There is no state of charge sensor. Nothing inside a cell reports how full it is. Every percentage figure a driver, an operator or a phone user has ever seen was produced by a model taking voltage, current and temperature as inputs and guessing. Understanding batteries at a professional level starts with taking that seriously.
The two definitions, stated properly
State of charge is the charge currently extractable from a cell, divided by the charge the cell can hold in its present condition, expressed as a percentage. The denominator is present capacity, not nameplate capacity. A cell that has faded to 80 per cent of its original capacity still reaches 100 per cent state of charge when it is full.
State of health is the comparison between present condition and beginning-of-life condition. It comes in two forms that are not interchangeable:
Capacity-based state of health is present usable capacity divided by rated or beginning-of-life capacity. This is what warranties are usually written against.
Resistance-based state of health tracks the rise in internal resistance, often measured as direct current internal resistance at a defined pulse and temperature. This is what determines whether the pack can still deliver its rated power.
A cell can hold 88 per cent of its capacity while its resistance has risen by 60 per cent. In a power-limited application, that cell has failed. In an energy-limited application, it has years left. Quoting a single state of health number without saying which definition it uses is one of the most common errors in the field.
How state of charge is estimated
A BMS estimates state of charge by combining methods: coulomb counting carries the estimate minute to minute, open-circuit voltage lookup corrects it after rests, and model-based filters such as the extended Kalman filter fuse both continuously. No method measures it directly.
| Method | How it works | Accuracy in practice | Main limitation |
|---|---|---|---|
| Coulomb counting | Integrates measured current over time | High over short periods | Drifts without correction, because current sensor error accumulates and there is no absolute reference |
| Open-circuit voltage lookup | Maps a rested voltage to state of charge using a stored curve | Good on sloped chemistries | Requires relaxation time, and fails where the voltage curve is flat |
| Model-based with a filter | Equivalent-circuit model plus an extended Kalman filter or similar, fusing current and voltage | Best general-purpose accuracy | Needs parameterisation across temperature and age, and computational headroom |
| Impedance-based | Uses the frequency response of the cell | Useful diagnostically | Needs excitation hardware, mostly a laboratory and increasingly an onboard diagnostic method |
| Data-driven | Trained on fleet or laboratory data | Strong within the training distribution | Degrades outside it, and is hard to certify |
Production systems combine methods. Coulomb counting carries the estimate between corrections, and an open-circuit voltage reading after a long rest resets it. The interesting engineering question is always what happens between corrections.
Why LFP is the hard case
The value of an open-circuit voltage lookup depends on the slope of the voltage curve. A steep curve means a small voltage error translates to a small state of charge error. LFP's curve is close to flat between roughly 20 and 90 per cent state of charge, so a measurement error of a few millivolts can correspond to tens of percentage points of state of charge.
Consequences worth knowing before you specify an LFP system:
The estimate relies heavily on coulomb counting, so current sensor quality and calibration matter more than they would in an NMC system.
Drift accumulates. A system that never reaches a full or empty endpoint has no opportunity to reset, so LFP home batteries and grid systems are usually programmed to complete a full charge periodically for calibration rather than for energy reasons.
Cell-to-cell divergence is harder to detect early, because the voltage differences that would reveal it are small in the flat region.
This is a case where the safety and cycle-life advantages of LFP come with a control problem attached, and it is the sort of trade-off that datasheets do not show.
Gross capacity, net capacity and what the display shows
Manufacturers reserve buffer at the top and bottom of the cell range. The top buffer protects calendar life by keeping cells away from their highest potential. The bottom buffer prevents over-discharge and gives the vehicle or system reserve.
So there are three different numbers in circulation for any pack: gross or nameplate capacity at cell level, net usable capacity after buffers, and the displayed percentage which maps the net window to a 0 to 100 scale for the user. A pack advertised at 82 kWh gross with 77 kWh usable will show 100 per cent at a cell state of charge somewhere below full. When comparing vehicles or storage products, ask which number you are being quoted.
Measuring state of health without taking the pack apart
The difficulty is that a clean capacity measurement needs a wide, slow, temperature-controlled cycle, and real assets rarely provide one. Several approaches work in the field:
Incremental capacity analysis differentiates capacity with respect to voltage during a slow cycle. Peaks in the resulting curve correspond to phase transitions in the electrode materials, and their movement and shrinkage indicate which degradation mode is running. Differential voltage analysis does the reverse and is often more useful on flat-curve chemistries.
Partial-cycle capacity estimation fits a model to segments of real operating data rather than requiring a full cycle.
Resistance tracking through routine pulses, which most systems already produce during normal operation, gives a continuous power-capability signal.
The diagnostic value of these methods is that they distinguish loss of lithium inventory from loss of active material at each electrode. That distinction tells you whether the pack is ageing in a way that will continue linearly or in a way that is about to accelerate.
The reporting requirement changes the stakes
State of health used to be an internal variable that manufacturers could define as they wished. Regulation (EU) 2023/1542 changes that in two steps. Since 18 August 2024, Article 14 has required the battery management system of stationary storage, light means of transport and electric vehicle batteries to hold readable data on state of health and expected lifetime. From 18 February 2027, the battery passport applies to electric vehicle, light means of transport and industrial batteries above 2 kWh, with state of health and durability data among the Annex XIII content. As of mid-2026 that date stands, although the delegated and implementing acts specifying the passport's technical details are still being finalised. That has two effects worth planning for.
Definitions become contestable. If two manufacturers report state of health on different bases, buyers, insurers and second-life purchasers will notice, and the comparison will be made whether or not it is fair.
Second-life valuation gets a data foundation. A repurposer who can read a documented degradation history rather than testing every module changes the economics of second life considerably.
If you are specifying a system today with a service life running past 2027, ask the supplier how they define state of health, at what temperature and rate they measure it, and what they will be able to report.
Informational and educational content only. Not professional, financial, legal, or engineering advice.