Industries / Food, farming and forestry / Food and beverage processing / Flour milling
Sifting and blending
After each grinding, mills sieve the ground wheat to sort flour from coarser pieces and bran, send each part on to the right next machine, and blend the many flour streams into the grades they sell.
4
Milestones achieved, from 1870 to 1960.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1870
Cleaning middlings with air
Spring wheat milled in Minneapolis gave flour "speckled with bran and lacking in gluten" that sold for less than winter wheat flour. In the "New Process", millstones set high cracked the kernels, and the coarse pieces of the starchy centre, called middlings, went to a purifier: a vibrating sieve where air blasts and suction carried off the light bran and let the white middlings pass, to be ground again into flour. The Historic American Engineering Record says it was first used at large scale in 1870 at Cadwallader C. Washburn's B Mill, and that by 1875 New Process spring wheat flour was worth up to $2.25 a barrel more than traditional winter wheat flour.
Credit
Cadwallader C. Washburn's B Mill, Minneapolis; the millers La Croix and George T. Smith, who improved the purifier
1887
Flat sieves in a stack
Carl (Károly) Haggenmacher, a miller in Budapest, filed the first patent for the plansifter in 1887. It sorts the ground wheat on stacked flat sieves, where mills had used reels: long turning cylinders covered with sieve cloth. Canada's science and technology museum says it "took up less space and required less power than several machines previously used for the same functions". A Hungarian biography says it needed about a third of the floor space of the centrifugal sieves used before, and that the Wörner works in Budapest built 200 of them in about a year.
Credit
Carl Haggenmacher; the Wörner works, Budapest
1942
Britain mills at 85%
The extraction rate is the share of the wheat's weight that ends up as flour; a higher rate keeps more of the bran and germ in it. On 15 December 1942 William Mabane told the House of Commons that "the present national loaf is made from national flour which is milled at an extraction of 85 per cent", and that no synthetic vitamins were added to it. Mills had to sift so that more of the grain went into the flour.
Credit
Britain's wartime Ministry of Food
1960
Sorting flour with air
Air classification uses moving air to sort flour particles by size, at sizes far below the openings of flour sieves. In Cereal Chemistry in 1960, C. R. Jones, J. R. Fraser and T. Moran described flour from English wheats split into fine (under 17 microns), medium (17 to 35 microns) and coarse (over 35 microns) fractions, which they called high- and low-protein fractions. A micron is a thousandth of a millimetre. A 1970 study in the same journal blended air-classified fractions back to 12% protein to test them in bread.
Credit
C. R. Jones, J. R. Fraser and T. Moran
Watch for
Blending to predicted baking
Equinom says its Manna software takes the full near-infrared readings that mills already collect, adds lab results, production logs and weather and field data, and trains machine learning models to predict water absorption, dough behaviour and loaf volume. Millers could then blend wheat or flour to a predicted baking result instead of an average protein figure.
Achieved whena mill or bakery that Equinom does not pay publishes predicted against measured results for lots the models were not trained on.
Claimed by
Equinom, in an article it sponsored in World Grain, June 2026
Sieve cleaners a detector finds
Sieve cleaners are small pads that bounce on the underside of each sieve to keep its openings from clogging; a broken one can end up in the flour. Bühler says its new NovaBlue II cleaners for plansifters contain ferrite so that magnetic detection systems can find pieces, are coloured blue to be seen by eye, and carry a wear indicator that shows when to replace them.
Achieved whena test by someone other than Bühler reports the smallest broken piece that standard magnets or metal detectors on a flour line catch.
Claimed by
Bühler, July 2026
Open
01
Camera check of sieve wear
A photo method that scores how worn a plansifter's sieve cloth is, so that a mill changes sieves when they are worn and not on a calendar. A cheap microscope or a phone with a macro lens photographs the cloth; an image program measures thread smoothing, open area and holes; and the score is checked against how much flour each sieve actually lets through.
The briefCredit · unclaimed02
An open mill-flow model
An open-source program that takes a mill's flow sheet, meaning its rolls, sifters and purifiers and how stock moves between them, and predicts the weight, ash and protein of every stream. Ash is the mineral left after burning flour; more bran means more ash, and many countries grade flour by it. The program then works out which streams to blend into each flour grade.
The briefCredit · unclaimed
Open 01 · the brief
Camera check of sieve wear
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A photo method that scores how worn a plansifter's sieve cloth is, so that a mill changes sieves when they are worn and not on a calendar. A cheap microscope or a phone with a macro lens photographs the cloth; an image program measures thread smoothing, open area and holes; and the score is checked against how much flour each sieve actually lets through.
Why it is still open
Sefar, a maker of sieve cloth, tells millers that flour sieves "have to be changed approximately once per year" because the thread surface becomes smooth, the mill loses yield of first-quality flour and capacity drops (talk at the IAOM Southeast Asia meeting, Manila, October 2025). It gives no way to measure that wear on site. Bühler patented a screen carrying a printed silver track whose electrical resistance changes when the cloth tears or thins (US patent 8,607,988, priority 2007), but no published test of how well it tracks the gradual smoothing that cuts throughput was found. Image methods for screen wear found in a search are patents for oil-drilling shaker screens and papermaking fabrics.
Built with what exists
A USB digital microscope or a clip-on phone macro lens; Fiji (ImageJ) or OpenCV to measure threads and openings; a laboratory sieve shaker and one reference flour for the throughput tests.
First step
Ask a nearby mill for the frames it removes at its next sieve change, plus two new frames, and photograph the same spots on each.
Done looks like
50 sieve frames (40 used ones of known age and 10 new) photographed and tested, with a published table of wear score against measured throughput, the correlation between them, and the photographs released.
Open 02 · the brief
An open mill-flow model
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source program that takes a mill's flow sheet, meaning its rolls, sifters and purifiers and how stock moves between them, and predicts the weight, ash and protein of every stream. Ash is the mineral left after burning flour; more bran means more ash, and many countries grade flour by it. The program then works out which streams to blend into each flour grade.
Why it is still open
Grant Campbell's group has published equations for how wheat breaks on the first and second sets of rolls (Mateos-Salvador, Sadhukhan and Campbell, 2013). Kalitsis and colleagues (Cereal Chemistry, 2022) took 10 flour streams from one wheat and showed that linear programming, a standard method for choosing the best mix under limits, met French flour-grade targets that the usual ash-curve method missed, worth an estimated $6.16 per ton of flour (their figure). A Bühler-funded doctoral thesis at the University of Padova by Filippo Dal Pastro modelled single roll and sifter steps and began a whole-mill simulator, but says that "due to confidentiality reasons, most of the technical details cannot be reported". No open-source model of a whole mill was found.
Built with what exists
Python with NumPy and SciPy; the published breakage equations; SciPy's linear programming solver for blending; stream data from a university pilot mill.
First step
Code the first-break equation from the 2013 paper and reproduce one of its published particle-size curves.
Done looks like
On a milling run kept out of the fitting, predicted ash within 10% of the lab value for at least 8 of 10 flour streams, with code and data under an open licence.