Industries / Mining and materials / Plastics and rubber
Mold making
Mold making is the design and building of the steel or aluminum molds that molding machines fill, including their channels for melt and cooling water.
6
Milestones achieved, from 1942 to 2014.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1942
Standard mold bases
Mold bases, the steel frame around the cavity, came ready-made in standard sizes with interchangeable parts, so mold makers machined only the cavity. The Plastics Hall of Fame says this greatly cut the lead time for tooling during and after World War II.
Credit
Ivar T. Quarnstrom, Detroit Mold Engineering Company (now DME)
1943
Cutting hard steel with sparks
Electrical discharge machining (EDM) wears metal away with sparks in an insulating liquid, so how hard the steel is no longer limits the cut. Mold makers use it for deep ribs and sharp inside corners, and a 2014 Stratasys paper notes that hard molds are "usually made from tool steel with a CNC milling machine or via electrical discharge machining".
Credit
B. R. and N. I. Lazarenko, a married couple in the USSR, who began by studying how sparks wore down tungsten electrical contacts
1965
Hot runners made workable
A runner is the channel that carries melt from the machine to the cavity, and in ordinary molds it freezes and is cut off as scrap. A hot runner keeps that channel heated and the plastic molten, so there is no runner to regrind and no wait for it to cool.
LessonEarly hot runners burned or streaked many plastics, and even heat along the melt path was the problem to solve.
Credit
Jobst Gellert, whom the Plastics Hall of Fame credits with the first commercially viable hot runner patent in 1965; he and Waltraud Gellert founded Mold-Masters in Toronto in 1963
1995
Cooling that follows the part
Metal 3D printing can build cooling channels that follow the part's shape at an even distance, called conformal channels, which straight drilling cannot do. The work began in Emanuel Sachs's group at MIT with a 1995 master's thesis. In December 2025 Plastics Engineering reported that a Michigan mold builder's printed steel inserts, with other changes, cut one product's cycle from 52 s to 41 s, about 20%. Those figures come from the customer and the printer maker.
Credit
Emanuel Sachs's group at MIT (Edward Wylonis, Xiaorong Xu)
2013
Robot polishing falls short
An EU project from 2010 to 2013 tried laser polishing plus robot finishing on freeform molds and saved at least 2.5 minutes per square centimeter. Its final report found "no clear advantage in the quality of the surface", and said the robot cell was "rather not economic" for any of the partner firms.
LessonA polishing process that works in trials still has to beat a skilled hand polisher on cost.
Credit
the POLIMATIC consortium, led by Fraunhofer
2014
Printed plastic molds
Mold inserts 3D printed in Digital ABS let designers mold real test parts within days. Stratasys says they suit "runs ranging up to 100 parts", and the contract molder Nypro called its trials "very successful". A 2020 UMass Lowell study found printed resin inserts lasted at most 78 cycles before cracking.
LessonPrinted plastic inserts fail suddenly by cracking, so they suit prototypes, not production.
Credit
Stratasys, with Nypro Healthcare
Watch for
Molds designed by software
Atomic Industries says it uses AI to design and build a mold for each part. It says its software "generates high-performance solutions in a matter of days, rather than the months typically required", and that it ships production parts to large manufacturers.
Achieved whencustomers or an independent auditor publish lead times and first-trial results for a set of Atomic molds, next to conventional quotes for the same parts.
Claimed by
Atomic Industries, September 2025
A free moldability checker
The FreeCAD DFM Workbench is an open-source add-on for the free CAD program FreeCAD. It checks 29 injection molding rules, including draft angle (the taper that lets a part slide out of the mold), undercuts and wall thickness. The FreeCAD blog called it a work in progress.
Achieved whenan independent group runs it on a set of parts with known molding problems and publishes how many it catches and how many false alarms it raises.
Claimed by
Ryan Kembrey, reported by the FreeCAD blog, April 2026
Open
01
Robot versus hand polisher
A cobot cell that polishes a curved steel cavity insert from its machined state to a named SPI finish grade (an industry scale of mold surface finish). Then a fair test against a skilled hand polisher on identical inserts, measuring time and roughness.
The briefCredit · unclaimed02
Public test for mold cooling
A public benchmark of about 20 part shapes. Each has a fixed mold block, material data, molding conditions, a standard heat simulation script and scoring rules: cooling time, temperature spread across the part, pressure drop in the channels, and the thinnest steel between channel and cavity. Any cooling design can then be scored, whether a person, a rule or an AI drew it.
The briefCredit · unclaimed
Open 01 · the brief
Robot versus hand polisher
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A cobot cell that polishes a curved steel cavity insert from its machined state to a named SPI finish grade (an industry scale of mold surface finish). Then a fair test against a skilled hand polisher on identical inserts, measuring time and roughness.
Why it is still open
The POLIMATIC project found "the finishing in the tooling industry is predominantly done by manual polishing". Robot demonstrations exist. A 2008 study reached a mirror-like finish on an aluminum bottle mold. A 2025 study reached an average roughness (Ra) of 0.017 µm, but on a flat steel block, and noted its settings must be re-tuned for each new abrasive. No published head-to-head with a hand polisher on a curved steel cavity was found.
Built with what exists
A Universal Robots UR5e cobot, whose built-in force sensor reads ±50 N; a FerRobotics Active Contact Flange for steady polishing force; a stylus roughness tester; the SPI B-1 grade (600-grit paper) as the target. Optionally, record a polisher's motions and train a policy with the open LeRobot library (Note: "No download").
First step
Get two identical curved P20 steel test inserts, measure roughness at marked points, and time a hand polisher finishing one of them to B-1.
Done looks like
Five inserts polished by robot and five by hand. A published table gives time per insert, roughness at 10 marked points, and photos. The robot reaches the B-1 range on at least 4 of 5 inserts.
Open 02 · the brief
Public test for mold cooling
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public benchmark of about 20 part shapes. Each has a fixed mold block, material data, molding conditions, a standard heat simulation script and scoring rules: cooling time, temperature spread across the part, pressure drop in the channels, and the thinnest steel between channel and cavity. Any cooling design can then be scored, whether a person, a rule or an AI drew it.
Why it is still open
A 2025 review in Computer-Aided Design says "there is still a lack of systematic and comprehensive classification, comparison and evaluation methodologies" for conformal cooling. It puts cooling at 50% to 80% of total cycle time. Vendor claims are tested on their own parts.
Built with what exists
FreeCAD for parts and channels; Elmer, an open-source solver for heat transfer and fluid flow; Python for scoring. Part shapes can be drawn new or linked from the ABC dataset of one million CAD models; those models stay under their creators' copyright, so link to them rather than copy them.
First step
Model a cup-shaped part with straight-drilled channels in FreeCAD, run one heat case in Elmer, and publish the files.
Done looks like
20 published cases with geometry files, boundary conditions and a scoring script that runs on a laptop. At least 3 design approaches are scored, including a straight-drilled baseline. A third party rerunning the script gets results within 2%.