Industries / Mining and materials / Steel and metals / Tool and die
Fitting and tryout
Fitting and tryout is the work of putting a new die or mold together, running it in a press, and reworking it by hand until its parts match the drawing.
4
Milestones achieved, from 1943 to 2003.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1943
Presses made for spotting
Spotting is fitting the two halves of a die so that they meet evenly: the toolmaker closes them, sees where they touch and grinds down the high spots. A die spotting press is built for this work alone. Herbert S. Fullerton of the Baldwin Locomotive Works patented one in which the upper platen first drops under a controlled flow of oil until the dies touch, then is lowered by motor-driven screws so that the corner nearest a high spot stops first, with switches that sense a few thousandths of an inch of movement. His patent says the accuracy of earlier spotting presses was limited by how accurately each machine had been built.
Credit
Herbert S. Fullerton, Baldwin Locomotive Works
1999
Judging dies by the assembly
Car makers had approved each stamping only when it met its own tolerances, so toolmakers reworked dies for panels that would have assembled well anyway. In "functional build", the stampings are welded into a subassembly first, and a die is reworked only if the assembly misses its target. Jay Baron reported that in 13 door assembly processes the door inner panel moved on average 40% closer to its nominal dimensions once assembled. A 2003 model by David W. Glenn of the University of Illinois found that such rework is "often unnecessary because in-spec assemblies can be made from out-of-spec panels".
LessonA panel that is off its drawing can still make a good assembly, but Baron notes that many stampers avoid the method because it is harder to manage when several suppliers build the dies.
Credit
Patrick C. Hammett, Shannon M. Wahl and Jay Baron, University of Michigan
2002
Tryout timed in two countries
The Center for Automotive Research benchmarked four North American and two Japanese die makers that supply car companies. Tryout took 7.6 weeks on average in the North American shops against 2.8 weeks in the Japanese ones, and a full set of dies 34.4 weeks against 19.5. The Japanese shops kept 0.5 to 1.0 tryout presses for each CNC machine tool, against 1.0 to 4.2 in the North American shops. The report put the difference down to simpler parts, fewer late engineering changes, more tryout done in the customer's production press, and dies accepted by functional build.
LessonThe report found no reason to think North American toolmakers less capable; the time went into complex parts, late changes and how dies were accepted.
Credit
Manufacturing Systems Group, Center for Automotive Research (Altarum)
2003
Tryout moves onto the screen
Die shops began to run forming simulations before cutting steel, checking a part for splits and wrinkles and working out its trim line in software instead of hitting laser-cut test blanks in the press. Terry Swan of Tomco Tool & Die in Belding, Michigan, wrote in 2003 that this let his shop fully assemble 95% of its tools before their first hit. In 2019 Oldenburg Metal Tech of Saukville, Wisconsin, said that simulating a die's motions on screen had greatly reduced the mistakes it used to find in the tryout press.
LessonSwan warned that a simulation is only as good as the material data entered, and that it does not catch die parts hitting each other or changes in sheet thickness along a coil, so CAD clearance checks and adjustable inserts are still needed.
Credit
Tomco Tool & Die; Oldenburg Metal Tech
Watch for
Half the tryout loops
AutoForm, a maker of software that simulates sheet metal stamping, says its software "helps reduce physical trial loops by 50-60%". The report gives no method or sample.
Achieved whenan independent study counts physical tryout loops on comparable dies built with and without the software, and publishes the counts.
Claimed by
AutoForm, in its ESG report for 2025, April 2026
Relieving surfaces before tryout
In tryout, toolmakers grind a die until it presses evenly on the sheet. Researchers at Heilbronn University of Applied Sciences use the pressure map from a forming simulation to find which die surfaces actually need to touch the sheet, and recommend setting the rest back by 0.4 mm before machining. They say this can cut the die surface to be modeled and spotted by up to 60%, and that validation with data from real tools is under way.
Achieved whena tool shop publishes tryout hours or spotting loops for dies built this way against comparable dies built the usual way.
Claimed by
Florian Göltl, P. Zimmermann, A. Birkert and N. C. Stache, Heilbronn University of Applied Sciences, September 2025
AI drafts finishing steps
Kiko Mold, a shop in Gifu, Japan, that builds about 300 one-off plastic molds for cars a year, says it gave generative AI more than two years of its records of adjusting molds after completion, so that the AI reads a customer's correction sheet and proposes the finishing steps a skilled finisher would plan. It says entry time per mold fell from 10 to 5 minutes, and that gaps in hand-entered past records made the AI's answers hard to check.
Achieved whenthe shop or an outside party publishes how often the AI's plans match what finishers actually did, over a run of molds.
Claimed by
Kiko Mold, with Kuroda Seisakusho and Gifu University, in a report to Gifu Prefecture's DX consortium, February 2026
Open
01
Open springback correction loop
Free, open programs that take a 3D scan of a stamped part, compare it with the CAD model, and move the die surface the opposite way to cancel springback. Then a real check: dies for one simple part, such as an aluminum hat-shaped channel, are cut, corrected and run at a partner press shop or university lab.
The briefCredit · unclaimed02
Public log of tryout loops
A shared record of die tryout. For each loop it holds the defect, where it is on the part, what was changed on the die, the hours spent and whether the change worked. An AI model turns toolmakers' handwritten notes, photos and voice memos into these records, and the records are published without names.
The briefCredit · unclaimed
Open 01 · the brief
Open springback correction loop
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Free, open programs that take a 3D scan of a stamped part, compare it with the CAD model, and move the die surface the opposite way to cancel springback. Then a real check: dies for one simple part, such as an aluminum hat-shaped channel, are cut, corrected and run at a partner press shop or university lab.
Why it is still open
The only public springback compensation code found, from a semester thesis at the Technical University of Munich, works on results from the commercial AutoForm solver, was tested in simulation only, and says its results "are not a general guarantee for other geometries". A 2020 analysis of the NUMISHEET 2016 benchmark found that springback comparisons still depend on how the fixture and measurement points are defined.
Built with what exists
CalculiX or code_aster, free finite element programs that model contact and permanent bending, for the forming and springback steps; Python with Open3D to line up scans with the CAD model; FreeCAD to edit die surfaces; a handheld structured-light 3D scanner or a scanning service; a machine shop to cut the corrected die inserts.
First step
Model a simple U-channel bend in CalculiX, compare its springback angle with values in published studies, and write the scan-to-CAD comparison on a strip bent in a bench vise.
Done looks like
For one part and one aluminum grade, two correction loops bring at least 90% of scanned points within ±0.5 mm of the CAD model on five parts in a row, with code, scans and die files published so that a second lab can repeat the loop.
Open 02 · the brief
Public log of tryout loops
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A shared record of die tryout. For each loop it holds the defect, where it is on the part, what was changed on the die, the hours spent and whether the change worked. An AI model turns toolmakers' handwritten notes, photos and voice memos into these records, and the records are published without names.
Why it is still open
The public figures on tryout are averages from small surveys, such as the Center for Automotive Research's 2002 benchmark of six die makers, and estimates, such as a minimum of four rework iterations at $25,000 to $125,000 each for some panel types, published on AutoForm's Forming World site in 2022. Vendor claims of 50 to 60% fewer loops cannot be checked against such figures. No public dataset of individual tryout loops was found.
Built with what exists
a vision-language AI model such as Claude or Gemini to read notes and photos; Whisper, an open speech-to-text model, for voice memos; a form tool such as KoboToolbox; a fixed list of defect types (split, wrinkle, springback, surface low, trim line).
First step
Ask one die shop for the notes from one finished die's tryout, and draft the record format from them.
Done looks like
At least 200 loop records from at least 30 dies at three or more shops, with the AI-filled fields matching the toolmaker's own check on at least 90% of fields in a 50-record sample, published under an open license with the number of loops per die by part type.