Industries / Mining and materials / Mining and quarrying / Frac sand
Sizing and testing
Sizing screens dry sand into grades named for the sieves that bound them, such as 40/70, and testing checks each grade's shape, strength and cleanliness by the industry's standard methods.
6
Milestones achieved, from 1963 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1963
A chart for grain shape
Krumbein and Sloss published a chart of drawn grains for judging by eye a grain's sphericity, how close it is to a ball, and its roundness, how smooth its corners are. Frac sand standards still use the chart and ask for a score of 0.6 or more on both.
Credit
Krumbein and Sloss, cited by Mary Ellen Benson and Anna B. Wilson of the USGS
1983
Standard tests for frac sand
A committee organized by the American Petroleum Institute (API) published Recommended Practice 56, standard ways to test frac sand, including a crush test in which sieved sand is loaded into a steel cell, pressed to a set stress and sieved again to weigh the broken fines. By 2009 Terrence Palisch and four co-authors warned that many engineers had forgotten the test's intent, to show the stress at which crushing becomes excessive, and misapplied its results when choosing sand.
Credit
American Petroleum Institute
2006
One international test standard
ISO 13503-2, adopted by API as RP 19C in 2008, replaced API's separate practices with one set of methods; the USGS says it replaced RP 56 and RP 58, and the instrument maker HORIBA says RP 56, 58 and 60. It rates strength by the K-value, the highest stress in thousands of pounds per square inch (psi) at which a sample makes no more than 10% fines by weight, and asks frac sand to lose no more than 3% of its weight in acid and to read no more than 250 nephelometric turbidity units (NTU), a measure of the cloudiness left by clay and silt.
Credit
International Organization for Standardization (ISO) committee TC 67; American Petroleum Institute
2019
Median size beats mesh labels
Within a standard grade the largest grains can be twice the size of the smallest, so two sands with the same label can perform differently. R. D. Barree, R. J. Duenckel and B. T. Hlidek reported from thousands of laboratory conductivity tests, which measure how easily fluid flows through a packed layer of grains, that the median grain diameter predicts conductivity well, so non-standard blends can work and plants that sieve less finely can ship more; that year Black Mountain Sand sold a 40/140 blend it called "coarse 100 mesh," and ConocoPhillips' Eagle Ford team said it saw too little difference among 100-mesh sources to pay for a particular one.
Credit
R. D. Barree, R. J. Duenckel, B. T. Hlidek; ConocoPhillips
2019
China swaps ceramic for sand
Ceramsite, a ceramic proppant, is stronger and costlier than quartz sand. PetroChina researchers found that grains in horizontal wells bear only 50% to 60% of the stress they bear in vertical wells, and report that in the fields they studied, including Xinjiang's Mahu shale oil and southwest China's shale gas, quartz sand rose from under 30% of proppant in 2014 to 69% in 2019, cutting yearly costs by more than 1 billion yuan with no effect on production.
Credit
Bo Cai and nine colleagues, PetroChina Research Institute of Petroleum Exploration and Development and PetroChina Xinjiang Oilfield Company
2024
Labs disagree on crush
I. J. Renkes, G. M. Money and A. E. Garza split 30/50, 40/70 and 100-mesh sand from a plant in Kermit, Texas, among four independent laboratories under changing sample labels, with each sample tested 30 times by API 19C. Fines differed by up to 70% between laboratories and up to 30% within one, and the spread grew as the sand got finer; in 2014 J. Getty and C. R. Bulau had found gaps between laboratories of up to 2,000 psi and traced much of the noise to how the cell is loaded.
LessonThe authors concluded that the standard crush test, often used to choose a supplier, "is not serving the industry's needs.".
Credit
I. J. Renkes, G. M. Money, A. E. Garza; J. Getty, C. R. Bulau
Watch for
Image size for shape scores
A dataset classifies 7,874 grains of rounded Ottawa sand and angular Jumunjin sand into Krumbein and Sloss's roundness and sphericity classes by two computer methods at five image sizes, and its authors say the smallest grain image that gives a reliable score differs by method and by grain shape.
Achieved whena peer-reviewed paper sets out the minimum image sizes and another laboratory reproduces them on frac sand.
Claimed by
Young-Woo Song (Korea Railroad Research Institute), Wooyoung Choi and Young-Hoon Jung (Kyung Hee University), September 2026
Open
01
People versus software on shape
A public set of microscope images of frac sand grains from several mines and grades, each grain scored on the Krumbein and Sloss chart by at least five trained people and by open software, so the spread among people and the gap between people and software can be measured.
The briefCredit · unclaimed02
Median size for every product
A free public table of the median grain diameter, called d50, and the full sieve results of every frac sand product whose maker publishes a data sheet, grouped by labeled grade, showing how far products with the same label differ.
The briefCredit · unclaimed
Open 01 · the brief
People versus software on shape
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of microscope images of frac sand grains from several mines and grades, each grain scored on the Krumbein and Sloss chart by at least five trained people and by open software, so the spread among people and the gap between people and software can be measured.
Why it is still open
The chart is still applied by eye, which HORIBA says "results in large subjective differences from operator to operator". Xu and colleagues in 2024 checked their software against the chart's own drawings, not against people scoring real sand, and the 2026 Ottawa and Jumunjin dataset compares two computer methods with no human scores.
Built with what exists
A USB digital microscope or a flatbed scanner; Meta's Segment Anything model to outline grains; OpenCV and scikit-image; the computational-geometry method of Zheng and Hryciw; a web form for the human scorers; Hugging Face Datasets to publish. Wear a respirator rated N95 or better when pouring dry sand.
First step
Buy three graded sands, photograph 200 grains of each under the microscope, and score 50 of them by hand to time the work.
Done looks like
At least 2,000 grains from at least six products (20/40, 40/70 and 100 mesh, from northern white and in-basin mines), five human scores per grain, and published figures for agreement among people (intraclass correlation) and for the average gap between the software and the people's median score, with images and code.
Open 02 · the brief
Median size for every product
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free public table of the median grain diameter, called d50, and the full sieve results of every frac sand product whose maker publishes a data sheet, grouped by labeled grade, showing how far products with the same label differ.
Why it is still open
Barree, Duenckel and Hlidek tied conductivity to d50 using thousands of laboratory tests that are not published as open data, and makers publish data sheets one product at a time; no compiled public table turned up on Hugging Face, Zenodo or the open web, only paid market reports.
Built with what exists
A browser agent such as Browser Use to collect data sheets (see the Note "Deleting 80% to start over"); a language model to read the sieve tables in the PDFs, with a share checked by hand; Python to find d50, the fiftieth mass percentile as API 19C defines it, from each cumulative sieve curve.
First step
Download data sheets from ten producers, compute d50 for each product by hand, and compare it with the model's reading.
Done looks like
At least 100 products from at least 30 mines in at least three countries, each d50 linked to its source sheet, a hand check of 10% of entries with no copying errors, and the range of d50 within each labeled grade published.