Industries / Construction and property / Heavy civil / Asphalt production
Quality testing
Quality testing designs a mix in the laboratory, setting how much binder it needs and checking that it will resist ruts and cracks, and then tests samples from the plant against that design.
7
Milestones achieved, from 1905 to 2020.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1905
Voids counted in the mix
Clifford Richardson, owner of the New York Testing Company, set out how to make asphalt mixes in "The Modern Asphalt Pavement" (1905). Mixes were then judged largely by eye, as in the pat-paper test, in which hot mix was tipped onto brown grocery paper and the binder's spread judged visually. Richardson also calculated voids in the mineral aggregate (VMA), the space between the stones that binder and air share. VMA became a design criterion in the Asphalt Institute's mix design manual, MS-2, only in 1962, after Norman McLeod argued for it in papers from 1956 to 1959.
LessonA measure in print since 1905 took 57 years to become a rule.
Credit
Clifford Richardson; Norman McLeod
1932
Binder set by surface area
Francis Hveem became a resident engineer for the California Division of Highways in 1927. By 1932 he had a method that set how much binder a mix needs from the surface area of its stone and sand, using factors published by the Canadian engineer L. N. Edwards in 1918, and he paired it with a stability test on his own device, the stabilometer.
Credit
Francis Hveem, California Division of Highways
1943
A test for wartime airfields
Bruce Marshall of the Mississippi highway department developed a mix design method in the 1930s and early 1940s. It compacted samples with a drop hammer instead of a hand rammer and calculated their air voids. In 1943 the US Army Corps of Engineers in Vicksburg, Mississippi, hired him, and during the Second World War the Corps used his portable method to design asphalt for airfields. State highway departments took it up after the war, and the 1974 edition of MS-2 made it the standard in practice.
Credit
Bruce Marshall; US Army Corps of Engineers
1993
Mixes designed by gyration
SHRP, which ran from 1987 to 1993, planned a mix design system in three levels, two of them built on performance tests meant to predict rutting and cracking. Those tests could not be put into practice, and SHRP's performance-based mix specification was never used by state agencies. What came into use was Level 1, volumetric design: samples are compacted in a gyratory compactor, a press that kneads the sample by turning it under load, and the binder content is set from air voids, VMA and the share of voids filled with binder. It replaced the Marshall and Hveem methods as the main US method from the mid-1990s and is specified in AASHTO M 323.
LessonThe part that was ready shipped, and performance tests waited 27 years, until AASHTO's balanced mix design practice of 2020.
Credit
SHRP; FHWA's technical working groups
1990s
Binder burned, not dissolved
To check how much binder a plant sample holds, laboratories used to dissolve it out with hazardous solvents. In the mid-1990s NCAT developed the ignition method, which burns the binder off a loose sample in an oven and weighs what is lost. Laboratories took it up quickly, and as AASHTO T 308 it is required or allowed by most US state transportation departments. Each mix and each oven needs its own correction factor. In 2019 NCAT reported that these factors vary widely even between ovens of the same brand, that some aggregates such as dolomite lose mass when heated, and that some agencies and contractors share factors between ovens against the standard.
LessonA faster, cleaner test still has to be calibrated to each machine.
Credit
NCAT, Auburn University
2017
A one-minute cracking test
Fujie Zhou, Soohyok Im, Lijun Sun and Tom Scullion, mostly of the Texas A&M Transportation Institute (TTI), published IDEAL-CT, a cracking test that squeezes a cylinder of mix 150 mm across and 62 mm high along its diameter at 50 mm a minute, at room temperature. It needs no cutting, gluing or notching, runs on ordinary strength-test equipment and finishes in under a minute; the authors report that its results matched cracking measured on FHWA's accelerated loading facility, two Texas highways and the MnROAD test road. ASTM published it as standard D8225 in 2019.
LessonSuperpave's stiffer binders and volumetric checks were the answer to the rutting of the 1980s, but they carried no cracking test, and in 2024 Gerry Huber of Heritage Research Group wrote that cracking "has become THE ISSUE".
Credit
Fujie Zhou and colleagues, TTI
2020
Mixes judged by performance
Balanced mix design (BMD) designs a mix with performance tests on suitably aged samples that address several kinds of damage, such as rutting and cracking, with allowance for traffic, climate and the mix's place in the pavement. An FHWA expert task group first defined it in 2015, and AASHTO published it as standard practice PP 105-20 in 2020. Oklahoma completed four phases over seven years, with 33 pilot and working projects, and now uses IDEAL-CT statewide alongside the Hamburg wheel-tracking test, which rolls a loaded wheel over a sample to measure rutting.
Credit
FHWA's BMD task force; AASHTO; Oklahoma Department of Transportation
Watch for
A robot runs the tests
TTI built AMAZE (Asphalt Mixture Automated Testing System with Zero Intervention) for the Texas Department of Transportation, which cited trouble hiring laboratory staff and the risk of injury from hot asphalt, saws and test machines. A robot arm moves each specimen between units that measure air voids, bring it to temperature and run IDEAL-CT, a rutting test and an indirect tensile strength test. TTI reports that two runs of the robot gave almost the same air voids on seven specimens, while two experienced technicians disagreed more on two of them. Humboldt now sells the system as a new product and says it "removes human errors".
Achieved whena state agency or a laboratory outside TTI and Humboldt publishes a side-by-side comparison with technicians on plant-produced samples, or an agency accepts its results for acceptance testing.
Claimed by
TTI (project completed August 2023) and Humboldt (product page seen November 2026)
Open
01
Cracking models tested across labs
One open table of published IDEAL-CT results, each with the mix behind it (stone grading, binder grade and content, recycled content, air voids, aging), pooled from many laboratories, and a scoring rule that trains a model on some laboratories and tests it on the others.
The briefCredit · unclaimed02
Open air-void robot cell
An open, documented robot cell that runs the bulk specific gravity test on room-temperature specimens (weighing a compacted sample dry, under water and surface-dry, to work out its air voids), with a public log comparing it with technicians.
The briefCredit · unclaimed
Open 01 · the brief
Cracking models tested across labs
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
One open table of published IDEAL-CT results, each with the mix behind it (stone grading, binder grade and content, recycled content, air voids, aging), pooled from many laboratories, and a scoring rule that trains a model on some laboratories and tests it on the others.
Why it is still open
Published models are fitted and tested within one laboratory's data. Lan Ngoc Nguyen and colleagues (PLOS ONE, 2023) trained on 107 records from one laboratory in Hanoi and tested on a random split of the same records, and they say more data are needed. Small tools on GitHub, such as pysphalt for Brazilian data, follow the same pattern; this research found no pooled set from several laboratories.
Built with what exists
Python, pandas and scikit-learn; an AI document reader such as Claude to pull tables out of state agency reports and theses; ASTM D8225 to check each record's test conditions.
First step
List the state BMD reports that print CT Index values with mix details, starting with Oklahoma's, and transcribe one.
Done looks like
At least 1,500 IDEAL-CT records from at least 10 laboratories in at least 3 countries, each linked to its source, and published leave-one-laboratory-out scores for at least three published model types against a simple baseline.
Open 02 · the brief
Open air-void robot cell
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open, documented robot cell that runs the bulk specific gravity test on room-temperature specimens (weighing a compacted sample dry, under water and surface-dry, to work out its air voids), with a public log comparing it with technicians.
Why it is still open
The one such cell found, TTI's AMAZE, is sold as a closed product, and its published comparison with technicians covers seven specimens.
Built with what exists
A collaborative robot arm able to lift a 150 mm compacted specimen; a laboratory balance with an under-balance hook and a room-temperature water bath; a damp-towel station for surface drying; a camera; robot learning software such as Hugging Face's open LeRobot library to teach the drying motion from recorded demonstrations, or NVIDIA's open GR00T 2 robot model, which by NVIDIA's own test succeeds 95% of the time after five demonstrations of a task (the Note "Five demonstrations per task").
First step
Film a technician running the test on 10 specimens and time each step.
Done looks like
100 specimens, each measured twice by the cell and twice by each of two technicians; the cell's two air-void readings within 0.2 percentage points of each other on at least 95 specimens, and its average within 0.3 points of the technicians' average; all data, drawings and code published.