The Cell Lot Change You Must Catch Before Mass Production Begins

The cell lot change you have to catch is the one nobody announces. The mass-production cells are not the cells that were qualified, the part number on the purchase order never moved, and the first time anyone finds out is a field return eighteen months later. Three controls stop it: a first article approval that verifies the production cell lot against the qualified pilot lot at the cell level, a substitution control that makes any change of cell manufacturer, model, or lot a written buyer decision rather than a supplier judgment call, and a traceability validation that confirms the records still capture lot, firmware, and process data once the pilot crew is gone and volume takes over.

I have sat in that meeting more times than I want to count. Pilot build is finished. Engineering signs the approval sheet. Purchasing cuts the production PO. Three weeks later a pallet arrives, the packs look identical to the pilot units, the label carries the same internal part number, and nobody in the room can tell you which cell is inside.

The way I find out is not by asking. I walk to the incoming inspection bench, take one pack out of the carton, and read the cell date code through the shrink wrap or off the laser mark on the can. I do that before I open the traveler. If the date code is younger than the pilot build, the question is simple: when did the switch happen, and who approved it. If the answer is a shrug, the conversation is over and we are now in containment mode.

That habit is the whole article in miniature. Everything below is about making the same catch earlier, on paper, before the pallet ships.

The pilot build's real deliverable is a fingerprint, not a pack

When a pilot build finishes, the unit that gets photographed for the customer deck is not the valuable output. The valuable output is the record behind it. Every pack off that line should carry a serial number, and that serial number should point to four things: the exact cell lot, the BMS firmware revision, the process settings used on that build, and the date.

I do not care how the supplier stores it. A spreadsheet, a paper traveler, an MES record, a shared drive full of scanned forms. What I care about is whether I can hand them one serial number a year later and get an answer in minutes instead of days. If they have to go dig through a folder to reconstruct which cells went into unit 14, the record was never a record. It was documentation.

The difference matters because of where this record gets used. It is not used at pilot sign-off. It gets used when a pack comes back from the field, the customer wants to know whether the failure is isolated, and you need to compare the failed unit's construction against the baseline you approved. Without the pilot fingerprint, every field failure is a fresh investigation starting from zero.

First article approval: check the cell, not the enclosure

The first article inspection sheet for a pack will usually be thorough about the wrong things. Dimensions. Torque values. Hipot results. Enclosure fit. All of it legitimate, none of it answering the question that actually matters at this stage: is the cell inside this approved unit the cell that was qualified?

What I want to see on the FAI, specifically:

• Cell manufacturer, model, and lot code written out in full. Not the internal pack part number, not the distributor's part number. The cell maker's own identity.
• A photograph of the cell label or laser mark from the unit being approved, not from a stock image of the cell model.
• The BMS firmware version read out of the device itself, not copied from a release note.
• A plain statement that the production lot matches the qualified pilot lot, or an explicit description of what changed and who approved the change.

That last line is the entire point. First article approval is not a quality ritual. It is the moment where the production cell lot either gets confirmed against the qualified lot or gets flagged as a deviation. If your FAI form has no field for cell lot, the form is telling you the supplier does not treat cell identity as a controlled characteristic. That is worth knowing before you scale.

I also treat firmware the same way as cells. A BMS firmware revision change between pilot and mass production is a substitution, even if the hardware did not move. Cell parameters, protection thresholds, state-of-charge algorithms, and CAN message content all live in that firmware. If version 1.4 was qualified and version 1.5 ships, the qualification no longer describes the product.

Substitution control is a written-approval problem

A supplier swapping cells is almost never malicious. It is a shortage, a minimum order quantity, a lead-time squeeze, a better price from a broker, or simply a new hire who did not know the pilot lot was frozen. The failure is structural, not moral, which is why the control has to be structural too.

The rule I write into the program is blunt: no different cell manufacturer, model, or lot enters mass production without written buyer approval before it ships. Written meaning dated, attributable to a named person on the buyer side, and filed against the part record so it surfaces in the next audit. A phone call is not approval. A verbal assurance at a factory visit is not approval. An email is approval only if somebody files it.

What I want the supplier's quality lead to understand is that "same specification" is not the same cell. A 21700 cell with matching nominal capacity and chemistry from a different manufacturer is not a like-for-like swap. Impedance, weld characteristics, venting behavior, dimensional tolerance on the can, and the way the cell behaves under the BMS's coulomb counting can all differ. The datasheet lines that match are the ones that were easy to match. The rest is where your field returns come from.

There is an honest tradeoff here. Enforcing lot-level freeze on every cell lot, not just every cell model, costs money and creates friction when the supplier genuinely cannot source a matching lot. Sometimes a documented lot change is the right call. The control is not "never change." The control is "never change without the buyer knowing." Silence is the thing you are buying protection against.

Traceability degrades quietly when volume arrives

The pilot line and the mass-production line are different systems that happen to make the same product. At pilot, an engineer or a technician scans every unit by hand, types the cell lot into a field that was built for the purpose, and the record ends up immaculate because there were fifteen units and one person cared about each of them.

At mass production, the scan has to happen inside a takt time measured in seconds. That is where records come apart. The failure modes are consistent:

• Labels that read cleanly when scanned one at a time at a bench, but produce a soft beep or a no-read when the operator is moving and the label has a scuff or a wrinkle over the barcode.
• Traveler fields that quietly become free text because the dropdown list was never updated with the new cell lot, so the operator types something approximate.
• End-of-line test records that keep the pass or fail flag but drop the raw data underneath it, so the weld resistance curve or the cell voltage at receipt is gone by the time you need it.
• Serial number sequences that develop gaps because reworked units get re-serialized without the original being closed out.

If the supplier moves from manual assembly to automated assembly between pilot and mass production, this gets sharper. Automation can improve traceability enormously, but only if the data path was designed. I have seen automated lines where the station controller held the cell lot in a buffer that got overwritten every shift, and nobody had set up the export. The line was collecting data and throwing it away. The manual pilot line, for all its slowness, had better records.

The validation test I actually run

The test is not a document review. It is a reconstruction.

Pick three units at random from a production shift. Pull their records. Then try to rebuild the build from the record alone: which cell lot, which firmware, which workstation, which process settings, which test result. Do not call anybody. Do not ask the supplier's engineer to explain. If a field is blank or ambiguous and you cannot resolve it from the record, the record failed.

I run that test at pilot, again about two weeks into mass production once the line has settled into its real rhythm, and again after any significant change: new automation, a new shift, a new cell lot, a new firmware revision, a change of line supervisor. The two-week check catches the most, because that is when the pilot crew has gone home and the production crew is running the process on its own.

A note on retention: those pilot records are your baseline for comparing field failures, so they need to live at least as long as the product's warranty period. Not the supplier's standard document retention window, which may be shorter. If the warranty runs three years and the records get purged at eighteen months, the comparison you need most is the one you can no longer make. Put the retention requirement in writing and verify it, because retention policies drift toward whatever the file server defaults to.

The checklist line that needs two signatures

Most pilot-to-mass-production checklists I see treat traceability as a checkbox: traceability, present, yes. That line gets signed by whoever is holding the clipboard.

Make it a real item with real evidence. Traceability readiness, verified by pulling three serial numbers and reconstructing the build, signed off by the buyer's quality lead and the supplier's quality lead. Two names, two dates, one shared record. When both quality functions have personally walked the reconstruction test, the conversation about cell lots stops being a procurement topic and becomes a quality topic, which is where it belongs.

What this does not solve

Full cell-level traceability on every lot has a real cost, and some suppliers, particularly smaller pack builders, cannot deliver it without changing their whole information system. Some cells pass through distribution layers where lot definition itself is fuzzy, and a supplier may genuinely not be able to tell you more than the date code range. You have to decide where on that spectrum your program needs to sit, based on what a field failure would cost you and how much of your product liability sits on being able to say which cells went where.

What is not negotiable is knowing which side of that line you are on. A supplier who cannot provide cell lot traceability and says so is a manageable risk. A supplier who cannot provide it and says they can is the one that puts you in the meeting eighteen months from now.