EV battery second life with AI — a misclassified module is not scrap, it is a loss event.

The problem

The pack's history lives on a screen, its real condition on the shop floor and its destination on a sheet from the buyer.

Anyone managing EV battery second life is caught between three worlds that do not talk to each other:

1. Each pack's BMS — cycles, depth of discharge, fast-charging events, peak temperatures. It is all there, but in the manufacturer's CAN/UDS protocol; if you handle packs from three brands, you have three diagnostic tools and three screens.
2. The physical condition — pack deformation, terminal corrosion, signs of thermal runaway in a module, impact damage. Only the person who opens the pack sees that, and it gets written on a report that almost never makes it into the dossier.
3. The buyer's criteria — stationary ESS tolerates a low SoH and a high cycle count; a scooter does not; an industrial tool demands yet another range. Every customer, its own rule, on a sheet held on their side.

The technician sorting a pack does it with incomplete information — and in a hurry, because the flow of packs reaching the end of their first life is on a steep curve. A doubtful module sent into an ESS installed in a building is hard cost and, worse, brand damage. The right decision exists in the data; the problem is that the data is not in the same place when someone has to sign for it.

How it fits the IRIS system

iLEAN does not force you to unify BMS systems — it reads them as they are and leaves the final decision to the person.

The EV second-life problem is not a lack of technology in the plant, it is fragmented information at the critical moment (the pack's arrival at the shop). iLEAN acts as the putty between the BMS, the visual inspection and the customer, without asking you to change diagnostic tools or protocols.

Edge sees the pack and every module. Connect reads the BMS however it comes — documented DBC or the manufacturer's screen. The agent cross-references with the buyer's criteria and assembles the dossier. The person signs — critical packs do not leave on their own.

The three iLEAN pieces applied to EV battery second life:

• Edge — a terminal with machine vision (CNN) over the inspection area. It detects pack deformation, terminal corrosion and signs of thermal runaway, module by module. It works without a network. If the plant loses WiFi, Edge keeps inspecting and holding doubtful packs.
• Connect — captures the BMS history through a documented DBC when the manufacturer provides one, or by reading the diagnostic software screen when it does not — the information enters the system in a uniform format, whatever brand it comes from. And it captures what arrives from outside: the buyer's email with a new SoH threshold, the WhatsApp message from the dealership flagging a previous thermal event that was never recorded.
• Agent — cross-references the calculated SoH, the Edge visual inspection, the BMS history and the buyer's criteria. It proposes the destination (ESS / scooter / recycling), builds the dossier per pack and alerts the manager. The person validates; a critical pack is never labeled on its own.

See the full IRIS architecture →

Before and after

Manual sorting vs. cross-referenced sorting with iLEAN

Aspect	Manual inspection + spreadsheet	With iLEAN Edge + Connect + Agent
Reading the BMS	One screen per brand, data copied by hand	Direct integration or visual reading of the software itself, uniform data
Visual inspection	The technician's eye, loose photos, a paper report	Edge sees every module and logs every defect with photo and timestamp
Buyer's criteria	A PDF from the sales rep, a different version per customer	Live rules inside the agent, updated at second zero
Destination decision	The technician decides alone, under time pressure	The agent proposes with cross-referenced data; the person signs
Dossier per pack	Assembled after the fact, with data missing	Assembled on the fly, complete, ready for the customer
Traceability if it comes back	Searching through archives, days of work	History by serial number, instantly

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your operation. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Remanufacturing shop with a mid-sized mixed flow of EV packs (several brands, output to ESS and secondary applications).
• Edge pilot in the intake area (camera over the inspection station + integration with the dominant BMS diagnostic tool). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, driven by the reduction in misclassified packs that come back (buyer returns) and by sorting speed, not by operator hours.
• The hard lever is a single module with undetected thermal risk reaching an installed ESS. One unit avoided pays for the pilot several times over.

And IT's reasonable doubt

"What if the AI misreads the BMS history and classifies the pack wrong?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualizes a piece of data from one system into another (reading the BMS screen and posting it into your database), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN proposes the destination and the person signs. The three safety rings exist precisely for this.

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