The vintage oven that now talks back

The problem

The curve that gives the battery its energy disappears every time the screen refreshes.

The formation/charge curve is literally what gives the battery its energy — a curve out of range directly predicts low capacity or CCA, but today nobody correlates curve to final test result because the panel's data dies every time the screen rotates.

• The curing chamber or the formation rectifier is typically over a decade old, isolated from the network, and shows humidity, temperature or charge current every few seconds on a local screen. Many were installed when nobody expected them to share data with anything.
• The formation charge curve is literally what puts energy into the plate. A curve out of range is a direct predictor of low capacity or low CCA at final test. It is the step that decides whether a battery meets its rating or comes back under warranty.
• Yet nobody correlates curve to test result, because the panel's data dies every time the display rotates to the next value. The operator may glance at it, but no record survives that can be compared with the final test bench.
• Replacing the equipment to get that data means hundreds of thousands of dollars and weeks of downtime in the plant's most sensitive chemical step. So the plant keeps running equipment it cannot learn from, and postpones a decision it cannot afford. Free lead after curing and capacity after formation finally get a process history to be compared against, lot by lot.

How it fits the IRIS system

Connect in photo mode on the panel — the oven stays exactly as it is.

Connect photo mode over the panel with a fixed industrial camera. Vision OCR with an anchored LLM reads every frame, structures the full time-series curve, and cross-references it with the ERP to know which lot was in process at that moment. Zero modification to the original equipment.

No ports opened, no PLC touched, no retrofit. A camera reads what the operator already sees, and the plant finally keeps the curing or formation curve of every lot next to its final test result. It is the same oven, the same rectifier and the same operator, with memory added from the outside.

See the full IRIS architecture →

Before and after

The isolated curing and formation panel versus the panel being read

Aspect	Today	With iLEAN Connect
Formation charge curve	Visible on screen, then lost	Full time series stored per lot
Curing humidity and temperature profile	Spot readings on a sheet	Continuous record per chamber load
Low-CCA batteries at final test	No link to the process	Root cause by lot in minutes
Which lot was in the rectifier circuit	Reconstructed from memory	Cross-referenced with the ERP
Changes to the original equipment	—	None: external camera only
Replacement CAPEX over $300K	Pushed onto this year's plan	Deferrable indefinitely

Zero correlation between curve and final test result → root-cause analysis by lot in minutes. Urgent equipment replacement → indefinitely deferrable.

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 5-10 months.
• On top of that, avoided replacement CAPEX above $300K for each oven or rectifier you keep running. That makes it the cheapest way to learn from equipment you were planning to replace anyway.
• Scrap reduction from correlating curve defects with final test, a correlation that is impossible to make today.
• And root-cause analysis on a low-capacity lot goes from guesswork to minutes.

Estimated payback 5-10 months, plus avoided replacement CAPEX (>$300K) and scrap reduction from curve-defect correlation. *Figure to validate*.

And the fair question from the production manager

“Can a camera reading an old panel be trusted as process data?” — here the task is anchored: fixed display, values in known positions and expected physical ranges, where the best models drop below 1.5% error [1]. Any frame whose value falls outside the possible range for that formation program is discarded instead of stored, so the curve never carries an impossible point.

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