Industries / Industrial equipment and services / Industrial machinery / Industrial filtration
Liquid filtration
This work makes the presses, drum filters, cartridges and sterilizing filters that take solids or germs out of liquids in factories, mines and drug plants.
6
Milestones achieved, from 1853 to 1999.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1853
The press filter
Potteries used to drive water out of clay slip by boiling it. Grace's Guide, citing The Practical Mechanic's Journal of 1867, credits the London engineers Needham and Kite with inventing the press filter in 1853. At the 1862 London Exhibition the firm showed a patent filter press with 600 square feet of drainage area, worked at 10 to 100 pounds per square inch, and said it was used by the largest potteries, brewers, oil refiners and colour makers in the United Kingdom.
Credit
Needham and Kite, Phoenix Iron Works, Vauxhall
1909
A filter that never stops
A press works in batches: fill, squeeze, open, empty. Edwin Letts Oliver of Grass Valley, California, patented a drum that turns partly submerged in a slurry; suction draws the liquid through the cloth on its surface and air pressure pushes the solid cake off, so filtering and washing run without a break. His patent names ores, tailings, cyanide solution, sugar and wood pulp. US patent 919,628 was filed on 8 September 1908 and granted on 27 April 1909.
Credit
Edwin Letts Oliver
1958
Presses that empty themselves
Opening a plate press and scraping out each cake was slow hand work. Kenichiro Kurita and Seiichi Suwa of Kurita Machinery in Japan built an automatic horizontal filter press that improved cake removal. Wikipedia calls it the world's first and dates it to 1958; McLanahan, a US press maker, gives 1959.
Credit
Kenichiro Kurita and Seiichi Suwa, Kurita Machinery Mfg
1967
A smaller germ slips through
The first membranes made in the United States for sterilizing liquids were rated at 0.45 micrometers (µm, thousandths of a millimeter). In 1967 Bowman and colleagues found a small bacterium, then called Pseudomonas diminuta, that these filters did not hold back. The rating for sterilizing filters was halved to 0.22 µm, or 0.2 µm by some makers, according to a filtration text published by PDA, the Parenteral Drug Association.
LessonA filter's rating holds only for the organisms it has been tested against.
Credit
Bowman and colleagues
1987
A test defines sterile
"Sterilizing grade" got a number: the filter must hold back at least 10 million colony-forming units (live bacteria able to grow) of B. diminuta per square centimeter of filter area. The Health Industry Manufacturers Association recommended this in 1982, and the US Food and Drug Administration (FDA) adopted it as the definition in 1987, according to the PDA text. ASTM International's test method F838 sets out the challenge, and FDA's 2004 guidance on aseptic processing names B. diminuta (ATCC 19146), about 0.3 µm across, as the usual test organism for 0.2 µm filters.
Credit
Health Industry Manufacturers Association; US FDA; ASTM International
1999
Shrunken bacteria get through
S. Sundaram and colleagues showed that Ralstonia pickettii, a common water bacterium, passed 0.2 µm nylon and 0.22 µm PVDF sterilizing filters in drug solutions. In a 2001 study, bacteria from a natural water source passed eight 0.2 and 0.22 µm filter types from four makers, all rated to retain B. diminuta. The PDA text counts some 25 such cases and links them to cells that shrink in nutrient-poor liquids.
LessonThe standard test organism is a model, not a guarantee; a filter for a given drug has to be challenged with what can live in that drug.
Credit
S. Sundaram and colleagues
Watch for
Robots change press cloths
Diemme Filtration, an Italian press maker in the Aqseptence Group, says its X-Change robot, carried on an overhead crane, replaces filter cloths unattended on its GHT2500F and GHT5000F presses, with "a total cloth replacement of around 10 minutes". It says its Optycare camera system scans a whole cloth in less than 20 seconds at 0.2 mm per pixel and uses a machine-learning model to find defects, and that the first Optycare was installed by May 2026.
Achieved whena plant operator or another outside party publishes measured cloth-change times and defect-detection rates from months of running.
Claimed by
Diemme Filtration, at a webinar reported in July 2026
Open
01
An open cloth defect library
A public, labeled set of images of used filter-press cloths showing tears, pinholes, worn seams and blinding (pores clogged with fine solids), with a baseline model that finds and sizes each defect.
The briefCredit · unclaimed02
A table of filter failures
One open table of every published case in which a sterilizing-grade filter let bacteria through, giving the organism, the liquid, the filter material and rating, pressure, run time, challenge level and outcome, each row linked to its source.
The briefCredit · unclaimed
Open 01 · the brief
An open cloth defect library
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, labeled set of images of used filter-press cloths showing tears, pinholes, worn seams and blinding (pores clogged with fine solids), with a baseline model that finds and sizes each defect.
Why it is still open
Camera inspection of press cloths exists inside one maker's product. Diemme, which built its scanner with the University of Bologna, trains its model on its own scans, its first unit was installed in 2026, and its FILTECH 2026 paper says cloth monitoring still relies heavily on manual inspection. Sameer Prabhu and colleagues at Luleå University of Technology, studying the cloth on an iron ore mine's rotary drum filters in 2025, had to make up their failure images, because preventive maintenance meant real failures could not be collected. Searches of Europe PMC, arXiv and Crossref on 19 November 2026 found no public set of labeled cloth images.
Built with what exists
Used polypropylene cloths from plants that dewater non-hazardous slurries such as quarry fines, clay or drinking-water sludge; a phone camera or a line-scan camera on a slide rail with low-angle light; labeling in CVAT or Label Studio; an open detection or segmentation model such as YOLO, trained in PyTorch.
First step
Ask a local quarry or water treatment plant for the cloths it throws away at its next change, and photograph 50 of them under side light.
Done looks like
At least 1,000 labeled cloth images from at least three plants, released under an open licence, and a baseline model whose recall and false-alarm rate for defects of 1 mm and larger are reported on cloths from a plant held out of training.
Open 02 · the brief
A table of filter failures
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
One open table of every published case in which a sterilizing-grade filter let bacteria through, giving the organism, the liquid, the filter material and rating, pressure, run time, challenge level and outcome, each row linked to its source.
Why it is still open
The cases are spread across journal papers from 1980 onward. The PDA's filtration text points to some 25 cases gathered by Sundaram and colleagues in 2001, and later studies report new ones one at a time. Searches on 19 November 2026 found single studies but no maintained public table.
Built with what exists
The free Europe PMC and Crossref interfaces to pull candidate papers; a large language model to draft rows from abstracts and full texts, with every row checked by a person against the paper; a CSV file published on GitHub or Zenodo with a DOI.
First step
Export all Europe PMC records that mention bacteria passing 0.2 or 0.22 µm rated filters and enter the first 30 into the table.
Done looks like
At least 50 cases, each with a source link and a second reader's check, published with a short summary of which liquids, filter materials and conditions recur.