Industries / Mining and materials / Plastics and rubber
Tires and rubber goods
This work turns natural and synthetic rubber into finished products: tires first, then hoses, belts, seals, gloves and molded parts, by mixing the rubber with fillers and chemicals, shaping it, and heating it until it sets.
7
Milestones achieved, from 1839 to 2020.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1839
Heat and sulfur cure rubber
Raw rubber turned sticky in heat and stiff in cold. Heating it with sulfur, a step now called vulcanization, made it tough and stable. Charles Goodyear found the effect in 1839 and received US patent 3,633 on 15 June 1844. Thomas Hancock had filed a British patent on 21 November 1843, and the Linda Hall Library calls who came first "a contentious subject".
LessonA patent filed first in one country can shut the inventor out of another: Goodyear could not get a British patent, and lost when he challenged Hancock.
Credit
Charles Goodyear (US); Thomas Hancock (UK)
1888
The air-filled tire
In 1845 Robert William Thomson patented a hollow rubber belt filled with air, which National Museums Scotland calls "the first pneumatic rubber tyre". In 1888 John Boyd Dunlop developed what the same museum calls "the first practical pneumatic tyre", for bicycles, and he lost his patent rights in 1890 because Thomson's came first.
LessonAn invention can arrive before its materials. Thomson found thin rubber hard to get, and the air-filled tire waited about 43 years.
Credit
Robert William Thomson; John Boyd Dunlop
1912
Carbon black toughens tread
Carbon black is a fine black powder. Mixing more of it into tread rubber bound the rubber together and made treads last much longer, and a Michelin representative quoted by Jalopnik put the gain at "a tenfold increase in wear resistance". Credit is disputed. A 1992 ASTM article credits S. C. Mote of Silvertown, England, in 1904, and says Diamond Rubber Co. of Akron used it in tires from 1912. Birla Carbon credits B. F. Goodrich, which sought a supply in 1911.
Credit
S. C. Mote; B. F. Goodrich and Diamond Rubber Co
1939
The O-ring seal
An O-ring is a rubber ring with a round cross-section that sits in a groove and seals by being squeezed. Niels Christensen's US patent 2,180,795 was granted on 21 November 1939. The US Army Air Corps tested O-rings in 1940, and within two years they were the specified seal for nearly every suitable use in aircraft hydraulics.
LessonA part can work in service before anyone knows why. Christensen's explanation was wrong, and researchers at Vought-Sikorsky and Lockheed found the real mechanism in 1941.
Credit
Niels A. Christensen
1946
The radial tire
In a radial tire the body cords run at right angles to the direction of travel, under a steel belt. By Michelin's own account its X tire has "around twice as long a service life" as earlier tires. Michelin applied for the patent on 4 June 1946, and in 1951 the Lancia Aurelia B20 was the first car to leave the factory on Michelin radials.
Credit
Michelin. Arthur Savage filed an earlier radial car-tire patent in 1915
2000
Firestone recall and TREAD Act
On 9 August 2000 Firestone recalled 14.4 million 15-inch tires, mostly on Ford Explorers, after treads came away from the tire body, which in some cases led to rollovers. The US TREAD Act, signed on 1 November 2000, made makers report foreign safety recalls within 5 days and gave regulators an early-warning system. It also led to a new tire standard, FMVSS No. 139, in 2003.
LessonWarning signs spread across complaints and foreign recalls went unnoticed until makers had to report them.
Credit
Bridgestone/Firestone and Ford (the failure); the US Congress and NHTSA, the US road-safety regulator (the response)
2020
A tire additive kills salmon
6PPD is a chemical added to tires to protect the rubber from ozone. In December 2020 a University of Washington Tacoma and Washington State University team reported in Science that it reacts with ozone to form 6PPD-quinone, which kills adult coho salmon in urban streams before they spawn. The US tire makers' trade body, USTMA, says there is still "no commercially available alternative" with comparable safety and performance.
LessonAn additive has to be judged by what it turns into on the road and in water, not only by what goes into the mixer.
Credit
Zhenyu Tian (lead author), Edward Kolodziej and Jenifer McIntyre
Watch for
A 6PPD replacement
Flexsys, a maker of rubber chemicals, says it has "the industry's first viable alternative to 6PPD". It named no molecule and published no data. In August 2026 it said it had narrowed the choice to two molecules and signed an agreement with the US Geological Survey (USGS) for independent tests on aquatic life. Lanxess announced its own alternative, Vulkanox 4060, in September 2026.
Achieved whenan independent lab such as USGS publishes results showing the chemical and what it breaks down into are safe for coho salmon, and a tire maker sells tires made with it.
Claimed by
Flexsys, November 2025 (update August 2026); Lanxess, September 2026
Abrasion limits for car tires
Abrasion is the rubber a tire sheds as it wears. In 2024 UN regulators endorsed a lab-drum method and a road-convoy method to measure it, and the EU's Euro 7 rules will use them. In 2026 the UN working party on tires (GRBP) endorsed a package for a new UN Regulation 186 on car-tire abrasion. New tire types would need approval under it from 30 June 2028 and all types from 30 June 2030, with a tighter second stage from 2033. The 2024 plan to set limits by September 2025 slipped.
Achieved whenthe UN's World Forum (WP.29) adopts the regulation and tire types are approved against measured limits after 30 June 2028.
Claimed by
UNECE GRBP, reported September 2026
Open
01
Diagnose failed O-rings from photos
A public set of photos of failed O-rings, each labeled with how it failed, and a model that names the failure from a phone photo and says how sure it is. Use only failures that can be made safely on a bench: compression set (the ring stays flattened), swelling from soaking in oil, heat ageing in a lab oven, cuts from fitting, abrasion and spiral twist. Include new rings as a control.
The briefCredit · unclaimed02
An open rubber recipe book
A public table of rubber recipes taken from open-access papers, with each ingredient given in "phr", parts by weight per 100 parts of rubber. Each recipe carries its measured hardness, tensile strength, stretch at break and cure time. Add a baseline model that predicts those properties from the recipe.
The briefCredit · unclaimed03
Recall check for any tire
An open table of every US tire recall since 2000 with the tire codes it covers, and a phone tool that reads the code from a sidewall photo and checks it against the table. The code is the Tire Identification Number (TIN), a 13-character code on the sidewall made of a plant code, a manufacturer code and a week-and-year date.
The briefCredit · unclaimed
Open 01 · the brief
Diagnose failed O-rings from photos
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of photos of failed O-rings, each labeled with how it failed, and a model that names the failure from a phone photo and says how sure it is. Use only failures that can be made safely on a bench: compression set (the ring stays flattened), swelling from soaking in oil, heat ageing in a lab oven, cuts from fitting, abrasion and spiral twist. Include new rings as a control.
Why it is still open
Seal makers' guides list 12 or 13 failure modes with visual signs for a person to match by eye. The only phone app found, Olypsys, measures and identifies O-ring sizes. The machine-vision tools found check that rings are present and seated on production lines, and none diagnoses failures from the field.
Built with what exists
A phone with a clip-on macro lens; Meta's DINOv2 image model (code and weights under Apache 2.0) with a small classifier trained in PyTorch on one graphics chip; a lab oven, steel clamp plates and common oils.
First step
Buy 200 identical nitrile O-rings. Photograph 50 new ones on a grey card under fixed light. Clamp 50 between steel plates at a fixed squeeze, heat them in a lab oven within their rated temperature for three days, and photograph them.
Done looks like
At least 1,200 photos across at least six classes, published under an open license with notes on how each failure was made. On 300 held-out photos, the model's labels are compared with three seal engineers' labels, and agreement is published for each class. The bar: the model agrees with the engineers' majority about as often as the engineers agree with each other.
Open 02 · the brief
An open rubber recipe book
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of rubber recipes taken from open-access papers, with each ingredient given in "phr", parts by weight per 100 parts of rubber. Each recipe carries its measured hardness, tensile strength, stretch at break and cure time. Add a baseline model that predicts those properties from the recipe.
Why it is still open
A 2022 study says "Training of such models requires data, which is equal to financial resources in industry". It built its model on 11 private blends and published no data. Rubber World reported in August 2026 that recipes are still sketched "based on experience" and then mixed, cured and tested in "round after round" until they pass. The closest open database found, NanoMine, covers polymer nanocomposites, not full curing recipes.
pmc.ncbi.nlm.nih.govrubberworld.combrinsonlab.pratt.duke.edu
Built with what exists
The OpenAlex research catalog to find papers; a small open language model fine-tuned on one graphics chip to pull recipe tables out of papers, such as Liquid AI's LFM3 (Note: "One GPU to fine-tune"); scikit-learn for the baseline; Zenodo to publish.
First step
Pick one family, natural rubber filled with carbon black. Collect 30 open-access papers and enter the recipes from 10 of them by hand, which fixes the columns and gives a checked test set for the extractor.
Done looks like
1,000 recipes from at least 100 papers, each row linked to its paper and table. A hand check of 100 random rows, with the error rate published. A baseline model whose error is reported on papers it never saw in training.
Open 03 · the brief
Recall check for any tire
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table of every US tire recall since 2000 with the tire codes it covers, and a phone tool that reads the code from a sidewall photo and checks it against the table. The code is the Tire Identification Number (TIN), a 13-character code on the sidewall made of a plant code, a manufacturer code and a week-and-year date.
Why it is still open
The industry's lookup by TIN covers "ONLY" members of the US tire makers' trade body, USTMA, about 80% of US shipments. Tire makers send everyone else to NHTSA to search by brand or model, and a November 2026 recall notice still told owners to search that way. NHTSA's public recall file has no field for TIN ranges. Phone TIN scanning is sold to businesses by Anyline and is not open.
recallinfo.ustires.orgustires.orgyokohamatire.comwbiw.comstatic.nhtsa.govanyline.com
Built with what exists
NHTSA's recall data file; the TIN format in US regulation 49 CFR 574.5; PaddleOCR (Apache 2.0) to read the text; a small model fine-tuned on one graphics chip to pull TIN ranges out of recall text, such as Liquid AI's LFM3 (Note: "One GPU to fine-tune"); the USTMA lookup as a cross-check.
First step
Download NHTSA's recall file, filter it to tire recalls, count campaigns by year, and copy the TIN ranges from 20 recent recalls by hand as a test set.
Done looks like
Every tire recall campaign in NHTSA's file from 2000 to 2026 is in the table with its TIN ranges, or flagged where the record gives none. For 200 TINs from member brands, the tool gives the same answer as the USTMA lookup. On 300 sidewall photos of loose tires, taken with permission at a shop or scrap yard, the share read correctly is published, and the tool says "cannot read" rather than guess.