TL;DR: A supplier’s COA is only as useful as the test conditions printed on it — if discharge rate, temperature, and sample size are missing, treat it as a marketing document, not a quality record.
TL;DR: In our incoming inspection protocol, we reject entire lots when more than 3 out of 30 sampled cells fall outside ±2.5% of the stated nominal capacity — a threshold that catches grade-mixing before it reaches pack assembly.
What a Thermal Runaway Event in Rotterdam Actually Traced Back To #
In Q3 2023, a European systems integrator deployed a 48V 200Ah lithium iron phosphate pack bank configured as 4S2P. The packs had passed visual inspection, and the supplier had provided COAs showing 206Ah capacity per cell. Eighteen months into field operation, one pack triggered thermal runaway during a routine charge cycle. The integrator traced the root cause not to a BMS failure or a wiring fault, but to internal resistance mismatch between two parallel cell strings — one string had been built using cells from a different internal batch, with DCR values averaging 1.47 mΩ versus 0.89 mΩ in the adjacent string. The COA the factory had provided showed a single DCR value for the entire shipment. No per-cell data. No batch separation. The integrator had no way to know until the pack failed.
The total cost: €214,000 in recalled units, replacement logistics, and lost installation time. The integrator’s internal post-mortem found that the supplier had silently introduced cells from a secondary source when their primary cell supplier couldn’t meet a Q3 delivery commitment. Nothing in the COA — and nothing in the purchase order’s incoming inspection clause — had required the supplier to disclose batch provenance.
That’s the failure mode that supplier qualification for series-parallel configurations is actually designed to prevent. Not obvious quality problems. The subtle ones.
The COA Fields That Predict Pack-Level Failure Before Assembly #
A COA from a Shenzhen-based pack house typically includes: nominal capacity, nominal voltage, internal resistance (DCR), self-discharge rate, and cycle life claim. Most of those numbers are usable. Most of them are also incomplete without the test conditions that generated them.
Capacity figures are the first place to scrutinize. If the COA states 100Ah without specifying discharge rate and temperature, you’re looking at a number that could have been measured at 0.2C in a 25°C climate chamber — a best-case condition that understates what real cycling at 0.5C or above will deliver. In our experience auditing cell lots over the past four years, cells derate by 6–9% moving from 0.2C to 0.5C discharge under otherwise identical conditions. A pack design spec built around the 0.2C figure will underperform in any application cycling faster than that.
Internal resistance is the second field to interrogate. DCR measured at 1kHz AC impedance versus DCR measured via DC pulse (the IEC standard method under IEC 60896-22) can differ by 15–30% on the same cell. If the COA doesn’t specify which method was used, ask. If the supplier doesn’t know which method their lab used, that tells you something about their QC process that no amount of paperwork will fix.
Cycle life claims deserve the most skepticism. The UN 38.3 test protocol does not test cycle life — it tests transport safety. A 2,000-cycle claim on a COA needs its own supporting test report: the C-rate for both charge and discharge, the depth of discharge, the ambient temperature, and the end-of-life capacity threshold. Without those four parameters, 2,000 cycles could mean almost anything. We’ve seen suppliers quote 80% EOL at 0.2C/0.2C in 25°C — a number that collapses to 1,347 cycles at 0.5C/0.5C in a 35°C enclosure. That delta matters enormously for a portable BESS deployed in the Middle East or Southeast Asia.
The field most commonly missing from COAs for series-parallel configurations is cell matching tolerance. For a 2P or higher parallel configuration, you need to know the spread of open-circuit voltage (OCV) across the shipped lot, not just a nominal value. A ±5mV OCV spread is acceptable for 2P assembly. Above ±15mV, circulating currents during charge will cause one string to chronically overcharge while the other undercharges — a slow degradation path that shows up at cycle 300, not cycle 30.
| COA Field | Minimum Acceptable Specification | Red Flag Condition |
|---|---|---|
| Capacity | Rate + temperature stated (e.g., 100Ah @ 0.5C, 25°C) | Rate or temperature absent |
| DCR | Method stated (AC 1kHz or DC pulse per IEC 60896) | Single value, no method |
| OCV spread | ±5mV or tighter for parallel groups | Not stated; or “matched” without threshold |
| Cycle life | C-rate, DoD, temperature, EOL threshold all present | Cycle number with no test conditions |
| Batch/lot number | Single lot ID per shipment, traceable to production date | Multiple lot IDs mixed; or no lot traceability |
The most commonly overlooked field in our QV-03 qualification checklist is OCV spread. Factories know buyers check capacity and DCR. Few check matching tolerance — and some factories don’t grade cells for matching at all unless contractually required to do so.
Decision Framework — When the COA Determines Your Pack Architecture #
If the COA shows strong lot homogeneity — OCV spread under ±4mV, DCR spread under ±0.15mΩ, and single-batch provenance — then a 2P or 4P parallel configuration is low-risk without active cell-level monitoring at the string level. This is the scenario where passive balancing BMS at 60–80mA is sufficient, and the BMS engineering guidance under IEC 62619 for secondary lithium cells covers the protection layer adequately.
If the COA is strong on capacity but silent on DCR spread, the approach changes. You’re flying partially blind on internal resistance matching. For a 4S2P or larger pack, I’d push for a pre-shipment DCR audit: 10% sample, measured DC pulse at 50% SOC, with a pass criterion of ±0.2mΩ within each parallel group. The cost is a 3–5 day delay and a modest lab fee, typically under $400 for a 30-cell sample. That’s the right trade when the alternative is a field return.
If the supplier can’t produce a traceable lot-level COA at all — meaning they’re offering a blanket document covering multiple production runs — do not proceed to parallel group assembly. A single COA covering cells from different months means different calendrical aging states entering the pack simultaneously. The faster-degrading cells will pull down the parallel group’s effective capacity over time, and the BMS will misread SOC because cell divergence wasn’t accounted for in the initial pack model. For UL 9540A compliance — relevant for any pack sold into the US market — traceability to the cell level is increasingly a hard audit requirement, not a best practice.
For buyers sourcing from Dongguan-area pack manufacturers specifically: the region has strong process capability for 18650 and 21700 cylindrical formats but uneven practice on prismatic cell matching documentation. Our 2024 audit of 9 Dongguan suppliers found that 4 of them did not grade prismatic cells by OCV before pack assembly — they relied on the cell manufacturer’s own binning. That’s acceptable only if you have a direct relationship with the cell manufacturer and can verify their binning procedure. For a pack house buying spot inventory, it’s not a safe assumption.
End with a specific non-obvious boundary: the COA-based qualification logic above applies cleanly to LFP chemistry. For NMC or NCA cells in parallel configurations, the stakes on OCV mismatch are higher — the flatter charge curve means a small OCV spread represents a larger SOC difference than it would in LFP. For NMC, tighten the OCV matching threshold to ±3mV and confirm it in writing with the supplier before the first purchase order.
See our cell technology sourcing guides for cell-level grading criteria across LFP, NMC, and NCA formats.
Sourcing Guidance for Buyers #
When evaluating Chinese suppliers for series-parallel pack configurations, the first document to request is a per-lot COA with lot-specific test data — not a generic product specification sheet. If the supplier sends you a datasheet instead of a COA, or if the COA shows the same values for every shipment across six months, those are signals that their QC output is generated, not measured.
A qualification red flag specific to this category: suppliers who cannot distinguish between “cell capacity” and “pack capacity” in their documentation. It sounds elementary, but a supplier who quotes 200Ah as both their 4S1P pack capacity and their individual cell capacity is either confused about their own product or hoping you are. Either condition is disqualifying.
For incoming inspection, our protocol uses a 30-cell sample from each lot. We measure OCV at nominal SOC and DCR via DC pulse at 50% SOC. Pass criteria: OCV within ±3mV of lot mean, DCR within ±0.18mΩ of lot mean, with a maximum of 3 cells failing either threshold before the lot is flagged for full 100% inspection. Capacity spot-check uses a 5-cell sample at 0.5C discharge, 25°C ± 1°C, with a pass threshold of 98% or greater of the COA nominal value. This protocol takes roughly 14 hours of bench time per lot. For buyers who want tighter assurance, our secondary check runs the same 5-cell capacity sample at 1C to catch cells that grade-slip under higher current demand.
For related pack-level design considerations, the battery pack design guides cover how cell matching tolerance propagates into thermal management and balancing circuit requirements.
FAQ
What’s the minimum information a COA must contain for me to approve a cell lot for parallel configuration?
At minimum: nominal capacity with discharge rate and temperature stated, DCR with measurement method specified, OCV spread across the lot (not just a nominal value), and a single traceable lot number. If any of those four are absent, you’re making an engineering decision without the inputs that decision requires. Some buyers accept a partial COA and compensate with 100% incoming inspection — that works, but it shifts the cost burden entirely to your receiving team.
Can I use cells from two different lots in the same parallel group if both lots pass my incoming inspection?
It depends on the calendar age gap between the lots and the chemistry. Two LFP lots manufactured within the same 60-day window, both passing OCV and DCR criteria independently, can be combined in 2P — the risk is low. If the lots are more than 90 days apart or from different cell manufacturers, I wouldn’t combine them in a parallel group regardless of how well they individually pass spec. The initial self-discharge curves diverge enough to create measurable SOC offset at cycle 200+. Our dataset here covers LFP prismatic only — we don’t yet have enough NMC parallel aging data from mixed lots to make the same call confidently.
Do Chinese suppliers typically provide the cell-level OCV spread data, or do you have to request it separately?
You have to ask, and you should ask before placing the order, not after. Most Shenzhen and Dongguan pack houses don’t include OCV spread in their standard COA format — it’s not a customer requirement often enough to be in their default template. When we request it, roughly half of suppliers provide it from their existing QC records within 24 hours; the other half either don’t measure it systematically or measure it at a different SOC point than what’s useful for matching. Make OCV spread at 50% SOC part of your purchase order quality clause, with the threshold written in — “±5mV maximum within any parallel group” is specific enough to enforce and realistic enough that a competent supplier can meet it without regrading their entire inventory.
Published by compactbess.com Technical Team | Request a sourcing consultation
