Industries / Industrial equipment and services / Industrial machinery / Industrial filtration
Membranes
This work makes thin porous sheets and hollow fibers that let water or small molecules through while holding back particles, bacteria or viruses, and packs them into modules for plants.
5
Milestones achieved, from 1927 to 2017.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1927
Membrane filters become a business
Richard Zsigmondy, a chemist at the University of Göttingen, and his colleague Wilhelm Bachmann made synthetic membrane filters with set pore sizes, and Zsigmondy patented the method in 1918. On 30 May 1927 Zsigmondy, the entrepreneur Wilhelm Sartorius and other industrialists founded Membranfiltergesellschaft to develop and make them. Zsigmondy gave the firm his patents. Sartorius says the firm became the core of its bioprocess business.
Credit
Richard Zsigmondy, Wilhelm Bachmann, Wilhelm Sartorius
1969
Membranes replace the settling tank
In a membrane bioreactor (MBR), a membrane takes the place of the settling tank that separates treated water from the sludge of bacteria that clean it. Dorr-Oliver sold the first ones in the late 1960s, for sewage on ships, with flat membranes in a loop outside the tank; a 2022 review gives 1969. High membrane cost, pumping energy and fouling kept them rare.
LessonA membrane that separates well can still fail on running cost.
Credit
Dorr-Oliver Inc
1989
Membranes sunk in the tank
K. Yamamoto and co-workers in Japan hung hollow-fiber membranes inside the aeration tank and drew clean water through them by suction, so no pump had to circulate the sludge. A 2019 review by Li Ping and colleagues says this greatly cut energy use, and that Zenon in the early 1990s and Kubota after it sold immersed modules, Zenon with hollow fibers and Kubota with flat sheets.
Credit
K. Yamamoto, M. Hiasa, T. Mahmood and T. Matsuo (Water Science and Technology, 1989)
1989
Filters that remove viruses
Asahi Kasei began selling Planova, hollow fibers of cuprammonium cellulose (cellulose dissolved in a copper and ammonia solution and spun into fibers), which a review by Shoichi Ide of Asahi Kasei Medical calls the first way to remove viruses from protein solutions by membrane filtration. Makers of blood-plasma medicines now use such filters as a routine step. In 2020, member companies of the Plasma Protein Therapeutics Association pooled 754 virus validation studies and found that removal depended mainly on filter pore size and virus size.
Credit
Asahi Kasei
2017
China builds large MBRs
China's first MBR pilot, in 1999, treated 10 cubic meters of industrial wastewater a day, and its first plant of 110,000 cubic meters a day opened in Beijing in 2009. A national survey of membrane makers and operators by Li Ping and colleagues at Renmin University of China counted about 192 MBRs of at least 10,000 cubic meters a day, with a total capacity above 11.17 million cubic meters a day, by 2017.
Credit
Chinese membrane makers and plant operators; survey by Li Ping and colleagues
Watch for
Nanofiltration in hollow fibers
Nanofiltration is a tighter grade of membrane that holds back many dissolved substances, such as colour and traces of chemicals. NX Filtration, a Dutch maker, says its hollow-fiber nanofiltration plants use more than 90% fewer chemicals and more than 70% less energy than conventional technology. It says a repeat order in 2025 doubled the capacity of the largest such installation in the world, run by the water utility SAPAL in Mexico, and lists a 34 megaliter-a-day plant.
Achieved whenthe utility or another outside party publishes measured energy and chemical use for the plant against a conventional plant on the same water.
Claimed by
NX Filtration, on its website as read in November 2026
Robots cast membranes
Researchers at the University of Toronto, led by Jason Hattrick-Simpers and Jay Werber, built a platform in which a liquid-handling robot (Opentrons OT-2) mixes polymer solutions and a robot arm (UFactory xArm7) blade-casts films, immerses them to form pores and loads them into a compression tester, at less than an hour per sample. They used polysulfone with PolarClean, a less toxic solvent.
Achieved whenthe work passes peer review or another lab reproduces it.
Claimed by
Hongchen Wang and colleagues, University of Toronto, August 2025 preprint
Open
01
Close the membrane recipe loop
Add an automatic filtration test, measuring water flow and the share of a standard protein held back, to a robot membrane-casting line, and let an optimizer choose each next recipe, aiming for a fluorine-free ultrafiltration membrane that matches a commercial PVDF one.
The briefCredit · unclaimed02
A membrane autopsy library
A public library of photographs, microscope images and lab results from fouled membranes taken out of service, labeled by what blocked them (biofilm, mineral scale, silt and clay, organic film), and a model that names the foulant from an image.
The briefCredit · unclaimed
Open 01 · the brief
Close the membrane recipe loop
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Add an automatic filtration test, measuring water flow and the share of a standard protein held back, to a robot membrane-casting line, and let an optimizer choose each next recipe, aiming for a fluorine-free ultrafiltration membrane that matches a commercial PVDF one.
Why it is still open
The Toronto platform makes and tests a membrane in under an hour but measures only compression. Its authors call filtration testing time-consuming and say their future work will "close the loop". A March 2026 review by Rigoberto Advincula and Jihua Chen of Oak Ridge National Laboratory treats self-driving labs for membranes as a step still to be taken. PVDF, the usual polymer for these membranes, is a fluoropolymer, and fluoropolymer chemists Bruno Améduri and colleagues wrote in 2023 that fluoropolymers "could be impacted" by the EU restriction on PFAS proposed in February 2023.
Built with what exists
An Opentrons OT-2 liquid handler and a UFactory xArm, as in the Toronto build; a doctor blade for casting; a small stirred filtration cell with a pressure sensor and a balance logged by computer; a UV-visible spectrometer to measure bovine serum albumin held back; an open optimization library such as BoTorch or Ax; non-fluorinated polymers such as polysulfone, polyethersulfone or polyacrylonitrile, with PolarClean in place of the more toxic usual solvents, in a fume hood.
First step
Cast 10 polysulfone films by hand, measure their water flow in a stirred cell, and write the test steps as a script a robot could follow.
Done looks like
At least 100 membranes made and filtration-tested with no hands on the samples, all data published, and one recipe whose water permeance and albumin retention are within 10% of a named commercial PVDF ultrafiltration sheet in the same test cell.
Open 02 · the brief
A membrane autopsy library
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public library of photographs, microscope images and lab results from fouled membranes taken out of service, labeled by what blocked them (biofilm, mineral scale, silt and clay, organic film), and a model that names the foulant from an image.
Why it is still open
A membrane autopsy, cutting open a spent module to find what blocked it, is done one case at a time. Graciela Gonzalez-Gil, Johannes Vrouwenvelder and colleagues wrote in 2021 that information from full-scale autopsies that guides plant operation "is scant in the formal literature". KAUST researchers trained image models that classify biofilm on seawater membranes with 90% accuracy, but a search on 19 November 2026 found no open library covering other foulants.
Built with what exists
Retired spiral-wound or hollow-fiber modules from drinking-water or water reuse plants (non-hazardous feeds only); a bench stereo microscope or phone microscope; scanning electron microscope time with element analysis (SEM-EDX) at a university core lab; labeling in Label Studio; an image classifier fine-tuned in PyTorch.
First step
Ask two local water or reuse plants for modules they are replacing, and write a one-page protocol for cutting and sampling them.
Done looks like
At least 500 labeled images from at least five plants, each tied to a lab result such as loss on ignition, ATP (a measure of living cells) or EDX elements, released openly, with a classifier whose accuracy on images from one held-out plant is published.