Industries / Consumer goods / Candles and home fragrance
Pouring and moulding
This is the work of turning wax into finished candles: dipping wicks, pouring wax into moulds or containers, pressing powdered wax or pushing it through a die, then cooling, trimming and checking each candle for defects.
7
Milestones achieved, from 1462 to 2025.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1462
Guild power over bad candles
A charter from Edward IV, dated 8 March 1462 in the Privy Council's register, gave London's Tallow Chandlers' Company the right to seize and destroy inferior goods of its trade, so a guild, not the buyer, policed candle quality in the City. The company's own timeline records that in 1709 a shop owner won damages after it broke his candles, and that its searches were suspended the following year.
LessonQuality policing by a trade body ended once a court found it had gone too far.
Credit
Worshipful Company of Tallow Chandlers
1823
Cooled moulds with pistons
Candles had been made by "dipping the wick into molten tallow, or pouring tallow over a hanging wick", in the Science Museum's words. Water-cooled moulds, first patented by Thomas Binns in 1801, and a piston system that pushed set candles out, first used by Joseph Morgan in 1823, turned moulding into a batch process run by machine. Grace's Guide says Morgan, a Manchester pewterer and candle-mould maker, patented his machine in 1834, and quotes a claim that with it "three men and five boys" could make two tons of candles in twelve hours.
Credit
Thomas Binns; Joseph Morgan
1888
Harder waxes suit machine moulds
Tallow is soft and hard to release from moulds. After 1850, harder fuels such as stearin and paraffin made it possible to mass-produce uniform candles in moulding machines. A half-size working model made in 1888 by Price's Patent Candle Company of Battersea, London, moulds 96 candles at each cast, and the Science Museum says larger machines managed up to 500.
Credit
Price's Patent Candle Company
1955
Cold extrusion around a wick
Daniel James Keogh of Price's patented a way to push cold, solid wax through an opening around a moving wick, forming a continuous wax rod that is cut into candles, with no melting and no cooling. The patent, granted in the US in 1955 from a 1950 British filing, reports about 300 small candles a minute. Extrusion is now a standard method: a 2014 patent by another firm, Elevance Renewable Sciences, describes wax "typically introduced into the extruder in prilled form", meaning as small beads, and then sliced to length.
Credit
Daniel James Keogh, Price's Patent Candle Company
1972
Pressing powder instead of pouring
A Hallmark Cards patent on pressing candles from powdered wax, filed in 1970 and granted in 1972 to the inventor Ronald R. Renoe, said cooling times of up to 12 hours were common for large cast candles, and that pressing cut the process to "substantially less than one minute". A Swedish design by Hugo Karlsson, from 1974, fed the wick through the pressing piston. Bolsius, the Dutch maker, says Lambertus Bolsius invented the pressed tealight in 1959, a claim no outside source opened here confirms. Pressing is now routine: the 2014 Elevance patent describes pressing wax powder at 1,000 to 4,000 psi and pushing the candle out of the mould.
patents.google.compatents.google.comen.bolsius.compatents.google.com
Credit
Ronald R. Renoe, Hallmark Cards; Hugo Karlsson; Bolsius (its own claim)
2021
Cobots stack tealight packs
At Gala Germany in Wörnitz, part of the Gala Group, one line makes about 30,000 tealights an hour and packs about 1,000 packs an hour, according to a German trade blog on robots. A collaborative robot, or cobot, one built to work beside people without a cage, took over stacking the packs onto pallets at the end of the line; it was accepted in January 2021 and a second followed in October 2021. The site runs 15 robots in all.
Credit
Gala Germany GmbH; Bayer, the integrator that installed the cobots
2025
EU output falls, duties follow
The European Commission found that EU candle production fell from 679,505 tonnes in 2021 to 484,048 tonnes in the year to September 2024, with plants using 45% of their capacity, down from 55%, while Chinese imports rose from 13% to 22% of EU consumption. It imposed provisional anti-dumping duties of up to 70.9% in August 2025, and the Brussels Times reported definitive duties of 56 to 60% in February 2026. The same regulation quotes a Chinese local plan saying the Dongwang candle cluster in Hebei supplies 70% of China's home market.
LessonThe EU industry had about twice the capacity it used.
Credit
European Commission, after a complaint by EU producers (more than 80 companies employing about 9,597 people directly, by its count)
Watch for
Robotic pouring with camera checks
A Chinese exporter says its lines use "robotic pouring arms with millimeter-level accuracy", centre wicks within 0.5 mm, hold fill weights within 2 g either way, and use cameras to inspect every candle for surface flaws before packing.
Achieved whena customer audit, trade press visit or independent test confirms the robots and the camera checks in routine production.
Claimed by
Ounis Candle (Ounis (Shijiazhuang) Import & Export Co., Hebei), August 2025
Neural networks check candle jars
Researchers trained image-recognition models mostly on pictures of good candle jars, because labelled faulty ones are hard to collect, and report 100% accuracy on their own test set.
Achieved whenthe method runs on a production line, or another group repeats it on a different set of jar images.
Claimed by
Azeddine Mjahad and Alfredo Rosado-Muñoz, University of Valencia, in the journal Mathematics, August 2025
Small-shop taper dipping machine
Engineering students at Camosun College in British Columbia built an automated dipping machine for Little Mill, a one-person taper candle business, from igus bearings, linear guides and lead screws; igus says it lets the owner "dip tapers consistently" and helped her move into wholesale.
Achieved whenthe maker or an outside party reports steady output and reject rates from the machine.
Claimed by
igus Canada, the parts supplier, December 2025
Open
01
Grade container-candle wax surfaces
A camera rig and open model that photographs the top and sides of container candles after cooling and labels sinkholes (hollows that form as wax shrinks), frosting (white crystal bloom on plant waxes), wet spots (patches where wax has pulled away from the glass) and cracks, with a published, labelled photo set.
The briefCredit · unclaimed02
Pour conditions versus defects dataset
A public dataset that logs pour temperature, jar temperature, room temperature and cooling rate for each candle, alongside graded defects at 24 hours and 30 days, for common waxes.
The briefCredit · unclaimed03
Learned wick setting, small batches
A low-cost robot arm that learns from a handful of human demonstrations to place and centre one to three wicks in mixed jar types before pouring, so a small shop does not need a fixed machine for each jar.
The briefCredit · unclaimed
Open 01 · the brief
Grade container-candle wax surfaces
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A camera rig and open model that photographs the top and sides of container candles after cooling and labels sinkholes (hollows that form as wax shrinks), frosting (white crystal bloom on plant waxes), wet spots (patches where wax has pulled away from the glass) and cracks, with a published, labelled photo set.
Why it is still open
The only public candle anomaly image set, Amazon's VisA, shows groups of candles with flaws such as scratches, colour spots and missing parts, not the surface defects of container candles. The 2025 Valencia study inspects jars, not wax. A commercial inline system checks only wax drips on the glass and wick positions. Searches of Zenodo, Hugging Face and GitHub found no candle surface defect dataset.
Built with what exists
A Raspberry Pi High Quality Camera or a phone on a fixed stand with a diffuse light box; Label Studio or Roboflow for labelling; the open anomalib library, with models such as PatchCore, for spotting defects, plus a small classifier for defect type; Hugging Face for hosting the photos.
First step
Pour 40 test candles with deliberately varied cooling (a fan, a still shelf and a chilled shelf) to produce each defect, and photograph them under one fixed light setup.
Done looks like
At least 1,000 labelled photos from at least three waxes (paraffin, soy and a coconut blend), published under an open licence; precision and recall for each defect type on a held-out set poured by a different maker; a public scoring script.
Open 02 · the brief
Pour conditions versus defects dataset
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public dataset that logs pour temperature, jar temperature, room temperature and cooling rate for each candle, alongside graded defects at 24 hours and 30 days, for common waxes.
Why it is still open
A Europe PMC search for "soy wax" returns 60 records and none on candle frosting, wet spots or sinkholes. Supplier advice comes from in-house trials without raw data, such as CandleScience's note that pours at 150 °F (65.6 °C) and above "had severe adhesion problems" and that fragrance oil and dye "can greatly affect the amount of frosting".
Built with what exists
K-type thermocouples with MAX31855 amplifier boards on an Arduino or Raspberry Pi logger; an infrared thermometer; a FLIR ONE thermal camera for cooling maps; a temperature-controlled cabinet or a logged room; the grading rig above; Zenodo for the data.
First step
Fix one wax and one jar, pour 20 candles across five pour temperatures, and log them.
Done looks like
At least 200 candles across three waxes and two jar sizes, each with a full temperature log and photos at 24 hours and 30 days, released openly with a short analysis of which conditions predict each defect.
Open 03 · the brief
Learned wick setting, small batches
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A low-cost robot arm that learns from a handful of human demonstrations to place and centre one to three wicks in mixed jar types before pouring, so a small shop does not need a fixed machine for each jar.
Why it is still open
A 2021 University of Massachusetts Amherst student project set out to automate preparing multi-wick jars, which were done by hand, and built a cobot prototype for one jar format. Commercial machines such as the ZONESUN ZS-ACW2, listed at US$34,000, insert wicks at high volume for set formats. No published test of an arm trained by demonstration doing this job was found.
Built with what exists
Hugging Face's LeRobot software with an SO-101 arm and an imitation-learning policy such as ACT, or NVIDIA GR00T 2, an open robot model that learns a task from a few demonstrations (see the Note "Five demonstrations per task"); a top-down camera and OpenCV to measure how far each wick is off centre; standard wick stickers and centring bars. The work is cold: the robot handles no hot wax.
First step
Record 10 remote-controlled demonstrations of setting a single wick in one jar type and train a first policy.
Done looks like
20 filmed attempts on each of three jar types, with wick offset measured from photos; at least 18 of 20 within 2 mm of target; the tally, videos and training data published.