The Tab Welding Evidence a Battery Pack Audit Should Require
A tab welding audit stands or falls on four things a plant can put in your hands at the weld station: parameter logs that tie a recipe to a specific serial number, destructive test records with the actual coupons still attached, calibration and maintenance history for every welder on the floor, and a live production run you did not schedule. Any one of those missing, and the rest of the workstation documentation is decoration — clean, dated, and worthless.
I start at the machine, not the conference table. The quality manager wants to walk me through the manual first; I ask to go to the line while it is running. What I want in the first ten minutes is not a document. It is the sound of the station.
An ultrasonic welder has a particular sound — a hard, quick buzz with a small bounce at the end when the horn releases. A resistance spot welder gives a soft thump and then a tick as the electrodes lift. Stand there long enough and you hear a rhythm, and the rhythm tells you whether the operator is placing and welding deliberately or just feeding parts. It also tells you when something changes. I have heard a line shift from a clean buzz to a duller, longer groan over the course of an afternoon, and that shift was a horn tip going away.
The three joints, and why each asks for different evidence
Most pack lines run two or three of these processes, sometimes all three under one roof.
Ultrasonic welding is what you will see on foil tabs — the thin aluminum and copper tabs coming off a pouch cell — welded down to a busbar or a collector. The horn descends, applies force, and vibrates at high frequency until the metals scrub together and bond. Amplitude, energy or weld time, and down force are the parameters that matter. The things that drift are tooling: horn tip wear, anvil flatness, contamination on the foil or the busbar face.
Resistance welding shows up on cylindrical cells, where a nickel strip gets spot welded to the can. Current, weld time, and electrode force. The evidence lives or dies on electrode tip condition and how often the tips get dressed.
Laser welding is increasingly common for busbar-to-terminal joints, especially aluminum to copper, and it usually happens inside a light-tight enclosure with an interlock. You may not be able to watch it directly. Power, weld speed, focus position, and shielding gas flow are the parameters, and the consumable that quietly degrades is the cover glass in front of the optics.
I am not there to validate metallurgy. That is not an audit. What I am checking is whether the plant knows its own window — the range of settings it considers in control — and whether it can show me the window held for the packs it already shipped.
Parameter logs: written by the machine or by a person at 6 a.m.?
Every plant has a log. That is not the question. The question is whether the log is generated by the controller or typed in by someone at the end of a shift from memory.
Ask three things. Is the recipe locked, and who holds the password? What is the alarm band around each parameter, and does the machine stop the cycle or just flag it? And can you see, for a given serial number, both the setpoint and what the machine actually delivered?
That last one matters more than people expect. Some controllers record delivered values — actual energy, actual force, actual weld time. Others record only the recipe number. If the log shows me a setpoint and nothing else, I am reading a wish, not a record. A plant that knows this will usually volunteer the distinction before I ask.
Then the move that separates the real programs: ask for the last twenty alarm events or rejected welds and match each one to a serial or lot. If the answer is that alarms get cleared and nobody keeps them, you have just learned what happens when a weld goes bad at 2 a.m. on a Saturday.
Destructive test records, and the coupons behind them
Peel tests, pull tests, cross sections — the names vary and the plant will use whichever one its procedure uses. What I care about is physical.
A cross section arrives in an epoxy mount with the date and machine number written on the side in marker. A peel-tested tab comes back as a bag of torn coupons with the substrate still attached. If those objects exist, the test is real. If the only artifact is a spreadsheet with a row of numbers and no operator initials, I assume the numbers were transcribed from a previous month.
Two checks I run every time.
First, ask them to cut a coupon while I stand there. A plant with a working program will do it in fifteen minutes at the bench. A plant without one will tell me the lab is offsite and it will take three days. The second answer is not a scheduling problem.
Second, look at where the coupon came from. The test piece has to come off the production machine, with the production recipe, at the interval the control plan specifies. I have seen plants weld their hourly coupon on a spare bench welder sitting in the corner with a gentler recipe dialed in, and then file the result next to the production records. That coupon proves nothing about the packs on the pallet.
And when the frequency in the work instruction says every two hours, count the coupons in the day's bin. If there are six coupons for a ten-hour shift, someone is welding tests on paper.
Finally, look for a failure. If the peel test log for the last two years shows no out-of-spec result anywhere, either the process is genuinely bulletproof or the test is not being run at the stated frequency, or it is being run and quietly repeated until it passes. Ask what containment looked like the last time one failed. Watch whether the answer names serials or names a process.
Calibration and maintenance: the counters do not lie, the paper sometimes does
At the weld station I want to see the force gauge, the load cell, the power supply, the chiller, and the sensor that tracks tooling life. For laser stations, cover glass hours. For resistance stations, the tip dressing counter.
What I am comparing is the paper record against the machine's own internal counters. When those two disagree, that is the moment the tone of the audit changes. Either the log was backfilled, or a counter was reset, and neither is good news for the packs already in the shipping area.
The other check is whether the calibration matches the plant's own procedure. I do not bring my own interval. I read theirs, then I look at whether the sticker on the instrument, the certificate in the folder, and the serial number on the certificate all refer to the same device. I have found current calibration certificates for a gauge that was not the gauge on the bench — same model, different serial.
Same logic for preventive maintenance. Does the interval in the maintenance record match the interval in the work instruction? If the work instruction says dress the electrodes every 500 welds and the maintenance log shows a dressing every 1,200, one of those documents is fiction, and you can usually tell which by asking the operator.
What the operator is actually looking at
Ask to see the visual acceptance criteria at the machine. Everyone has a laminated sheet. The tell is what is printed on it.
The common failure is a standard that shows only good welds — a perfect nugget, a clean surface, a uniform row of spots. That gives the operator a reference for what right looks like and nothing for what wrong looks like. So they approve welds by feel.
What I want on that board is the defect set: blowholes and porosity, incomplete penetration, expulsion and spatter, excessive deformation or a crushed tab, undercut around the nugget edge, heat discoloration running past the joint, a weld that is visibly off-center on the busbar. These are the signatures of process drift, and if the station has never photographed them from its own tear-downs, the operator's judgment is the process. Operator judgment goes home at the end of the shift.
Then the work instruction itself. Compare the revision letter on the laminated sheet at the machine against the master in the document control system. Every disagreement I have found this way was months old.
Operator qualification follows the same pattern. Who certified them, on which machine model, on what date, and what does recertification look like? Cross-training matrices taped to the wall are everywhere; what matters is whether the matrix has been updated since the last line change and whether the person standing at the welder is the person the matrix says should be there.
Traceability should get you below the shift
Pull a serial number off a finished pack on the floor and ask for its weld record. Then time how long it takes.
The gap I find most often is that the weld log is per shift, and the shift produced 300 or 400 packs. That is a batch, not traceability. Ask the direct question: if one pack comes back failed, can you narrow the suspect population below the whole shift, or do you have to quarantine the day? A plant that can answer that has thought about it. A plant that cannot has been getting away with a binder.
The run you did not schedule
A staged demonstration is choreographed. Two o'clock on a Tuesday is not.
Watch for the small things. Does the operator glance at the parameter screen before starting a cycle, or cycle blind? Does anyone pull the hourly coupon while you are standing there, or does the bin look like it was filled on Monday and not touched since? If a coupon pull is scheduled during your visit and it does not happen, ask when the last one was, and then ask to see it.
And then the move I make on every audit, which almost always produces a reaction. Ask the line lead to make a bad weld on purpose. Take a piece of scrap, run it deliberately outside the window — low energy, low force, short time — and watch whether the machine alarms and whether anyone notices.
You learn more from one deliberate bad weld than from an hour of reading logs. Either the controller catches it, the alarm is visible, and the operator reacts, in which case the plant just showed you its process is real. Or nothing happens, the part drops into the good bin, and you have measured exactly how much of the weld quality program exists outside the binder. I have had plants refuse this request. I have also had an operator catch it instantly and pull the part before I could ask, and that told me more about the line than the entire document review.
What actually changes the decision
A small pack shop running low volume may not have a metallurgical lab, and I do not hold that against it. A bench vise, a peel test, a dated bag of coupons, and a log that matches the machine counters will carry a lot of weight with me.
What does not carry weight is documentation that cannot be traced to an object or a serial number. A cross section exists in one place and one time. A parameter log with an actual-energy column exists in one place and one time. A binder can be assembled the week before an audit. That asymmetry is the whole reason I spend the first hour at the weld station instead of in the meeting room — because the physical evidence at that bench cannot be written retroactively.
