Industries / Mining and materials / Steel and metals / Forging
Dies
Dies are the blocks of hardened steel whose cut-out shapes, called impressions, give a forging its form; this work covers cutting them, predicting how metal will fill them, and tracking how they wear out.
4
Milestones achieved, from 1921 to 2007.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1921
A tracer copies the die
The brothers Sidney A. and Joseph F. Keller went into business in 1896 designing and making dies for the silverware trade. Joseph Keller then built a weight-controlled machine for copying forging dies, the E-1 Automatic Die Cutter, and the firm sold its first machine to the Crescent Tool Company in 1915. John C. Shaw, who joined in 1916, designed an electric tracer control, in which a tracer follows a model of the impression and the cutter copies its shape into the steel block; the first Keller machine of that type was shipped to the Budd Company in Philadelphia in March 1921. Tracer control later gave way to computer numerical control (CNC).
Credit
Sidney A. Keller, Joseph F. Keller and John C. Shaw, Keller Mechanical Engineering Corporation
1988
Dies tried out in software
ALPID is a finite element program, one that splits the metal into many small cells and calculates how each deforms; it was developed at the Battelle Memorial Institute with US Air Force sponsorship to analyze metal flow in dies of any shape. In 1988 T. E. Howson and H. E. Delgado of the Wyman-Gordon Company in North Grafton, Massachusetts, used it on a superalloy turbine disk that had coarse grains and high "sonic noise", background echoes in ultrasonic testing, in some places. The simulation showed that the metal had become too hot there, partly from the heat of its own deformation, and the forge changed its process. They wrote that simulation was "now an integral part of forging design" and was replacing the costly cycle of trying out dies, testing and redesigning.
Lessonmetal heats itself as it is squeezed, so the furnace setting does not fix the temperature inside the forging.
Credit
Battelle Memorial Institute (ALPID, building on S. I. Oh's analysis of dies of arbitrary shape); T. E. Howson and H. E. Delgado, Wyman-Gordon
2005
Most large forges simulate
Former Battelle staff who had developed ALPID and its successor DEFORM founded Scientific Forming Technologies Corporation (SFTC) in 1991. A survey in Forging magazine's November/December 2005 issue, as quoted by John Walters in February 2006, found that 80% of large forging companies (250 or more employees), 75% of those with 100 to 249 and 50% of those with 50 to 99 used process simulation, shares Walters says had more than doubled in a decade.
Credit
Scientific Forming Technologies Corporation (DEFORM); survey by Forging magazine
2007
Die wear computed and checked
Die wear, the slow loss of shape as hot metal slides and presses over the impression, is a major reason hot forging dies fail, and the hot workpiece softens the die surface by heating it past the steel's annealing temperature. Bernd-Arno Behrens and André Hundertmark at IFUM, Leibniz Universität Hannover, built a finite element method that calculates wear including that softening. They calibrated it on several industrial forging dies whose wear was measured optically over long operating times, and checked it on an industrial process with a complex die. They wrote that the number of forgings a die is designed to make was still "generally based on the know-how of the engineer".
Credit
Bernd-Arno Behrens and André Hundertmark, IFUM, Leibniz Universität Hannover
Watch for
Laser corrects anvil wear
On a high-speed forging machine, which shapes a hot bar between two anvils, the control reads the bar's diameter from the anvils' position, so worn anvils give wrong readings. Kamil Sikora reports a camera system at a forging plant that builds a 3D model of the shaft as it is forged and feeds a correction into the machine's Siemens control, with software written at the University of West Bohemia in Pilsen and protected by Czech utility model UV 32 227. In a week of measurements in December 2024 the anvils lost 2.9 mm, from 117.2 mm to 114.3 mm, over 137 working hours, but the paper gives no figures on how much the forgings' final dimensions improved.
Achieved whenthe forge or an outside party publishes final-diameter errors and machining allowances for the same products before and after the correction, over months of production.
Claimed by
Kamil Sikora, Institute for Research and Development in the Field of Thermomechanics and Thermal Engineering, Ostrava, December 2025
Open
01
A library of worn dies
A public library of 3D scans of retired hot-forging dies, each compared with its design model and labeled with the number of forgings it made, the die steel and its hardness, the surface treatment, the lubricant and the part it forged; and a first model that predicts where, and how deep, a new die will wear.
The briefCredit · unclaimed02
Forging drawings from part models
A program that takes the 3D model of a machined steel part and draws the forging it should be made from, adding machining allowance (extra metal to be cut away later), draft (the slight taper that lets a forging come out of the die), corner radii and a parting line (where the two die halves meet), with a public test of its drawings against those of forging engineers. The forging drawing is what the die impression is cut from.
The briefCredit · unclaimed
Open 01 · the brief
A library of worn dies
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public library of 3D scans of retired hot-forging dies, each compared with its design model and labeled with the number of forgings it made, the die steel and its hardness, the surface treatment, the lubricant and the part it forged; and a first model that predicts where, and how deep, a new die will wear.
Why it is still open
Die life is hard to forecast: Muhammad Younas and Anwar Sheikh of King Fahd University wrote in 2002 that the life of forging tools sometimes varies by several orders of magnitude. Marek Hawryluk, Jacek Ziemba and Łukasz Dworzak of Wrocław University of Science and Technology developed a 3D scanning method that compares worn forging tools with their CAD models and tracks wear against the number of forgings made (2017), and Behrens and Hundertmark calibrated their wear model on industrial dies measured optically (2007). No scans from either study were found in a public repository, and searches of Zenodo, Hugging Face and GitHub in November 2026 found no public set of worn forging-die scans.
Built with what exists
A handheld structured-light scanner such as a Creaform HandySCAN or Shining 3D EinScan, with phone photogrammetry as a low-cost comparison; CloudCompare, open-source software that colors the difference between a scan and a CAD model; Open3D for batch processing; a gradient-boosted regression model as a first baseline. Dies are scanned cold, after they leave service.
First step
Ask one closed-die forge for five retired die inserts with their stroke counts and design files, and scan them against the CAD.
Done looks like
At least 100 retired dies from at least three forges, published under an open license with deviation maps and metadata; scanner error checked against a coordinate measuring machine (CMM) on ten of them; a model that predicts maximum wear depth on dies from a forge it was not trained on, with its error published.
Open 02 · the brief
Forging drawings from part models
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A program that takes the 3D model of a machined steel part and draws the forging it should be made from, adding machining allowance (extra metal to be cut away later), draft (the slight taper that lets a forging come out of the die), corner radii and a parting line (where the two die halves meet), with a public test of its drawings against those of forging engineers. The forging drawing is what the die impression is cut from.
Why it is still open
The rules are written in design handbooks and applied by engineers. Knowledge-based design programs have tried parts of the job since at least 1988, when Juipeng Tang, Soo-Ik Oh, Taylan Altan and Richard A. Miller published "A knowledge based approach to automate forging design". J. Kulon, D. J. Mynors and P. Broomhead (2006) described a tool that guides a user from a component profile to forgeable geometry and a hot forging die design. For die casting, J. Madan, P. V. M. Rao and T. K. Kundra (2006) chose parting lines from a part's STEP file and checked the results against experts, noting that the choice is normally made "intuitively" by an expert. Searches of Crossref and GitHub in November 2026 found no public tool that starts from a full 3D part model, and no public test of any such tool against forging engineers.
Built with what exists
An AI model that writes CAD code, such as Claude Fable 5.2, which Anthropic says designed three working circuit boards (see the Note "Three working devices"); CadQuery or FreeCAD for the geometry; a script that checks every face for minimum draft and radius; the public ABC dataset of CAD models as a source of machined parts to try.
First step
Pick five machined parts with public STEP files, write the allowance, draft and radius rules as a checklist, and ask a forging engineer to draw one of them by hand as the reference.
Done looks like
30 machined parts, each drawn as a forging by the program and by a forging engineer; three other forging engineers, not told which drawing is which, rate each as usable, usable with small changes, or unusable; the program reaches the first two ratings on at least 24 of 30, with forging weights within 5% of the engineer's.