Industries / Construction and property / Engineering and architecture / Geotechnical engineering
Lab testing
Lab testing measures soil samples brought in from site, such as how wet they are, how coarse or fine their grains are, how they behave as they get wetter, and how strong and how compressible they are, by set procedures so that results from different laboratories can be compared.
5
Milestones achieved, from 1915 to 2021.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1915
A falling cone measures clay
John Olsson of the Swedish State Railways' Geotechnical Commission invented the fall cone test in 1915: a cone is let fall onto clay, and the depth it sinks shows how strong the clay is. He used it to measure sensitivity, how much strength a clay loses when it is remolded, and a "fineness number" that is today called the fall cone liquid limit (the water content at which a clay stops behaving like a plastic and starts to flow).
Credit
John Olsson, Swedish State Railways
1932
A cup replaces the hand
Albert Atterberg, a Swedish chemist and agricultural scientist, set the water contents that mark a clay's change from solid to plastic to liquid, now called the Atterberg limits. In 1927 Karl Terzaghi found that Atterberg's hand-made liquid limit test did not give the same result with different operators, and asked Arthur Casagrande to fix it. In 1932, after five years of trials, Casagrande finished a metal cup that a crank lifts and drops 10 mm onto a hard base until a groove cut in the soil closes; ASTM's liquid limit standard, D4318, still uses it.
LessonA test done by hand gives the operator's answer, so the motion has to be fixed by a machine.
Credit
Arthur Casagrande, at Karl Terzaghi's request
1952
One soil classification
In 1942 Casagrande designed an Airfield Classification to teach US Army officers to judge soils for airfields. Several US government agencies adopted a modified version in 1952, and it became the Unified Soil Classification System, which sorts soils into groups with two-letter symbols, such as CL for lean clay, from their grain sizes and Atterberg limits. ASTM publishes it as standard D2487.
Credit
Arthur Casagrande; the US government agencies that adopted it
2019
Shear waves in the lab
ASTM published D8295-19, a standard method for bender elements: thin piezo-ceramic plates set into the ends of a soil specimen that send a tiny shear wave through it and pick it up at the other end. The travel time gives the soil's stiffness at very small strains, which analyses of earthquakes, machine vibration and wind or wave loading need, and because the wave does no damage the reading can be repeated during a triaxial test (a test that squeezes a cylinder of soil inside a rubber sleeve under water pressure).
Credit
ASTM Committee D18 on soil and rock
2021
Fifteen labs, one soil
Fifteen laboratories around the world tested the same sandy silt gold tailings, the ground rock left after ore is processed, to find its critical state line, which tells engineers whether a soil at a given density and pressure will tend to squeeze together or expand when it is sheared. The organizers, led by David Reid and Andy Fourie of the University of Western Australia, did not tell the laboratories which methods to use. Four of the 15 results diverged from the rest and were left out of the main comparison; the laboratories that froze each specimen at the end of the test to measure its final water content agreed best, within 0.04 in void ratio (the volume of the pores divided by the volume of the grains).
LessonSmall choices in how a specimen is made and measured can move a result, and only shared samples reveal how much.
Credit
David Reid, Andy Fourie, Riccardo Fanni and the participating laboratories
Watch for
Robots pour sand samples
Researchers at the Karlsruhe Institute of Technology (KIT) used a six-axis robot to pour sand through a single nozzle into specimen molds, making more than 80 specimens from two sands. They report, in a study in Acta Geotechnica, that density measured at more than 1,000 points was significantly more uniform and repeatable than in specimens made by hand, and Professor Hans Henning Stutz says the work "opens up perspectives for self-driving laboratories in civil engineering".
Achieved whenAnother laboratory reproduces the gain in uniformity with its own robot, or a round robin shows robot-made specimens narrowing the spread of results between laboratories.
Claimed by
Karlsruhe Institute of Technology, June 2026
Open
01
A camera judges the cup
A camera system that watches the Casagrande cup test and decides, from the video, the blow at which the groove has closed along 13 mm (0.5 in), the point a technician now judges by eye.
The briefCredit · unclaimed02
Physics checks for lab data
An open checker that reads laboratory results in AGS files and flags values that cannot all be true at once, such as a degree of saturation above 100 %, a plasticity index larger than the liquid limit, or a group symbol that does not match the reported limits, with each flag citing the rule it broke.
The briefCredit · unclaimed
Open 01 · the brief
A camera judges the cup
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A camera system that watches the Casagrande cup test and decides, from the video, the blow at which the groove has closed along 13 mm (0.5 in), the point a technician now judges by eye.
Why it is still open
The cup was built because Atterberg's hand method gave each operator a different result, yet the end of Casagrande's test is still judged by eye. The closest published work, a 2023 Georgia Southern University conference paper, estimated the liquid limit of kaolin clay from video of the soil's vibrations rather than from the groove. This research found no published system that judges groove closure from images.
Built with what exists
A phone or USB camera fixed above a motorized cup; OpenCV to find the groove; a small image model trained on frames that people have labeled.
First step
Film 30 cup tests from above in a teaching laboratory, and have two people mark the closing blow in each video.
Done looks like
On three clays, each tested by ten technicians, the camera's call agrees with the median technician within one blow in at least 90 % of tests, and liquid limits from the camera's calls spread less between technicians than those from their own calls; code and videos published.
Open 02 · the brief
Physics checks for lab data
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open checker that reads laboratory results in AGS files and flags values that cannot all be true at once, such as a degree of saturation above 100 %, a plasticity index larger than the liquid limit, or a group symbol that does not match the reported limits, with each flag citing the rule it broke.
Why it is still open
The checkers for AGS files test their structure, not whether the numbers make sense, as a 2019 Geodrilling International article notes; the AGS4 validator's own guide says it "does not validate the items of data" and leaves that to "bespoke data quality assessment specifications". In a 2024 paper in Acta Polytechnica CTU Proceedings, V. Pavelcová and T. Janda set out rules for checking the laboratory data sent to ExCalibre, a web tool that calibrates soil models, but those rules serve that tool's tests. This research found no open checker for routine laboratory results in AGS files.
geodrillinginternational.comags4-validator-desktop-app-43aa52.gitlab.ioojs.cvut.cz
Built with what exists
Python; the AGS4 data dictionary; a language model to explain each flag in plain words, citing the rule behind it (see the Note "Findings that cite notes").
First step
Write the first ten rules and run them on one AGS file that a ground investigation contractor agrees to share.
Done looks like
Run over at least 1,000 laboratory records from at least three shared projects; a person checks 200 flagged and 200 unflagged records; published precision of at least 90 % and a count of real errors found.