Industries / Mining and materials / Steel and metals / Foundries and casting
Pouring and solidification
Pouring and solidification is the work of getting molten metal from the furnace into the mold, and of designing the channels and reservoirs so that the metal fills the mold cleanly and freezes without holes.
6
Milestones achieved, from 1917 to 2005.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1917
Pipe spun in a mold
Iron pipe had been cast standing upright in sand molds sunk in pits. Dimitri Sensaud de Lavaud, a Brazilian citizen living in Buffalo, New York, patented centrifugal pipe-casting machines with rotating molds cooled by circulating water, in which the spin throws the iron against the mold wall. His US patent 1,240,546 was filed on 15 December 1916 and granted on 18 September 1917. Wikipedia dates the invention to 1918 and says much pipe making then moved to centrifugal casting.
Credit
Dimitri Sensaud de Lavaud
1940
Freezing time from shape
Nicolas Chvorinov reported in the German foundry journal Giesserei that the time a simple casting takes to freeze is proportional to the square of its volume divided by its surface area. Engineers use the rule to check that a riser, also called a feeder (a reservoir of extra metal that tops up the casting as it shrinks), stays liquid longer than the casting it feeds.
Credit
Nicolas Chvorinov, who verified the rule experimentally in 1940, by the account of John Ågren in MRS Bulletin (1986)
1980s
Filling from below
Metal that falls or splashes into a mold folds its own oxide skin into the casting, where the folded film can later open as a crack. John Campbell of the University of Birmingham filled molds from the bottom while holding the inflow speed down, and Cosworth, the racing engine maker, adopted the method after its castings tested markedly better than anything before, by Kevin Cameron's account in Cycle World. Cosworth had been fighting sponginess around the exhaust valves of its DFX engine.
Lessonmany defects are made in the seconds of filling, not in the furnace.
Credit
John Campbell, University of Birmingham; Cosworth
1988
Simulation software for foundries
MAGMA was founded in Aachen, Germany, on 27 June 1988 by Peter R. Sahm, Preben N. Hansen and Erwin Flender to sell software that simulates how a casting fills and freezes. MAGMA's own anniversary article says its MAGMASOFT let foundry engineers design and optimize gating (the channels that carry metal into the mold) on a computer for the first time; that is the company's claim. By 1995 an outside benchmark of casting simulations drew materials data from MAGMA among others.
Credit
Peter R. Sahm, Preben N. Hansen, Erwin Flender
1995
Simulations put to test
For the 1995 Conference on Modeling of Casting, Welding and Advanced Solidification Processes, a University of Birmingham team cast a simple pure-aluminum plate, filmed its filling with real-time X-rays and asked modelers to predict it. Most programs got the 2-second fill and the main flow pattern right, but predicted freezing times scattered widely, and none could predict the bubbles that formed in the runner, which the authors think cause a large share of casting defects.
Lessondisagreement over basic materials data, including a viscosity quoted in the wrong units, hampered the comparison as much as the models did.
Credit
B. Sirrell, M. Holliday and J. Campbell, JOM (1996)
2005
Cameras steer the pour
The first automatic pouring systems that reacted during the pour used a laser to measure the metal level in the sprue cup, the funnel at the top of a mold; Inductotherm then used an industrial camera that watches the whole cup. Foundry Management & Technology reported in 2005 that Inductotherm's Visipour systems were pouring gray, ductile and malleable iron worldwide, mold after mold without an operator, and that Lodge Manufacturing poured 420 to 430 molds an hour with one, with a peak of 450.
Credit
Inductotherm
Watch for
Pouring that learns from pours
Viking Technologies (USA) and pour-tech AB (Sweden) say their EASYpour software module uses machine learning on records of earlier pours to set up the next one. In one unnamed job they report about 1 second cut from each pour and output raised from 313 to 343 molds an hour, nearly 10%. The article is written by Viking's president and names no foundry or outside check.
Achieved whena named foundry, or a party outside the two companies, publishes before-and-after mold rates and short-pour or scrap counts over months of production.
Claimed by
Viking Technologies and pour-tech AB, February 2026
Open
01
Open water-filling benchmark
A public set of videos of water filling clear acrylic copies of published gating systems, with the geometry files, the pouring head and flow rate for each, and ready-to-run cases for an open simulation code, so anyone can test a filling simulation against the same filmed flow.
The briefCredit · unclaimed02
Agent-designed gating, simulated
An AI agent that takes a 3D model of a part, proposes the gating channels and feeders, runs a filling and freezing simulation, reads the predicted defects, revises the design, and hands over the design with its simulation results for a foundry engineer to check.
The briefCredit · unclaimed
Open 01 · the brief
Open water-filling benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of videos of water filling clear acrylic copies of published gating systems, with the geometry files, the pouring head and flow rate for each, and ready-to-run cases for an open simulation code, so anyone can test a filling simulation against the same filmed flow.
Why it is still open
The 1995 Birmingham benchmark filmed aluminum filling with X-rays and found predictions "widely different" from measurement, but its paper shows the videos only as printed frames, which the authors say lose much of the detail. V. Jaiganesh and K. Prakasan (2016) found transparent water models "a reliable indicator of the mould-filling process" and saw substantial differences from commercial filling software, but the article page links no public data. A search in November 2026 found water-model studies of gating but no public set of filling videos with matching simulation cases.
Built with what exists
Laser-cut clear acrylic; a constant-head water tank; a phone or camera filming at 240 frames a second or faster; OpenFOAM's interFoam free-surface solver; GitHub and Zenodo for files and videos.
First step
Build an acrylic copy of the 1995 benchmark plate and its bottom-gated running system from the dimensions in the JOM paper, and film three water pours.
Done looks like
At least 10 gating geometries, each poured three times on video and published with geometry files and simulation cases under an open license, plus a table comparing simulated and filmed fill times and flow-front positions at fixed times for at least one open solver.
Open 02 · the brief
Agent-designed gating, simulated
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An AI agent that takes a 3D model of a part, proposes the gating channels and feeders, runs a filling and freezing simulation, reads the predicted defects, revises the design, and hands over the design with its simulation results for a foundry engineer to check.
Why it is still open
Vishvakarma Foundry AI, published in May 2026, computes first-pass gating and riser sizes from formulas and uses a language model only to explain them; its authors say it does not replace simulation and that its designs have not yet been checked by simulation or trial pours. PhysicsX reports AI models that tried gating alternatives for an unnamed aerospace manufacturer, with no figures. A search in November 2026 found no agent whose gating designs were compared, part by part, with engineers' designs in the same simulation.
Built with what exists
A model that can carry out multi-step design work, such as Claude Fable 5.2 (see the Note "Three working devices"); FreeCAD for geometry; OpenFOAM, or an academic license of MAGMASOFT or ProCAST, for simulation; Python to run the design loop.
First step
Pick five parts with published gating designs, model them in FreeCAD, and get one simulation to run from a script from start to finish.
Done looks like
20 parts that have engineer-made gating designs, each also designed by the agent, with both designs simulated in the same software and predicted shrinkage porosity, yield (casting weight over poured weight) and highest metal speed at the ingates (the openings into the casting) reported side by side, and two foundry engineers rating the designs without knowing which is which.