BRENNEN

Industries / Mining and materials / Steel and metals / Foundries and casting

Patterns and molds

Patterns and molds is the work of making the shape the metal fills: a pattern, a copy of the part, is pressed into packed sand or another material to leave a hollow, and cores, separate shaped pieces of bonded sand, are set inside to form holes and passages.

6

Milestones achieved, from 1707 to 1998.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1707

    Pots cast in green sand

    While running a furnace at Cheese Lane, Bristol, Abraham Darby invented a way of casting iron pots in "green sand", meaning damp sand packed around reusable patterns, and patented it in 1707. It made molds much quicker than the older method, in which a grass rope was wound around a core, a loam mold was built over it and the rope pulled out, and it let him cast thinner and cheaper pots.

    hmsjournal.org

    Credit

    Abraham Darby, by the account of the historian Peter King in Historical Metallurgy (volume for 2013, published 2015)

  2. 1944

    Resin-bonded shell molds

    Johannes Croning invented the shell, or Croning, process in Hamburg, Germany, in 1944. HA International, a maker of foundry binders, calls it the oldest core and mold making technology that uses synthetic resins to bind the sand.

    ha-international.com

    Credit

    Johannes Croning

  3. 1958

    Foam patterns left in sand

    Harold F. Shroyer's US patent 2,830,343, filed in April 1956 and granted on 15 April 1958, packs mold sand around a pattern of expanded polystyrene foam that stays in the mold. Molten metal poured in burns the foam away and takes its place, so the mold has no open cavity and the pattern never has to be drawn out.

    patents.google.com

    Credit

    Harold F. Shroyer

  4. 1964

    Flaskless molds on a line

    DISA, a Danish company, says it brought vertical molding to market with the DISAMATIC in 1964. The machine makes sand molds split along a vertical line, with no box (flask) around them, one after another on a conveyor; DISA now quotes up to 555 molds an hour without cores and 485 with cores. The design came from Vagn Aage Jeppesen, which DISA bought and developed.

    disagroup.comdisagroup.com

    Credit

    Vagn Aage Jeppesen; DISA

  5. 1968

    Cores hardened by gas

    Ashland introduced Isocure, a phenolic urethane "cold-box" binder, at the American Foundry Society's convention in 1968. Sand mixed with a phenolic resin and a polyisocyanate is shaped, and an amine carried in a stream of air or nitrogen is passed through it, so the core hardens at room temperature with no heated tooling; the examples in the patent gas each sample for 60 seconds.

    ask-chemicals.compatents.google.com

    Credit

    Janis Robins, inventor on US patent 3,409,579 (filed 1967, granted 5 November 1968); Ashland Oil

  6. 1998

    Molds printed from sand

    Researchers at the Technical University of Munich printed their first sand molds in 1998, founded Generis GmbH in 1999, and shipped the first sand printers to BMW and Daimler in 2002, by the account of ExOne, one of Generis's two successor companies. A print head lays down the mold layer by layer straight from a computer model, with no pattern. In April 2024 BMW said its Landshut plant prints up to 4,500 sand cores a day on 17 printers, bound with inorganic binders, for six cylinder head variants.

    exone.comnewsroom.bmwgroup.com

    Credit

    Technical University of Munich researchers and Generis GmbH, later voxeljet and ExOne

Watch for

  1. Printed molds for wind castings

    ExOne, which now owns voxeljet, says the Advanced Casting Cell it built with GE Renewable Energy and the Fraunhofer Institute for Casting, Composite and Processing Technology (IGCV) can print sand molds for castings up to 9.5 m across and over 60 tonnes, for parts such as the nacelle of GE's Haliade-X offshore wind turbine. It says printing cuts pattern and mold lead time from ten weeks or more to about two, and its history page says the printer, the VX9000, completed its first trial castings for the energy sector in 2025.

    Achieved whenGE, a partner foundry such as Hodge Foundry, or the US Department of Energy, which funded part of the program, reports a production wind or hydro casting poured in printed molds, with its lead time and inspection results.

    voxeljet.comexone.com

    Claimed by

    ExOne and voxeljet, September 2025

Open

  1. 01

    Parting lines chosen by software

    An open benchmark of at least 100 sand-cast part models, each with the parting line (where the two mold halves meet), the cores and the draft angles that a foundry engineer actually chose, plus a baseline tool that proposes them from the bare part model.

    The brief
    Credit · unclaimed
  2. 02

    Open green sand records

    A public data set of a year or more of daily green sand tests from working foundries (moisture, compactability, green compression strength, permeability, active clay), each matched to the castings scrapped that day for sand-related defects, with a baseline model that predicts sand-related scrap from the tests.

    The brief
    Credit · unclaimed

Open 01 · the brief

Parting lines chosen by software

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open benchmark of at least 100 sand-cast part models, each with the parting line (where the two mold halves meet), the cores and the draft angles that a foundry engineer actually chose, plus a baseline tool that proposes them from the bare part model.

Why it is still open

A 2025 paper from the Bandung Institute of Technology, Indonesia, says choosing the parting line and making patterns and core boxes can take up to 70% of the lead time for a new cast part. Its SolidWorks add-in scores candidate parting lines on five criteria, but it is shown on three parts and not compared with engineers' choices. A GitHub project created in May 2024 to automate the choice of mold parting lines holds only a license and a one-line description. A search of Zenodo in November 2026 found no public set of parts with foundry-chosen parting lines.

doi.orggithub.com

Built with what exists

Open CAD libraries that read STEP files (FreeCAD, CadQuery, Open CASCADE); the five criteria from the Bandung paper (projected area, undercuts, draw distance, dimensional stability, cope and drag volume) as a first rule set; an AI model such as Claude Fable 5.2 to propose and explain choices (see the Note "Three working devices").

  1. First step

    Ask one jobbing foundry for 20 past parts whose owners agree to release them, and record where the parting line and cores went on each.

  2. Done looks like

    At least 100 parts from three or more foundries, published under an open license with the owners' consent, and the tool's agreement with the engineers' parting direction and core count reported part by part.

Open 02 · the brief

Open green sand records

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public data set of a year or more of daily green sand tests from working foundries (moisture, compactability, green compression strength, permeability, active clay), each matched to the castings scrapped that day for sand-related defects, with a baseline model that predicts sand-related scrap from the tests.

Why it is still open

Studies that link sand tests to scrap use private data. Ziga Breznikar, Marko Bojinovic and Miran Brezocnik (University of Maribor and Pro Labor, 2024) joined 7,115 batches at one Slovenian foundry, named only as Company X, and reached an accuracy of 0.85 with a nearest-neighbors classifier, but published no data. The only open set found, posted to Zenodo in September 2025 by Manik Sonawane, is synthetic: 1,500 simulated mixer batches.

ijsimm.comzenodo.org

Built with what exists

The foundry's own sand laboratory and scrap logs; the Maribor paper's method of joining sand readings to casting batches by timestamp; Python with pandas and scikit-learn; Zenodo for publishing.

  1. First step

    Ask one green sand foundry for a year of its sand test log and its scrap log, and join them by date.

  2. Done looks like

    At least 365 days of sand tests and scrap counts from three or more foundries, released under an open license, with the baseline's error reported on a foundry held out of training.