What a Battery Pack Factory Audit Should Verify Before Production
A battery pack factory audit is a procurement risk screen, not a compliance ceremony. The one thing it has to prove is that a finished pack can be traced — on the spot, without advance warning — back through its end-of-line test record, its BMS firmware revision, its weld parameters, and its cell lot to one specific production order. When that chain holds and the factory can walk it for you under its own roof, you have a supplier you can scale with. When a link is missing, you have found out while it is still cheap.
I ask for the production traveler before I take the coffee. Not the quality manual in the frame on the wall, not the certificate — the router that follows one work order down the line, the one with the lot numbers written on it and initials in the corner. Everything I care about for the next few hours hangs off that document, and I want it in my hands before anyone has had a chance to bring me a nicer version.
So picture the floor. Fluorescent light, two half-built packs on fixtures, a spot welder firing every few seconds, someone at a laptop at the end of the line pulling test data off a rack. You're at station three with the router in one hand, and you point at the cell lot number printed on it.
Show me those cells.
That one request does more work than any checklist I've ever carried. Watch how long it takes. Watch whether the answer comes off a shelf, out of a system, or out of somebody's memory. A factory that can produce the receiving record, the incoming inspection data, and the cell re-sort sheets for that lot in a few minutes is a factory that has done this before with customers who ask hard questions.
Match the audit type to the stage
Three different things get called a factory audit, and buyers mix them up constantly.
A desk audit is paperwork, done remotely: QMS certificates, UN 38.3 test summaries, calibration records, production flow diagrams, org chart, any field failure or recall history. Its only job is to tell you whether a trip is worth booking.
A process audit is the one that matters at pilot and mass production. You're verifying that the controls — cell receiving, weld parameter setting and recording, BMS flashing and version control, end-of-line testing and data retention — actually function during a real build, rather than in a binder.
A product audit takes finished goods and tests them against the specification you're buying. It answers "is this pack good," not "can this factory make every pack good."
At sample stage, a desk audit plus product testing on the samples is usually enough. You're buying a design, not a production capability. At pilot, you want a process audit with a live run, because the pilot is where the process either exists or it doesn't. By mass production you're doing process audits again, but now you're also pulling historical records — EOL data, weld logs, traceability records from real orders — rather than watching a demo.
What to demand before you get on the plane
The desk package isn't a formality. It's how you decide what to look for on the floor.
Ask for the QMS certificate and read the scope statement, not just the logo. Plenty of pack assemblers hold a certificate that covers a different site, or a scope that stops short of the process you're buying. Ask for UN 38.3 test summaries for the cell and for the pack, and check that the pack summary reflects a configuration close to yours. Ask for a calibration list — every measuring and test instrument used on the line that will build your pack, with current calibration date and interval. Ask for the production flow diagram with inspection points marked, and then, on site, compare it to the actual line layout.
There's one thing I always do with the flow diagram. I ask them to mark, on paper, exactly where each cell lot number first attaches to a work order and where it stops being traceable. Most diagrams make that look clean. Most floors make it look messier.
Follow one pack from cells to carton
On site, stop touring and start tracing. Pick one work order and stay with it.
Incoming cells. Where did this lot come from — cell maker direct, authorized distributor, or open market? Is there a receiving inspection record, and does it cover voltage, internal resistance, and capacity on a sample? Do they re-sort or match cells before pack assembly, and if so, is the sorting data stored per lot? If they don't sort, ask how they handle a lot that spans more than one cell manufacturing date. Mixing cell dates inside a single pack is a traceability problem even when the pack tests fine.
Tab welding. This is where I spend the most time, because it's where process control either exists or is theater on a wall chart. Ask what welding technology and equipment, then ask the harder question: what were the parameters on the weld used to build this pack? Not what's loaded on the machine right now — what was used, and how do you know? A factory with real control has a weld record tied to the work order or lot: energy or current, weld time, electrode force, and often a monitoring trace per weld. A factory without it will show you the recipe screen and hope you don't ask again.
Then look at how the welds are verified. Pull tests or peel tests on a sample basis, X-ray or cross-section on setup or on some periodic basis, and a documented accept/reject threshold. Ask who signs off when a weld check fails, and what happens to the packs built since the last good check. That last question is what separates a control system from a testing habit.
BMS flashing and firmware. Ask to see the flashing station. You're looking for a log, not a bench: serial number, firmware version, checksum or hash, timestamp, operator or station ID. Then ask to see the approved firmware and configuration file and compare it against what is actually on the production PC. Config files are where variants hide. Two different packs can run identical firmware with a different parameter file, and a file named something like "standard" tells you nothing. The check you want is byte-level: the file on the line equals the file you approved.
End-of-line testing. Find out what is tested at EOL — open circuit voltage, internal resistance, insulation resistance or hipot, BMS communication and fault codes, sensor readback, balance function, whatever your specification requires — and then find out whether the pass/fail limits can be edited at the station. If the operator can change a limit, that's a finding, no matter how good the intentions are. Locked limits, versioned test programs, and a record stored per serial number is the shape you want. Ask to see the EOL record for a specific serial number from a lot built weeks ago, then ask how long those records are kept and who can retrieve them.
Final inspection and pack identity. Confirm that the label on the finished pack ties back to the test record and the work order. Check connector torque control — either a recorded torque driver or a documented torque with a calibrated tool. Check enclosure sealing where it applies, and look at the finished packaging as it will ship, including the UN packaging marks that dangerous goods shipping depends on.
Where the chain usually breaks
The same gaps show up, factory to factory, and they're almost never about the machines. They're about the record.
Cell lot traceability dissolves at the distributor boundary, or when cells get pre-sorted into bins and the bin loses its lot identity. Weld parameter logging exists as a screen and not as history. Firmware revision control is a folder on one engineer's laptop with no log and no comparison to the approved file. Test equipment carries a calibration sticker that expired two quarters ago, or the calibrated instrument sits in a drawer while an uncalibrated one is in use. And EOL data exists only as a pass/fail count, not as a record tied to a serial number.
None of these are exotic. All of them are invisible from a tour route.
Grade findings, then decide
A finding list nobody grades becomes a list nobody acts on. I sort everything into three buckets while I'm still in the building.
Critical: it breaks traceability, safety, or regulatory standing. No cell lot link to the work order. No record of weld parameters. An approved firmware mismatch. EOL limits editable at the station. Uncalibrated test equipment actively in use.
Major: the control exists but is unreliable or undocumented. Weld verification done but not recorded. EOL records kept but not retrievable by serial. Incoming inspection performed on cells but sample size never specified.
Minor: labeling, housekeeping, paperwork discipline.
Critical findings mean no approval. Not "conditional with a plan," not "we'll keep an eye on it." You can't audit around a traceability hole. It comes back as a field failure and a recall you can't bound. Major findings can support conditional approval, but only with containment on already-shipped product, a written corrective action with a due date, and a defined exit criterion somebody actually re-checks. Minor findings go on the scorecard and get watched at the next visit.
If you run a supplier scorecard, this is the cleanest place to use it: weight quality and process control findings rather than averaging them, so one critical can't get diluted by twenty clean rows.
Corrective actions and closing the loop
For anything major or critical, ask for a corrective action in a structured format — 8D or your equivalent — and read the root cause section carefully. "Operator error" and "we will retrain" are not root causes. A real one looks like: the weld log was stored locally on the machine controller with no export, so no history existed beyond the last build. The fix is a controller configuration change and a retention procedure; the evidence is a retrieved log from a new lot.
Then verify closure. A signed document saying "corrected" is a claim, not evidence. Effectiveness checks take one of a few forms: a re-audit of the specific area, a sample of new production records showing the control now works, or a lot-based check across the next two or three builds. Give it a date. Without a date, closure is just a story.
One day is not a capability study
An audit where you watch a line run for four hours and follow two or three packs will tell you more than a two-day tour of an idle factory. Production capability shows up in rhythm: how the operator handles a bad weld, what the line lead does when the tester throws a fault, whether the EOL record gets stored or skipped because the shift is behind. None of that happens on a stopped line.
So push for a scheduled production day. If they can't run your product or a close relative of it, watch whatever they are building — the controls are either there or they aren't. Then build in the follow-up. What you're really approving isn't the factory you saw on Tuesday. It's that factory's ability to keep the control system intact in month six, after the launch energy has faded. The records you pull six months from now will tell you which one you actually got.
