Foundries can use AI to build custom foundry management software for pattern and tooling records, charge mix and heat data, spectro results and pouring parameters, sand and moulding records, knockout and fettling stages, rejection analysis by defect and by stage, metal yield, heat treatment, test certificates and heat-wise traceability. Pentoggle is an AI platform that generates production-ready software from a plain English description, which means a foundry can build an application around its own furnaces, patterns and customers instead of adapting to a system built for a machining plant.
Most foundries in India already run Tally for accounting, some run Busy, and larger businesses may run SAP Business One. Those systems handle purchase, sales, GST and accounting well. This is not a proposal to replace them. Pentoggle builds the plant application around them, covering the workflows they were never designed for.
A foundry earns differently from a machine shop. A machine shop can inspect a part, find it wrong and often correct it. A foundry cannot. Once metal is poured, the casting is whatever the sand, the metal and the pouring produced, and the only remedy is to melt it again. That is why a foundry's economics sit in two places: how much saleable casting comes out of every tonne melted, and how early a bad casting is identified.
Key takeaways
- Rejection is decided at pouring and discovered much later, which is why rejection data is only useful when it is linked back to the heat that produced it.
- A casting rejected after machining has consumed every operation performed on it, so where a defect is found matters as much as how many there were.
- Metal yield, the saleable casting weight per tonne melted, is the second largest cost lever after the charge itself.
- Customers increasingly ask for composition and mechanical test certificates tied to the heat, and reconstructing them later is slow and unconvincing.
- The number that runs a foundry is rejection split by defect and by the stage at which it was found.
Why foundries are badly served by existing software
Tally records the scrap and pig iron purchase and the casting sales. It does not know that heat 412 ran with a charge mix adjusted for the scrap that was available that morning, that the castings poured after eleven o'clock came from metal held longer than usual, or that most of last month's blowhole rejections came from two heats on the same day.
Packaged ERP models a bill of materials and a routing. A foundry's inputs are a charge mix that changes with what is in the yard and what the spectro says, and the output is a heat that becomes many castings across several patterns. The unit that everything actually refers to, the heat, does not exist as a concept in most systems, so it gets forced into a batch or a production order and loses the parameters that matter.
The rest lives in the furnace logbook, the spectro printout, the sand lab register, the fettling tally and the rejection note.
Most foundries are running some combination of the first two columns below.
What foundries use today, and what they can build instead
| Logbooks and Excel | Packaged ERP | Application built with Pentoggle | |
|---|---|---|---|
| Heat records | Furnace logbook | Production batch without parameters | Charge mix, additions, spectro results, temperature and timing |
| Charge mix | Decided per heat, recorded loosely | Fixed formulation | Actual mix per heat with cost, against the target |
| Sand and moulding | Lab register | Not modelled | Sand parameters recorded and linked to the heats poured that shift |
| Rejection | Counted at fettling | Scrap quantity | By defect, pattern, heat and the stage where found |
| Metal yield | Estimated monthly | Not computed | Poured weight against saleable weight, per heat and pattern |
| Test certificates | Typed when requested | Generated from limited data | Produced from the heat record, with results already attached |
| Traceability | Reconstructed from books | Rarely configured | Casting to heat to charge to supplier, in one query |
What a foundry can build
Each of these can be built separately or combined. Most foundries start with heat records and rejection analysis.
Pattern and tooling register
Every pattern with its owner, castings per mould, condition, location and repair history.
Heat record
Furnace, charge mix by input with quantities, alloy additions, spectro results, tapping temperature, pouring temperature and the time the metal was held.
Moulding and sand records
Sand parameters from the lab, moulding line output, and which patterns were poured from which heat.
Casting output by pattern
Moulds poured, castings knocked out and the weight produced, per pattern and per heat.
Metal yield
Metal melted and poured against saleable casting weight, with returns from runners and risers accounted for.
Fettling and finishing stages
Knockout, fettling, grinding, shot blasting and heat treatment, tracked as stages so work in progress is visible.
Rejection analysis
Defect type, quantity, pattern, heat and the stage at which the defect was found.
Test and inspection records
Chemical composition, mechanical properties, hardness and any NDT, held against the heat.
Test certificate generation
Certificates produced from the heat record rather than assembled by hand when a customer asks.
Traceability
Casting to heat to charge inputs to supplier, and forward to every dispatch.
Rejection is decided at pouring and found much later
A blowhole, a shrinkage cavity, a cold shut or a sand inclusion is created in the few minutes around pouring. It is discovered at fettling, at final inspection, or in the worst case after the customer has machined the casting and cut into a defect that was invisible from the outside.
This gap between cause and discovery is what makes foundry rejection so hard to act on. By the time the defects are counted, the heat that caused them was days ago, the charge mix that produced it was adjusted for whatever scrap was available, the sand parameters that morning were recorded in a different register, and nobody can reassemble the picture. So the response becomes generic: tighten inspection, remind the operators, watch the numbers.
Attaching the heat number to every casting, and carrying it through knockout, fettling and inspection, closes the gap. Rejections then aggregate by heat rather than by day, and the pattern becomes visible. Defects concentrated in two heats point at those heats' parameters. Defects spread evenly across every heat on one pattern point at the pattern or the methoding. Those are completely different problems, and no rejection percentage can tell them apart.
Where a defect is found decides what it cost
A casting rejected at knockout has consumed metal, sand and furnace power. The same casting rejected after fettling has additionally consumed grinding labour and shot blasting. Rejected after machining, whether in your plant or your customer's, it has consumed everything, plus the machining cost, plus in some cases a customer's line stoppage.
Foundries generally count rejection as a single percentage, which treats all three as equivalent. They are not equivalent, and the difference between them is often larger than the difference between a good month and a bad one.
Recording the stage at which each rejection was found produces two useful things. It gives a weighted rejection cost rather than a count, which is the number that belongs in costing. And it tells you whether your inspection is positioned correctly, because defects consistently escaping to machining mean the check that would have caught them is happening too late or not at all.
Certificates are easier to produce than to reconstruct
Customers in automotive, pumps, valves and general engineering increasingly ask for a test certificate with each consignment: chemical composition from the spectro, mechanical properties, sometimes hardness or an NDT result, tied to the heat that produced the castings.
When the underlying records live in separate registers, producing that certificate is a small assembly job every time, done under time pressure at dispatch, by transcribing numbers from one book to another. It is slow, it is the kind of task where transcription errors happen, and a certificate with an error in it is worse than a delay.
If the spectro result and the test data are recorded against the heat when they are taken, the certificate is generated rather than assembled. The same records then serve the harder case, which is a customer asking six months later which heat a particular casting came from and what its composition was.
Why building this is now practical
A foundry with two induction furnaces and a moulding line has never been able to justify custom software. A development team, a specification document and a six month build were never going to be recovered on castings sold by the kilogram.
That has changed. With Pentoggle you describe how your foundry runs, including your furnaces, your grades and charge practice, your patterns, your fettling stages and the certificates your customers require, and get a working application. When you add a grade, take on a customer with a new test requirement, or start recording sand parameters against heats, you describe the change and the application updates. Most foundries start with the heat record and rejection analysis, because linking those two is what turns rejection from a number into a cause.
Where foundries choose Pentoggle
Built around the heat
The unit everything in a foundry actually refers to, with its charge, parameters and results held together.
Rejection linked to cause
Defects aggregated by heat and pattern rather than by day.
Works alongside Tally
Pentoggle handles the plant. Your accounting stays where your CA already works.
Certificates from records
Generated from the heat data rather than typed at dispatch.
Changes in days
A new grade or a new customer requirement does not become a three month project.
The one number that runs a foundry
Rejection, split by defect and by the stage at which it was found.
A single rejection percentage is the number most foundries report and the least useful one available. The split does two things a percentage cannot. Splitting by defect points at a cause, since gas defects, shrinkage and sand defects have different origins and different owners. Splitting by stage weights the loss, since a rejection at machining costs several times what the same rejection costs at knockout.
Track it per heat and per pattern, reviewed weekly. Watch metal yield alongside it, because the two together tell you whether a bad month came from making castings badly or from making too little saleable metal, which are separate problems with separate fixes.
Ready to build foundry software?
Rejection is not found at inspection. It is decided at pouring.