Test data is only as good as the conditions attached to it. A capacity figure without a rate, a cycle life figure without a depth of discharge, or a resistance figure without a pulse duration is decoration. Most of the discipline in battery testing lies in defining and controlling conditions rather than in the measurement itself.
Four purposes
Performance characterisation establishes what a cell or system does when new: capacity at various rates, energy, power capability, efficiency, self-discharge, and how all of these change with temperature.
Life testing establishes how those properties change with time and use, through cycle ageing and calendar ageing programmes.
Diagnostics determines why a cell has changed, separating loss of lithium inventory from loss of active material from resistance growth.
Abuse and safety testing determines what happens when the cell or system is pushed outside its design envelope, deliberately.
Each needs different equipment, different timescales and different sample sizes, and confusing their outputs is a common source of bad decisions.
Performance testing
Capacity is measured by charging under a defined protocol, resting, then discharging at a specified current and temperature to a cut-off voltage. The delivered charge is integrated. Because delivered capacity falls as rate rises, a capacity test is meaningless without its current, and a cell rated at a low rate will disappoint anyone who assumed the figure applied at higher rates.
Power capability is characterised separately, most commonly through hybrid pulse power characterisation. Pulses of defined magnitude and duration are applied in both directions at a series of states of charge, and the voltage response gives resistance and therefore the current the cell can pass before hitting a voltage limit. The output is a map across the state of charge and temperature space, which is what a system designer actually needs.
Efficiency, both coulombic and energy, is measured over full cycles. Coulombic efficiency in particular is a sensitive early indicator of side reactions, but measuring it usefully requires equipment accuracy well beyond a general purpose cycler.
Self-discharge is measured as open circuit voltage decay over an extended rest at controlled temperature. It is the screen that catches internal defects, and it is slow by nature, which is why the ageing area in a cell factory is as large as it is.
Life testing, and the trap inside it
Cycle life testing runs a defined profile repeatedly, with periodic reference performance tests at a fixed low rate and temperature to track capacity and resistance on a consistent basis. Without those reference tests, drift in ambient conditions contaminates the ageing signal.
Calendar ageing testing stores cells at combinations of temperature and state of charge, with periodic checks. A useful matrix needs several temperatures and several states of charge, which multiplies the sample count quickly.
Both take a long time. A protocol running to 4,000 cycles at a realistic rate occupies a channel for many months, and a calendar study intended to say something about a fifteen-year asset cannot be run in real time at all. Acceleration is therefore unavoidable, and it carries a specific hazard.
Accelerating with temperature assumes the mechanisms that dominate at elevated temperature are the same ones that will dominate in service. Waldmann and colleagues showed the dominant mechanism changes around 25 degrees Celsius, from plating-driven below to film-growth-driven above, so an accelerated test at elevated temperature can genuinely mispredict a cold climate application.
The second hazard is the knee. Fade is often close to linear for a long period before accelerating. A test that stops before the knee produces a clean linear fit, a tight confidence interval and an optimistic answer. Confidence in the fit says nothing about whether the extrapolation region contains a knee. Any lifetime claim built on extrapolated linear fade should state how far the test actually ran.
Sample size, which is where most programmes are weakest
Nominally identical cells from the same production lot differ measurably in initial capacity, resistance and, importantly, in ageing trajectory. Baumhöfer and colleagues documented this, and later work by Dechent and colleagues addressed the sample sizes needed to characterise it.
The practical consequences:
A single cell tells you about one cell. Comparing two cells and drawing a conclusion about two designs is not a valid comparison.
Sample size should be planned against the effect size you need to detect and the variance you expect, before the test starts.
Report the spread, not only the mean. A design with the same mean fade and twice the variance is a worse design, because warranty exposure sits in the tail rather than at the average.
Diagnostics
Electrochemical impedance spectroscopy separates ohmic, interfacial and diffusion contributions to resistance, localising where degradation is occurring.
Incremental capacity analysis and differential voltage analysis, derived from slow reference cycles, distinguish lithium inventory loss from active material loss at each electrode. Differential voltage analysis is generally more workable on flat-curve chemistries.
Isothermal calorimetry measures heat generation during normal operation, which feeds thermal system design. Accelerating rate calorimetry measures self-heating onset and heat release under adiabatic conditions, which feeds safety design.
Computed tomography and X-ray imaging inspect internal structure without disassembly, catching electrode misalignment, deformation and gas pockets.
Post-mortem analysis remains the reference method. Disassembly in an inert atmosphere followed by microscopy, diffraction and elemental analysis confirms what the non-invasive methods inferred, and it is how those methods were validated.
Abuse and safety testing
| Test | What it simulates | Typical level |
|---|---|---|
| Nail penetration or forced internal short | Internal short circuit from a defect | Cell |
| Crush and mechanical shock | Collision damage | Cell, module, pack |
| Overcharge and over-discharge | Control or charger failure | Cell, module |
| External short circuit | Wiring fault | Cell, module, pack |
| Thermal abuse, oven or hotbox | External fire exposure | Cell, module |
| Propagation testing | One cell failing among many | Module, unit, installation |
| Altitude, thermal cycling, vibration, shock | Transport environment, per UN 38.3 tests T.1 to T.4 | Cell, battery |
The distinction that matters commercially: cell-level abuse testing characterises initiation, while propagation testing characterises what the system does about it. A cell that passes nail penetration tells you very little about whether a container full of those cells will survive one of them failing. UL 9540A exists specifically to answer the second question for stationary storage, and the full test report at unit level is the document worth reading rather than the certificate. Regulators are moving the same way for vehicles: China's GB 38031-2025, applying to new type approvals from 1 July 2026, requires no fire and no explosion at pack level after a thermal runaway trigger, replacing the earlier 5-minute warning requirement.
Reporting, briefly
Every result should carry its conditions: temperature and its control tolerance, current or C-rate, state of charge window, cut-off criteria, rest periods, cell age and history, sample size, and the measurement boundary if the test is at system level. This is not bureaucracy. It is the difference between data another engineer can use and a number that has to be regenerated.
Informational and educational content only. Not professional, financial, legal, or engineering advice.