How to choose a cell sorting machine: OCV and IR grading for large ESS cells

What does the sorter do on an ESS line?

The sorter is the module line’s first quality gate. After robots depalletize the incoming cells, the sorter scans and tests each cell, rejects bad ones, and bins good cells by voltage, resistance and other values so the stacking station can build modules from matched cells.

It solves three problems:

• Incoming quality: cells with abnormal voltage, high resistance or damage from transport and storage never reach a module.
• Consistency: the closer the cells in a module, the less any single cell is over-charged or over-discharged, and the longer the module lasts.
• Traceability: every cell’s results are tied to its barcode, so a module or PACK problem can be traced back to the cell.

What does it measure?

Test	Purpose	Typical requirement	Fit
Barcode reading	Identify cells, bind data	Read rate ≥ 99.9 % (excluding defective labels)	Standard
Open-circuit voltage (OCV)	State of charge, abnormal cells	Voltage accuracy at the ±0.01 % of reading level	Standard
AC internal resistance (1 kHz)	Internal connection quality, consistency	3 mΩ range, 0.1 µΩ resolution	Standard
K-value (self-discharge)	Catch cells with high self-discharge such as micro-shorts	Two voltage readings 7–14 days apart	Optional, or from the cell maker
Thickness / dimensions	Keep stacked module dimensions in tolerance	Thickness at a defined pressure	Optional
Can insulation / withstand	Insulation between terminals and can	Per cell datasheet	Optional
Visual inspection	Scratches, dents, leakage, dirty terminals	Per appearance standard	Optional
Temperature	Temperature compensation of voltage	About ± 0.5 °C	Recommended

Reading accuracy specs: accuracy, resolution, repeatability

Large storage cells have very low resistance. A 314 Ah cell measures about 0.18 mΩ (180 µΩ) at delivery, and cells in one batch differ by only a few to a dozen microohms. So the first question is whether the instrument can resolve differences that small.

Spec	Good enough	Warning sign
IR range and resolution	3 mΩ range, 0.1 µΩ resolution	“0.1 mΩ resolution” only — over half the reading, useless for matching
IR accuracy	±(0.5 % rdg + 10 dgt): about ± 1.9 µΩ (about 1.1 %) at 180 µΩ	A percentage with no digits, or no range stated
Voltage accuracy	±(0.01 % rdg + 3 dgt): about ± 0.36 mV at 3.3 V	±0.1 % of reading (about ± 3.3 mV at 3.3 V) — coarse screening only
Repeatability	Repeated readings on one cell within about ± 0.1 mV and a few microohms	No repeatability data, or IR repeatability only stated to 0.01 mΩ

Two more points are easy to miss:

• Speed costs accuracy. On its fastest sampling setting, the same instrument’s IR error can grow two- to threefold. Takt and accuracy claims must refer to the same setting.
• Temperature moves voltage. A 1 °C change shifts OCV by several hundred microvolts. Without compensation, morning and evening, or winter and summer, readings cannot be compared.

On the 8 PPM line on this site, the sorting station’s voltage accuracy is ±(0.01 % rdg + 3 dgt), IR accuracy ±(0.5 % rdg + 10 dgt), and barcode read rate ≥ 99.9 %.

Bins and matching: setting the rules

National standards do not set cell-matching windows. GB/T 36276 sets requirements at batch and system level, for example that the initial-energy spread of a batch of cells stays within 4 % of the mean. Matching windows are usually agreed between the cell maker, the PACK maker and the end customer. Common practice:

Parameter	Common window	Notes
OCV spread	≤ 5 mV, ≤ 3 mV when strict	Maximum minus minimum within one module
AC IR spread	≤ 10 % of the mean	Some projects use (max − min) ÷ mean ≤ 10 %
Capacity spread	≤ 1 % of nominal	About 2.8 Ah for a 280 Ah cell; usually taken from the cell maker’s grading data
K-value	LFP commonly ≤ 2–3 mV per day	With rests under 48 hours, readings are mostly relaxation noise

Match the number of bins to how the line builds modules: sorters typically have 5–10 physical lanes (including NG), and software can grade more finely. Finer bins match better but need more buffer space and longer waits to complete a module.

K-value is calculated as (first OCV − second OCV) ÷ days between readings, in mV per day. It is the best indicator of micro-shorts but needs rest time and storage space, which is why many lines rely on the cell maker’s K-value data.

Takt, bins and changeover

• Leave takt headroom. The sorter sits at the head of the line; if it is slow, everything waits. Run it at least 20 % faster than the line, for example 10–12 PPM on an 8 PPM line.
• Large cells need new grippers. A 314 Ah cell weighs about 5.6 kg, a 587 Ah cell 10–10.6 kg. For large-cell compatibility, check handling payload, gripper size and test-fixture positions. See 500 Ah-plus cells: what changes.
• Probes wear. Spring probes need replacing about monthly at high test rates; confirm how they are changed (in minutes) and what spares cost.
• Put changeover in the contract. Agree target times for swapping grippers, locating fixtures and test programs when the cell format changes.

Data and traceability

• Each cell’s barcode, results, bin, test time and machine ID should go to MES in real time.
• Limits should be sent from MES, so every machine judges by the same values.
• Rejected cells should be logged by defect type and kept separately, for reconciliation with the cell maker.
• Run statistical process control (SPC) on the data: drifting voltage or resistance distributions flag problems before reject rates rise.

Buyer’s checklist

1. Test instrument model, range, resolution and accuracy (stating the sampling setting).
2. Committed repeatability and gauge R&R.
3. Takt (in cells) and number of bins against line takt plus headroom.
4. Cell formats covered (size, weight), changeover method and time.
5. Temperature compensation; whether the test area must be temperature-controlled.
6. Options: K-value, thickness, insulation/withstand, vision.
7. MES interface and data format.
8. Wear-part list, life and price; calibration interval and method.

Acceptance: proving it measures right

• Gauge R&R. Take about 10 cells spanning the parameter range and measure them repeatedly with different operators at different times. By common rules, GR&R ≤ 10 % is acceptable, 10–30 % conditionally depending on use, and over 30 % unacceptable; aim for at least 5 distinct categories.
• Reference-sample correlation. Measure a set of reference samples on the sorter, a lab instrument and the cell maker’s equipment to establish offsets.
• Continuous run. Run real cells continuously and measure takt, read rate and misjudgement rate (good cells rejected, bad cells passed).
• Traceability spot checks. Pick cells at random and compare MES data with a retest on the spot.

What we can supply

The 8 PPM line on this site has a cell scanning, testing and sorting station working with robotic depalletizing, NG rejection and refill. As a project needs, we can also supply higher-precision test instruments, K-value rest-and-retest solutions, thickness and visual inspection, or a stand-alone sorting machine through our supply-chain service.