome (patent); Keystone Portland Cement Company, later Atlas Portland Cement (first working kiln)
1953
Heating feed in the gas
A suspension preheater blows powdered raw meal into the hot exhaust gas rising from the kiln, then separates it again in a cyclone, a cone that spins the dust out of the gas. Three to six such stages are stacked in a tower. Because the heat passes into every particle at once, gas and powder reach the same temperature in a fraction of a second, and most of the heat that used to go up the stack is recovered. Humboldt (now KHD) ran the first such kiln at Beckum, Germany, in 1953, and by the end of the 1960s every major kiln supplier had its own version, by Moore's account.
Credit
Humboldt (KHD)
1973
A fire before the kiln
A precalciner burns part of the fuel, up to about 70% in common designs, in a vessel at the bottom of the preheater, so the raw meal is 90–95% calcined (its carbon dioxide driven off) before it enters the kiln. The rotary kiln then only has to sinter, so a small kiln gives a large output. Moore dates the first precalciner kilns to Japan in the late 1960s. In 1973 a Korea Cement Association journal reported that the Japanese engineering firm IHI had developed one, the "SF" process with its "Flash Furnace", together with Chichibu Cement. Nearly all new kilns now have precalciners.
Credit
IHI and Chichibu Cement, among other Japanese makers
1981
Operators' rules in software
Fuzzy logic lets a computer apply rules written in words, such as "if the burning zone is a little cold, raise the fuel slightly", the way kiln operators already worked. L. P. Holmblad and J.-J. Østergaard of the Danish kiln maker F.L. Smidth described fuzzy-logic control of a cement kiln in a 1981 paper for the International Federation of Automatic Control. In 1983 E. H. Mamdani and colleagues at Queen Mary College, London, wrote that the method was "now being used commercially for the control of cement kilns".
Credit
L. P. Holmblad and J.-J. Østergaard, F.L. Smidth
2021
China retires shaft kilns
Shaft kilns, small upright kilns fed from the top, made about 70% of China's cement from 1996 to 2000. A national policy to close small plants with high energy use and heavy pollution cut the shaft kilns' share sharply in 2001, and new suspension preheater kilns with precalciners overtook the older kinds in 2007. By 2021 they made about 99% of China's cement, the world's largest output. By the authors' estimate, the new kilns burn about half as much fossil fuel as the old ones, and by 2021 the switch had cut particulate pollution by about 53% against a case with no upgrade.
Credit
China's cement makers under national policy; measured by Xu Xiaozhen, Liu Gang and colleagues (Communications Earth & Environment, 2022)
2023
Waste as the main fuel
German cement kilns took 73.4% of their fuel energy from alternative fuels in 2023, up from 63.4% in 2014, by the figures of the German cement association VDZ. The fuels include sorted fractions of industrial, commercial and household waste, plastics, used tyres, meat and bone meal and sewage sludge, while coal, lignite and petroleum coke supply the rest. Making a tonne of cement in Germany took 2,807 MJ of fuel energy on average that year.
Credit
German cement makers; reported by VDZ
Watch for
Clinker from electricity alone
SaltX Technology of Sweden says it has made high-quality clinker with a fully electrified process, and with Holcim and Paebbl, a company that binds CO2 into building materials, it plans to test its electric clinker reactor at industrial scale at its research centre in Hofors, Sweden. The project has EU funding through the CET Partnership and a total budget of SEK 45 million, and is due to start in late 2026 and run about three years.
Achieved whenclinker made in the Hofors unit with electric heat only passes standard cement tests at an outside laboratory, and the results and the electricity used per tonne are published.
Claimed by
SaltX Technology and Holcim, June 2026
A claimed 10% fuel cut
Visionaize, a software firm in San Jose, California, says its machine-learning kiln optimizer cut specific heat consumption (the fuel heat used per kilogram of clinker) at an unnamed South Asian plant from 820 to 740 kcal/kg, about 10%, cut mill and kiln downtime by 60% and saves about $1.4 million a year. The system reads more than 250 process variables and gives operators recommendations rather than running the kiln itself.
Achieved whenthe plant is named and an outside energy auditor confirms the before-and-after heat consumption, corrected for changes in raw meal, fuel and output.
Claimed by
Visionaize, October 2026
Open
01
One test for kiln control
An open-source test bench for kiln control. It holds code for a published model of a whole kiln line (preheater, precalciner, rotary kiln and cooler), a fixed set of upsets (a swing in raw-meal chemistry, a drop in fuel heating value, wet waste fuel, a feeder stall), and a scoring script. The script reports fuel per tonne of clinker, the spread of free lime (lime left unreacted, the usual check of clinker quality) and the time with carbon monoxide in the exhaust. Run at least a tuned PI controller (the standard feedback loop), a model predictive controller and one machine-learning controller through it.
The briefCredit · unclaimed02
Check the savings claims
A free tool that takes hourly plant data from before and after a kiln optimizer is switched on (kiln feed, raw-meal chemistry, fuel rates and heating values, clinker free lime) and estimates the fuel saving after correcting for changes the optimizer did not cause. Test it on simulated plant data in which the true saving is known.
The briefCredit · unclaimed
Open 01 · the brief
One test for kiln control
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source test bench for kiln control. It holds code for a published model of a whole kiln line (preheater, precalciner, rotary kiln and cooler), a fixed set of upsets (a swing in raw-meal chemistry, a drop in fuel heating value, wet waste fuel, a feeder stall), and a scoring script. The script reports fuel per tonne of clinker, the spread of free lime (lime left unreacted, the usual check of clinker quality) and the time with carbon monoxide in the exhaust. Run at least a tuned PI controller (the standard feedback loop), a model predictive controller and one machine-learning controller through it.
Why it is still open
Researchers at the Technical University of Denmark and FLSmidth published the equations for such a model between 2024 and 2026 and used it to tune PI loops, but the 2025 paper's full text points to no code release, and none was found in this check. That paper says a dynamic model of the whole process suitable for control was not available before it. AI kiln claims such as Visionaize's are each measured on one plant, so methods cannot be compared. Moore notes why control is hard: holding free lime within ±0.5% would need the feed rate held within 0.25%, about ten times finer than kiln feeders can meter.
Built with what exists
Python with CasADi, or OpenModelica, to solve the model's equations; do-mpc for model predictive control; Gymnasium to connect the bench to reinforcement-learning agents, with PC-Gym, an open set of chemical process-control test environments that holds no kiln, as a pattern to follow.
First step
Code the 2024 rotary-kiln part of the model on its own (arXiv 2405.03200) and match the paper's published plots.
Done looks like
Public code that reproduces the steady state in the 2025 paper within 2% on its main temperatures and flows, ten published upset scenarios, and a results table for at least three controllers.
Open 02 · the brief
Check the savings claims
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free tool that takes hourly plant data from before and after a kiln optimizer is switched on (kiln feed, raw-meal chemistry, fuel rates and heating values, clinker free lime) and estimates the fuel saving after correcting for changes the optimizer did not cause. Test it on simulated plant data in which the true saving is known.
Why it is still open
Published claims compare plain before-and-after figures from one plant, such as Visionaize's 820 to 740 kcal/kg, and do not say how changes in raw materials, fuels or output were handled. Fuel mixes shift a lot: Germany's alternative-fuel share rose from 63.4% to 73.4% of fuel energy between 2014 and 2023.
Built with what exists
The test bench above, or any published kiln model, to make the simulated data; Python with statsmodels; the International Performance Measurement and Verification Protocol (IPMVP) of the Efficiency Valuation Organization as the framework for baseline correction.
First step
Take three published kiln-optimizer case studies and list which of the needed data each one reports.
Done looks like
On 20 simulated switch-overs with true savings between 0 and 5%, the tool's estimates fall within 0.5 percentage points of the truth, reported next to how far a plain before-and-after comparison misses on the same cases.