Grid casting start, no paper

The problem

The first link of the battery's traceability sits in a folder.

Grid casting is the first traceability link of the whole battery — if alloy or plate weight get logged wrong, the defect isn't caught until final test or, worse, until warranty. Paper logs can sit undigitized for hours, and when a non-conformance shows up weeks later, nobody can reliably reconstruct the alloy or pour temperature of that lot.

• Every shift, the grid caster starts with a paper log: lead-calcium alloy, target plate weight, pour temperature, book mold in use and the grid lot number. All of it handwritten, often with a correction in the margin when the pot temperature had to be adjusted.
• That sheet is the traceability origin of every plate that will later be pasted, cured, formed and welded into a battery, yet it stays undigitized for hours. Pasting and curing records exist too, but they all hang from a casting lot that is not yet queryable.
• A wrong alloy or a plate weight off target is not caught at casting: it surfaces at final high-rate discharge test or, worse, as a warranty claim months later. By then the grids have been pasted, cured, formed and shipped inside thousands of batteries.
• When a non-conformance shows up weeks after the pour, nobody can reliably reconstruct which alloy or which pot temperature that grid lot came from. The answer ends up being an estimate, and the containment is drawn wider than it needs to be.

How it fits the IRIS system

Connect in photo mode — the grid caster keeps the same log he fills in today.

Connect photo mode. The caster photographs the log with his phone at shift start, no new app, no habit change. An anchored model recognizes the log's layout and extracts alloy, weight, temperature and lot in seconds, landing in iLEAN's central memory before the pour even finishes.

Nothing changes at the caster: same sheet, same pen, one photo at shift start. What changes is that the grid lot already exists in central memory before the first rack of grids leaves for the pasting line. It is usually the first case approved, precisely because it asks nothing of the person at the pot.

See the full IRIS architecture →

Before and after

The grid casting log on paper versus captured at the pot

Aspect	Today, on paper	With iLEAN Connect
Grid lot in the system	Hours after the pour, if transcribed	Zero-second latency, before the pour ends
Lead-calcium alloy per lot	Handwritten, hard to query	Structured field tied to the grid lot
Target plate weight and pour temperature	A number in a folder	History queryable by caster and shift
Tracing a low-CCA complaint back to casting	Hunting through binders	One lookup by grid lot
Casting supervisor's transcription time	Every shift, by hand	Recovered
Grid caster's routine	—	Unchanged: one phone photo

Hours of latency between pour and digital record → zero-second. Manual reconstruction on a non-conformance → instant lookup by grid lot.

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 4-9 months, depending on daily grid lot volume.
• Most of the return comes from the casting supervisor's transcription time, recovered every single shift.
• From hours of latency between pour and digital record down to zero-second.
• And a non-conformance stops being a manual reconstruction: it becomes an instant lookup by grid lot.

Estimated payback 4-9 months depending on daily lot volume, mostly recovering the casting supervisor's transcription time. *Figure to validate*.

And the fair question from the production manager

“What if it misreads the alloy or the plate weight?” — reading fields from a log with a fixed layout is an anchored task, where the best models drop below 1.5% error [1]. On top of that, each value is checked against its physical range for that alloy and grid type, and the casting supervisor confirms the summary before it reaches the ERP.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.