Cell imbalance

What it is (precise)

Imbalance has two components, and they age differently. State-of-charge imbalance is a divergence in where each cell sits: the cells are offset from one another in how full they are, and a balancing circuit can realign them.

Parameter imbalance is a divergence in what each cell is — spread in capacity and internal resistance that no balancing current can touch, because it is the hardware itself that differs. In a working string both are present at once and coupled: parameter spread continuously regenerates state spread, which is why balancing is a permanent duty rather than a commissioning task.

The neighboring pages own the consequences and the cure. The series-physics law — the same current flows through every cell, so the first cell at a voltage limit terminates the string — is Weakest-cell limitation. The BMS machinery that bleeds or shuttles charge to realign the cells is Cell balancing. This page owns the phenomenon itself: where the divergence comes from, why the dominant stationary chemistry keeps it nearly invisible across most of the operating window, and what it quietly removes from the plant's sellable energy.

Why cells diverge

The spread is present before the first cycle. Cells come off a production line with a distribution in capacity, internal resistance and self-discharge rate; factories grade cells and match them into batches precisely because that distribution exists, and grading narrows it without eliminating it. A rack assembled from one matched bin still contains a fastest-fading cell and a highest-resistance cell — matching decides how far apart the extremes start, not whether there are extremes.

In service, temperature does most of the damage. Calendar aging roughly doubles for every +10 °C of sustained cell temperature, so a few degrees of persistent gradient across a rack — cells near the cooling inlet against cells at the far end, bottom modules against top — means the warm cells fade in capacity and grow in resistance measurably faster than their neighbors, year after year.

Self-discharge is temperature-dependent too, so the same warm cells also drift downward in state of charge faster between cycles. Because thermal gradients recur in the same physical locations every day, the divergence they drive is systematic rather than random: the plant's thermal map slowly writes itself into the cell population.

Resistance spread then feeds back on itself. Every cell in the string carries identical current, so a higher-resistance cell dissipates more I²R heat than its neighbors, runs warmer, and therefore ages faster — growing still more resistance. Self-discharge differences work on the state side of the ledger: each cell leaks charge at its own rate, and the offsets accumulate even while the plant sits idle, which is why a string can come back from a long outage measurably less balanced than it went in. None of these mechanisms is dramatic in isolation; imbalance is their compound interest.

Why it hides — LFP's flat curve

For the LFP chemistry that dominates stationary storage, the open-circuit-voltage curve is nearly flat at about 3.2-3.3 V across the middle of the SOC range. On that plateau, cells several percent apart in state of charge sit millivolts apart in voltage — at or below the ±5 mV accuracy of typical cell-voltage sensing — so in mid-window operation, real imbalance is invisible to the very instrument meant to show it.

The spread surfaces only where the curve steepens, near full and near empty: a string can read as tightly balanced at 50% SOC all week and then fan out sharply in the last few percent of a full charge.

Load adds the opposite distortion. Under current, each cell's terminal voltage is offset by its resistance, so resistance spread inflates the apparent voltage spread while power flows and lets it relax at rest — a loaded snapshot exaggerates imbalance exactly as the resting plateau conceals it. The honest reads are rest voltages taken near the ends of the window, or the BMS's own charge accounting. Watch, too, which cells the string terminates on: the cells that reach the limits first are the ones the spread is costing money through.

What it costs

The bill is written by the series rule. On charge, the highest cell reaches its upper voltage limit while its neighbors still have room, and the string stops with that room unfilled; on discharge, the lowest cell hits the floor while the rest still hold charge the meter will never see.

Energy is stranded at both ends of the window, and the loss scales with the spread: imbalance can hold usable system energy 1-3% below what the average cell state of health would suggest. Across the hundreds of racks of a full plant, that is megawatt-hours of installed capacity — degraded nowhere — that simply cannot be reached.

The cost also compounds contractually and in power. The witnessed capacity test terminates on the first limiting cell, so whatever imbalance the fleet carries on test day is embedded in the contractual baseline that every later warranty comparison uses. And the BMS computes rack charge and discharge power limits from its limiting cells, so a single drifting cell derates its whole rack's power capability before any energy shortfall appears — often the earliest operational symptom of a spread that the mid-window voltage display still says is not there.

How it shows up in specs, operations and contracts

A datasheet usable-energy figure assumes a balanced string at stated temperature and C-rate; nothing on the sheet tells you what happens as the population spreads.

The working telemetry is the max-min cell voltage delta near the window ends, per-cell rest voltages, and the balancing duty each module is drawing — a balancing system working harder every month is measuring a divergence rate, and rising duty concentrated on the same modules points at a thermal gradient or an outlier cell rather than ordinary drift. Before a capacity test, a rest-and-balance period is standard precisely so the measurement captures the battery rather than its spread.

Vendor questions worth putting in writing: what cell-to-cell spread the usable-energy guarantee assumes, what intra-rack temperature gradient the thermal design holds, and whether per-cell and limiting-cell telemetry is exposed to the owner rather than summarized into a fleet average. Averages are the standing trap — a healthy plant-level SOC can hide one rack whose spread is about to end the discharge. And since every divergence mechanism runs on temperature, an HVAC economy setting that widens the internal gradient is quietly an imbalance decision too.