A foundry twin is a living model of your process, fed by your own instruments. It holds the sand test, the pour and the rejection in one record, so the cause is a lookup rather than a meeting.
The honest version
Strip the marketing away and it is a simple idea. A twin is a model of your plant that stays honest because your own instruments feed it. Every heat becomes a record with the sand test, the chemistry, the pour and the outcome held together, so a rejection can be walked backwards to the stage where something first went out of range.
The knowledge that prevents scrap already exists in your foundry. It sits with the two or three people who have been there longest. When they are not on shift, it is not available, and when they retire it leaves. A twin is how that reasoning gets written down while they are still there to correct it.
The bench instruments you already own are what feed all three stages. The twin does not replace your lab. It makes your lab the thing everything else is checked against, which is a stronger position for it than the corner of the plant it usually occupies.
All three stages are running on a fully simulated foundry you can open now, with no account and nothing to install: see the live plant.
Versatile has built foundry sand testing instruments since 1967, and they run in more than 9,000 foundries across 40 countries. The twin was built the way it was because we have spent decades watching what the bench actually measures and what gets lost between the bench and the rejection note.
Stage two, on the floor
Not another dashboard to keep looking at. A job in front of the person who can do something about it, before the rejections and the process failures happen. The twin raises it as a maintenance request in the plant's own ERP, with the reasoning, the cost and the history attached. This is one from last week, as the fitter would open it, with the plant and the people blacked out.
Why this request exists. Mould hardness on line 2 has drifted from 95 to 91 (B-scale) since Thursday, falling each shift. Squeeze pressure now takes 2.1 s to reach set point where it took 1.4 s, inside the same cycle time. Read together, the hydraulic oil on the squeeze circuit is past its useful life.
If nothing changes this week: sand inclusion rejections are expected 2 to 3 points higher on Thursday's ductile run.
Last time this happened: MR/2026/0087, 11 to 13 March, same signature on the same machine. Sand inclusion rejections rose from 2.1% to 4.9% across three shifts before the oil was changed. 146 castings scrapped, 38 of them on the ductile run. Impact Rs 2,38,943 in metal, sand and machining time, before the customer credit note.
Advised: change the squeeze-circuit oil (HLP 68, 46 L) and the cylinder rod seal (90 x 110) before Thursday. Line 2 down window agreed with production, Wed 14:00 to 17:00.
Three things on that request are worth noticing. The fitter never had to find the problem. The works manager never had to be persuaded of it, because the request arrived with a price on it: the last time this machine showed the same signature, it cost Rs 2,38,943 before anyone changed the oil. And nobody had to remember March. The twin did, and it put the number in front of the one person who could sign a purchase order, before the shift had started.
The reasoning, the cost and the history are on it. A stores requisition. A purchase order. A fitter's name. The fitter has his work cut out, and it is in the ERP, not in somebody's head.
No miracles on that card. This is what the data crunching should do for you: a record that leads to an action, and an action that only a human can take.
How it is put together
A twin is only as honest as the readings underneath it. These are the five stages the loop needs, and what covers each one today.
Compactability, green compression strength, permeability, moisture, active clay and grain fineness. Tested at the line by V-CAT, or on the bench by the instruments you already run.
A number is only evidence if the instrument that produced it was in calibration. Every instrument carries a code, and the certificate can be checked by anyone, including your auditor.
Results reach the record without anyone retyping them, with the time they were actually taken. Roughly 193 machines across 82 foundries report this way today.
The fettler photographs the defect and taps what it is. No typing, no email, works without a network, in six Indian languages. Free for every foundry.
With the four above in place, each process parameter can be ranked against each defect on your own heats. Not an opinion in a meeting, an ordered list with the strength of each link.
Most plants are sold stage five first, and it disappoints, because a correlation drawn over readings nobody trusts tells you very little. Stages one to four are the slow part and they are where we can help most. Start there and the fifth stage becomes worth switching on.
Reference library
Free, no sign-up, no gate. Seven cover calibration and what an auditor will ask for. Ten cover the defects themselves, and what the sand was usually doing when they appeared. Three deal with reading a rejection record once you have one.
Before you ask anyone for a quote
These get asked on every call. The answers are here rather than behind a form.
There is no single price, and any supplier who gives you one before seeing your plant is guessing. The cost breaks into three parts, and most foundries already carry the first.
Instruments. If your bench is sound and in calibration, this may be nothing. If the strength machine is thirty years old and reads high, that gets replaced before anything else is worth doing.
Connecting them. Charged per machine, once. This is the part that stops results being retyped.
The platform. Charged by use, monthly, and it can be stopped. Defect logging on CaRe 101 is free for every foundry with no user limit, so a plant can begin recording outcomes today at no cost while it decides about the rest.
Less than most people expect. Nothing is rewired and no controller is reprogrammed. The twin reads a copy of signals your machines already produce, and where a machine produces none, the reading can be typed in or photographed from the log sheet.
What it genuinely needs is three things: instruments in calibration, somebody who records the rejection rather than only sweeping it up, and one person who owns the question. The third is the one plants underestimate.
Traceability from the first week. As soon as readings and rejections land in one place, any rejected casting can be walked back to its heat, its mould and the sand test that preceded it. Most plants find their first surprise in that walk within days, usually a parameter nobody was watching that had been drifting for months.
Ranked causes within the first quarter. Once a few hundred heats are in, each parameter is ranked against each defect on your own data, and the ranking tightens with every shift. Prediction, the heat flagged before the pour, switches on as the model earns it, in most plants within a few months. Each stage pays for itself before the next one starts.
From our own instruments
We went back through every reading our connected V-CAT II units produced over the five years to early 2025. Around 5.6 million individual readings across roughly 1.38 million test cycles. Compactability, moisture, green compression strength and permeability were taken on the same sample in the same cycle, so these are honest pairs rather than readings matched up afterwards.
Pool every plant together and the firmest relationship left standing is green strength against permeability, r = -0.27 on about 136,000 paired tests. Denser, wetter sand passes less air and takes more load, which any foundry man would tell you. Everything else averages out to nearly nothing.
That is the point. A rule of thumb that holds across the industry does not exist in 1.38 million real tests, which is why generic plant software so often disappoints. The answer for your plant is in your plant, and nowhere else.
Four units carry enough paired readings to stand on their own, each on its own moulding line. Moisture against compactability, measured by the same instrument model running the same test on the same kind of green sand, gives a different answer on every one of them. On one line it runs the other way entirely.
| Moulding line | Correlation r | Paired tests | Reading |
|---|---|---|---|
| Line A | +0.60 | tens of thousands | Water drives compaction hard |
| Line B | +0.28 | hundreds of thousands | Present but muted |
| Line C | +0.07 | hundreds of thousands | Almost no relationship |
| Line D | -0.42 | tens of thousands | It runs the other way |
On line A, adding water raises compactability sharply, which is what the textbook says should happen. On line C the same change barely moves it. On line D it pushes compactability down.
If your control rule came from a handbook, a seminar or another plant, this table is the reason it may not be working for you. The physics does not change between foundries. The muller, the return loop, the cooler, the ambient conditions and the way water is dosed all do, and on this evidence they change the relationship enough to reverse its direction.
Everything above links sand properties to each other. Add what was rejected, and the same arithmetic ranks sand behaviour against your actual defects: which reading moved before the blowholes, which one before the misruns. That record is what CaRe 101 collects on the floor, by photograph, and it is free for every foundry with no user limit, so a plant can start building it today before spending anything else.
Where to start
You do not need a platform to begin. Pick the defect that costs you most, check that the instrument which should have caught it is in calibration, and start writing down the rejection. We can help with all three, and the first two are ordinary work we have done for decades.