Industries / Food, farming and forestry / Food and beverage processing / Rendering
Cooking and separation
Cooking and separation is the plant work of grinding raw by-products, cooking the water out of them, and splitting what is left into liquid fat and a solid that is pressed, dried and milled into meal.
6
Milestones achieved, from 1900 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1900
Screw press squeezes out fat
Valerius D. Anderson patented a continuous screw press on 10 April 1900 and sold it as the Expeller. Anderson's company calls it the first continuous mechanical screw press: it pressed oil and fat out of material fed in without stopping, where earlier presses were loaded and emptied one charge at a time. The National Renderers Association's history says hand presses (first described in 1829) gave way to hydraulic presses about 1850 and then to Anderson's screw press. Anderson's company says the Expeller was first used for linseed oil and was later adapted to more than 80 oil- and fat-bearing materials.
patents.google.comandersonintl.comassets.nationalrenderers.org
Credit
Valerius D. Anderson, V. D. Anderson Company (now Anderson International)
1920s
Dry cookers keep the protein
Early renderers boiled material in open kettles or injected live steam into sealed tanks ("wet rendering") and skimmed the fat. Long contact with water raised the fat's free fatty acids, a sign of breakdown, and the leftover "stickwater", which carried dissolved protein, went to sewers and rivers. By the NRA's account, the dry rendering cooker introduced in Germany in the 1920s, a steam-jacketed drum that cooks material in its own juices, let the industry sell good protein as well as fat, and dry rendering then replaced wet rendering.
Lessonwater added to the cook has to be removed again, and what is thrown away with it is lost protein and a pollution problem.
Credit
German equipment makers, not named by the source; account by the NRA's Essential Rendering (2006)
Early 1960s
Cooking goes continuous
In the late 1950s George Epsy, a maintenance worker at Baker Commodities in Los Angeles, suggested to the owner, Frank Jerome, that ground raw material could be cooked while moving through steam-heated tubes. With Jack Keith of Keith Engineering they built a prototype, and the first single-vessel continuous cooker went into Denver Rendering Company in the early 1960s. Dupps and Keith Engineering then built the DUKE system, sold today as the Equacooker. In the NRA's 2006 estimate, US plants ran 219 continuous tube and disc cookers against 41 batch cookers.
Credit
George Epsy, Frank Jerome, Jack Keith; The Dupps Company
Late 1970s
Fish meal methods arrive
Atlas and Stord-Bartz, makers of fish meal plants, brought disc dryers, evaporators run on waste heat and mechanical vapor recompression (a compressor that raises waste steam back to useful pressure) to North American rendering. Low-temperature separation, taken from fish meal making, cut energy use on wet raw material and gave lighter, cleaner fat.
Lessonby the NRA's account, the fastest slurry evaporators saved the most energy but did poorly in European tests of BSE infectivity reduction, because material spent so little time in them; a design is judged on safety as well as fuel.
Credit
Atlas, Stord-Bartz (later part of Haarslev)
2020
Two die clearing a hydrolyzer
A hydrolyzer is a pressure vessel that cooks feathers with steam until their protein can be digested. On 10 August 2020 a hydrolyzer at Darling Ingredients' poultry rendering plant in Byram, Mississippi, clogged with pressure trapped inside. Maintenance workers loosened a bottom flange to let the steam out; the cover broke free and two men died of burns. OSHA cited the company for lockout procedures that did not "clearly and specifically outline" how to work on the machine, and the US Court of Appeals for the Fifth Circuit upheld the citation on 6 October 2023.
Lessona clogged pressure vessel needs its own written steps for releasing stored steam before anyone opens it.
Credit
OSHA investigation; Fifth Circuit opinion No. 22-60466
2021
Rendering's energy share measured
A survey of Australian red meat plants for the Australian Meat Processor Corporation (AMPC) and Meat & Livestock Australia, by All Energy Pty Ltd, found that rendering used about 37% of the energy of meat processing in 2019–20, about 1,200 MJ per tonne of hot carcass weight. The 2024 RACE for 2030 study that reports this puts rendering's use at 4.25 PJ a year, mostly from gas-fired steam.
Credit
All Energy Pty Ltd for AMPC and MLA; reported by Nihill and colleagues, RACE for 2030 CRC
Watch for
Cookers run by learned control
Haarslev Industries says its CORE advanced process control learns from a plant's own process history and adjusts key equipment, such as cookers and presses, in real time, cutting process variation by 30 to 70% with an average payback of about 12 months; a June 2026 Haarslev article repeats that CORE "typically reduces process variation".
Achieved whena renderer or an outside engineer publishes before-and-after logs from a named line showing the drop in variation and in energy per tonne.
Claimed by
Haarslev Industries, June 2026
Low-temperature line at Oakey
NH Foods Australia is replacing a 50-year-old rendering plant at its Oakey beef plant in Queensland with a continuous low-temperature line of up to 12 tonnes an hour. It estimates 2 to 5% more meal and tallow recovered and about 9,600 tonnes of CO2-equivalent a year less, with completion due in the first half of 2026.
Achieved whenthe line is running and its measured recovery and emissions are reported by someone outside the company, such as an auditor or AMPC's next environmental performance review.
Claimed by
NH Foods Australia, March 2026
Open
01
An open rendering energy model
A free, open-source model of the steam and electricity a rendering line uses, covering the main designs: batch dry cookers, continuous dry cookers, low-temperature wet lines with a decanter (a spinning separator) and a stickwater evaporator, and slurry evaporators. A plant enters its raw material (water, fat, protein and bone), its throughput and its fuel prices, and compares gas boilers, electric boilers and heat pumps with mechanical vapor recompression.
The briefCredit · unclaimed02
Cooker end point from logs
An open dataset of cooker runs from working plants, with logged jacket steam pressure, product temperatures, agitator current, feed rate, and the lab moisture and fat of the cooked material and the press cake, plus a model that predicts moisture at discharge and is compared with the operator's own call.
The briefCredit · unclaimed
Open 01 · the brief
An open rendering energy model
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free, open-source model of the steam and electricity a rendering line uses, covering the main designs: batch dry cookers, continuous dry cookers, low-temperature wet lines with a decanter (a spinning separator) and a stickwater evaporator, and slurry evaporators. A plant enters its raw material (water, fat, protein and bone), its throughput and its fuel prices, and compares gas boilers, electric boilers and heat pumps with mechanical vapor recompression.
Why it is still open
The 2024 RACE for 2030 study built its own model, but says the plant data used to check it "are obscured here due to confidentiality concerns", that "accurate and up to date data on industry trends" is lacking, and that a further analysis should come before any pilot. The NRA's handbook notes that single-plant renderers "must continually compare only against their past performance". A GitHub search in this session found no public rendering energy model.
Built with what exists
Python with the CoolProp library for steam properties; published check figures: the NRA's 0.76 pounds of steam per pound of water evaporated (theoretical best) to 1.50 (poor), and 800 to 1,500 BTU per pound of water; RACE's coefficient of performance (heat delivered per unit of electricity) of 3.2 for dry and 1.8 for wet rendering; Australia's 1,200 MJ per tonne of carcass weight.
First step
Code the steam balance of one continuous dry cooker and check it against the NRA's 0.76 pounds of steam per pound of water.
Done looks like
A public repository whose results match RACE's two coefficients and the NRA's steam range within 10%, and runs checked by engineers at three or more plants against their own meter readings, with the differences published.
Open 02 · the brief
Cooker end point from logs
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open dataset of cooker runs from working plants, with logged jacket steam pressure, product temperatures, agitator current, feed rate, and the lab moisture and fat of the cooked material and the press cake, plus a model that predicts moisture at discharge and is compared with the operator's own call.
Why it is still open
The NRA's handbook lists end-point control, stopping the cook when moisture is low enough for good pressing but before the protein is overcooked, among the four checks that set quality, and operators make that call. Haarslev's CORE learns from plant data but publishes neither data nor an outside test. The one peer-reviewed in-line study found, by Garrido-Varo and colleagues (2018), measured 346 meal samples with a near-infrared probe on a test conveyor, after cooking, not from cooker logs. Searches of GitHub and Europe PMC in this session found no public cooker dataset.
Built with what exists
Exports from the plant's existing data historian; routine lab moisture (oven) and fat (Soxhlet) results; Python with scikit-learn or XGBoost. Nothing in the plant's controls is changed.
First step
Ask one renderer for three months of historian exports and the matching lab moisture results, and check that the timestamps line up.
Done looks like
At least 300 runs from at least two plants published with their lab results, and a model that predicts discharge moisture within 1 percentage point on runs from a plant it was not trained on, scored against operators' calls on the same runs.