What Each UN 38.3 Test Proves and What Your Summary Skips

A UN 38.3 test summary is worth reading only when you can map every row of its results table back to one of the eight tests, know which of those tests were run on cells and which were run on the finished battery, and notice the test names that are not in the table at all. The document is a compressed record of specific cell and battery models pushed through a specific edition of the Manual of Tests and Criteria. It is not a certification and not a badge, and the "Passed UN 38.3" line on a supplier's product page is not a summary.

Two windows, one quote, one PDF

I do this late in the day, when the quote has stopped moving. Left window: the pack quote, cell model spelled out in a line item, pack model in the header. Right window: the test summary the supplier attached, anything from one page with a logo on it to eight pages of tables. Between them, a scratch file where I type the eight test names in a column before I read a single result. Altitude. Thermal. Vibration. Shock. External short circuit. Impact or crush. Overcharge. Forced discharge.

Eight lines of typing. It takes a minute, and it is the difference between auditing the document and being led through it. If you read the table top to bottom first, the document defines the universe of what was tested, and you will never notice what isn't there.

Then I read the table and cross names off. Whatever is still uncrossed when I reach the end is the actual story.

Here is what each one is really doing, and which hardware it is doing it to.

T.1 and T.2: pressure and temperature, working on the enclosure

Altitude holds the sample at low pressure for several hours at ambient temperature. For cells and for batteries. What it stresses is the pressure differential across the case, the seals, the gasket, the vent, the feedthroughs for the terminals. A welded aluminum housing with a properly sized vent shrugs at it. Stamped steel and a gasketed lid can go either way. What it does not tell you is anything about your pack at 5,000 feet in a hot building, because it is one mechanical event, not a service condition.

Thermal is 75°C for at least six hours, then minus 40°C for at least six hours, repeated ten times, then a settling period at room temperature. Cells and batteries. This is a materials test wearing a temperature costume. It is not measuring capacity fade. It is measuring whether the shrink on a cell splits, whether a weld cracks, whether an adhesive loses grip, whether a solder joint on the protection board opens after ten expansion cycles. Ten soaks is almost nothing against a pack that lives on a three-shift floor, and that is not a flaw in the test. It was never asked to do that job.

T.3 and T.4: joints and mounts

Vibration sweeps from 7 Hz to 200 Hz and back, fifteen minutes per sweep, twelve sweeps per axis, three axes. Nine hours on the shaker. Cells and batteries. What it stresses is every mechanical interface in the assembly: tab welds, nickel strip, busbars, connector retention, anything held by a screw whose thread locker was applied by a person at the end of a shift.

When the summary says vibration passed, I check whether the line is the cell or the battery, because those two fail differently. A cell can pass on a shaker and the battery built from it can fail, because the pack adds mass, length, and a different resonant response. The line in the table with the cell model next to it proves nothing about the pack.

Shock is a half-sine pulse applied three times in each of six directions, with the g level stepping down as the sample gets heavier. Cells and batteries. This is closer to a handling and freight test than a service test. It stands in for a carton coming off a pallet badly or a pack sliding off a bench. What it stresses is retention: brackets, mounts, standoffs, fasteners. It says nothing about a low-speed bump into a rack leg, which is a completely different pulse shape and duration.

T.5: the one that tests the protection path, not the cell

External short circuit sits the pack at 55°C, bridges the terminals through an external resistance below 0.1 ohm, holds it for an hour after the case temperature comes back to 55°C, then watches for six hours. Cells and batteries.

Of the eight, this is the one that is mostly a test of the protection chain rather than the cell. What is under stress is the overcurrent path: the FETs, the fuse, the contactor, the current sense, the firmware that decides when to open. If the pack has no protection board at all, then the external short is being survived by the cell's own internal resistance and construction, and you should know which of those two situations you are looking at, because the table uses the same word — pass — for both.

T.6, T.7, T.8: the three that only run on one side

This is where short tables get short, and it is also where short tables are sometimes completely correct.

Impact is cells only. A bar is laid across the cell, a weight is dropped on the bar, and the question is whether the cell's internal construction holds up. It runs on a bare cell, not on anything installed in a pack. So a battery summary with no impact line is not missing anything.

Crush is batteries only. A controlled force is applied to the pack and held. What it stresses is the module structure and the load path through the pack's own architecture. This one is genuinely configuration-specific: it is a statement about the exact assembly that was crushed, which is one more reason the model number line matters so much.

Overcharge is batteries only. The pack is driven at twice the manufacturer's recommended maximum continuous charge current and held past the normal cut-off. What is under stress is the charge-side protection working independently — the backup cutoff, the overvoltage comparator, the fuse in the charge path. If the protection opens at the right moment, nothing happens, and that is the pass. If the only thing standing between the pack and the overcharge is the microcontroller that has already failed, you learn it here or you learn it in the field.

Forced discharge is cells only. A cell is driven into reverse polarity by an external supply. This is the overdischarge scenario in a series string, where one weaker cell gets dragged below zero by the others. It runs on bare cells, one at a time, not on the assembled pack. A battery summary with no forced discharge line is correct.

Now count the rows

Two items across the five tests that apply to both gives ten rows. Impact on cells, crush on batteries, overcharge on batteries, forced discharge on cells gives four more. That is thirteen or fourteen rows depending on whether the lab writes impact and crush as separate lines or folds them into one, and both formats are fine. If the table has nine rows, I want to know which four are gone before I read the word pass anywhere on the page.

The lines above the table

The Manual lists what a summary has to contain, and it is a short list. Three of those elements do most of the work for a buyer.

The edition or revision of the Manual used for the testing. The requirements have been amended over time, and a summary that does not name the revision cannot be lined up against what your freight forwarder or your customer's safety reviewer is working from. This is the cheapest line in the document for a supplier to fix and one of the most commonly left off.

The cell and battery description. Model numbers, chemistry, physical description, rated capacity, voltage, watt-hour rating. I read this against the quote with both documents side by side, character by character on the model numbers. A revision letter in a cell part number is exactly the kind of thing that disappears when a supplier second-sources a cell to protect a lead time. If the cell model on the summary is not the cell model on the quote, the whole document describes hardware you are not buying. That sounds paranoid until the first time it happens to you, and then it stops sounding paranoid permanently.

The test report identification. This is the pointer back to the underlying report, which is where the curves, the photos, and the sample counts live. You will probably never be sent the report unless you ask for it. But the summary is supposed to carry an identifier so the report can be produced on demand, and a summary with no identifier cannot be traced to anything.

What none of the eight will ever tell you

Not one of the eight tests runs at your duty cycle. They are a transport-safety screening standard. Altitude, thermal, vibration, shock, external short circuit, impact or crush, overcharge, forced discharge — eight names, and not one of them is your load profile, your ambient, your vibration spectrum, your charge rate, or your depth of discharge.

So I use the summary for exactly one job: confirming that the hardware on the quote has been through the standard screening and that the document matches that hardware. Then I close it, and I go design the pilot that will actually exercise the duty cycle, because that is the only evidence that will still be standing in year three.