Lithium-Ion Battery Aging: Connecting Test Signals to Degradation Mechanisms

Two lithium-ion cells can lose the same amount of capacity for very different reasons. One may have consumed cyclable lithium in side reactions. Another may have isolated part of an electrode through cracking or loss of electrical contact. A third may still hold much of its lithium inventory but deliver less energy at the test rate because resistance has increased.

Capacity fade is a symptom, not a mechanism. A useful aging study asks what changed inside the cell, which operating stress caused it, and which measurements can distinguish that explanation from the alternatives.

This is why battery aging cannot be reduced to one curve or one universal lifetime model. Chemistry, electrode design, cell format, temperature, state of charge, current, pressure, and manufacturing history interact. The same cell can also move from one dominant degradation path to another as it ages.

Start with the cell's stress history

Before interpreting diagnostic data, reconstruct what the cell experienced. Calendar aging develops while the cell is stored, with time, temperature, and state of charge strongly influencing the reactions. Cycle aging adds repeated lithium transport, volume change, heat generation, and current-dependent polarization. In real use, the two occur together.

Aging data are difficult to compare when the protocol is incomplete. Record charge and discharge rates, voltage limits, constant-voltage hold criteria, depth of discharge, rest periods, temperature, applied pressure where relevant, and the frequency of reference performance tests. A capacity value measured at a high rate is not directly comparable with one measured slowly, because polarization changes how much of the stored charge is accessible before the voltage limit is reached.

For laboratory cells, fabrication history belongs in the same record. Slurry dispersion, coating loading, porosity, drying, electrolyte amount, separator choice, and stack pressure can all create apparent aging differences. Our guides to battery slurry preparation and repeatable coin-cell assembly cover two common sources of cell-to-cell variation.

Three ledgers describe most observable degradation

Researchers often organize degradation into loss of lithium inventory, loss of active material, and resistance or transport changes. These are modes used to account for performance loss, not single chemical reactions. Several mechanisms can contribute to each mode, and one mechanism can affect more than one ledger.

Loss of lithium inventory (LLI)

LLI means that less cyclable lithium is available to move between the electrodes. Growth and repair of interphase layers can consume lithium and electrolyte. Lithium plating can also remove lithium from normal intercalation, especially when charging conditions create high anode polarization. Portions of plated lithium may be recovered, while other portions become electrically isolated or react further.

SEI growth is important in many graphite-based cells, but it should not be declared the dominant cause without evidence. The balance depends on electrode chemistry, temperature, voltage window, formation history, and operating conditions.

Loss of active material (LAM)

LAM occurs when part of an electrode can no longer participate effectively. Particles may crack, lose contact with the conductive network, undergo unfavorable structural change, or become covered by resistive surface products. Silicon-rich anodes, layered oxide cathodes, graphite, and lithium metal do not age in the same way, so the physical origin of LAM must be tied to the materials actually used.

Mechanical damage does not always produce immediate capacity loss. Fresh surfaces can trigger additional side reactions, and contact may deteriorate gradually. This coupling is one reason a single diagnostic signal rarely identifies a unique mechanism.

Resistance and transport changes

Ohmic resistance, interfacial charge-transfer behavior, ionic transport through porous electrodes, and solid-state diffusion can all change with age. The result may appear as greater voltage polarization, lower power, more heat, or reduced usable capacity at higher rates.

Resistance growth and capacity loss are related but not interchangeable. A cell can retain substantial low-rate capacity while failing a power requirement. Conversely, lithium inventory can decline before a large direct-current resistance change is obvious.

Read aging signals in layers

A practical diagnosis builds from measurements that are easy to repeat toward measurements that are more specific or invasive. Agreement among independent signals is more persuasive than a complicated interpretation of one dataset.

Capacity, energy, and coulombic efficiency

Reference performance tests establish how much charge and energy the cell delivers under controlled conditions. Use the same temperature, rest time, rate, and voltage limits at each checkpoint. Tracking both charge and discharge energy can reveal growing polarization that a capacity-retention plot alone may hide.

High-precision coulombic efficiency can expose small parasitic losses, but the differences are often subtle. Temperature stability, current accuracy, channel calibration, and sufficient repeat cells matter. A changing efficiency does not by itself identify which side reaction occurred.

Incremental capacity and differential voltage

Incremental capacity analysis (ICA, dQ/dV) and differential voltage analysis (DVA, dV/dQ) transform voltage-capacity data so that shifts in electrochemical features become easier to see. Peak movement, area change, and altered spacing can help separate changes in electrode balance from loss of active material.

These methods are powerful but sensitive to protocol and processing. Current, temperature, voltage resolution, smoothing, and cell relaxation affect the result. Full-cell peaks also combine contributions from both electrodes. Strong interpretation usually relies on appropriate half-cell reference data or a validated electrode model rather than assigning every peak by visual inspection.

Electrochemical impedance spectroscopy

EIS probes the cell response across frequency, giving access to processes with different characteristic times. It can reveal systematic changes in ohmic, interfacial, and transport-related behavior and can provide rich inputs for statistical models.

An impedance spectrum is not a fingerprint with one automatic answer. Equivalent-circuit fits may be non-unique, and features can overlap. Temperature and state of charge can shift the spectrum even without permanent degradation. Report the measurement conditions, check fit uncertainty, and compare trends across controlled reference states.

A Nature Communications study showed how machine learning could extract health information from more than 20,000 EIS spectra. The authors also noted important limits: their testing cells used the same charge-discharge rate as the training cells, and broader operating-rate validation was still needed. That boundary is as important as the headline result.

When electrochemical clues are not enough

Post-mortem work can test hypotheses formed from cycling, ICA/DVA, and EIS. Depending on the question, researchers may use microscopy, X-ray diffraction, spectroscopy, elemental analysis, gas analysis, or three-electrode measurements. The best technique is the one that can disprove the suspected mechanism, not simply produce a detailed image.

Opening a cell can alter the evidence. Air exposure, rinsing solvent, drying, sampling location, and delay before analysis may change surface species. Plan the disassembly before the aging test ends, preserve matched pristine controls, and document exactly where each sample came from.

Safety note: destructive analysis of lithium-ion cells can expose reactive electrodes and flammable electrolyte. Fully assess the cell condition and use an approved inert-atmosphere or controlled-environment procedure, appropriate personal protective equipment, and the laboratory's waste and emergency protocols.

Prediction is not the same as diagnosis

Machine-learning models can estimate state of health or remaining useful life from early cycling curves, impedance, temperature, and operating history. Early-cycle prediction research has shown that informative signals exist before obvious capacity fade. More recent models attempt to transfer learning across varied temperatures, protocols, formats, and chemistries.

No error percentage applies universally. Performance depends on the training population, the end-of-life definition, the future duty cycle, and how far the target cell lies outside the training distribution. A model can predict an end point accurately without identifying the physical cause of aging; it can also identify a useful correlation that fails after a chemistry or protocol change.

For research use, report the train-test split by cell rather than by individual cycle, preserve truly unseen validation cells, quantify uncertainty, and test performance after deliberate changes in chemistry or operating condition. Physics-informed models can improve consistency, but physical constraints do not compensate for missing degradation mechanisms or biased data.

A practical blueprint for an aging study

1. Define the decision. State whether the study needs mechanism identification, lifetime comparison, a power limit, a safety warning, or an RUL model.
2. Build matched cells and controls. Use enough replicates to distinguish material effects from assembly scatter.
3. Separate stress and measurement cycles. Apply the aging condition consistently, then use standardized reference tests for comparison.
4. Collect complementary signals. Combine capacity and energy with selected voltage analysis, resistance, EIS, temperature, or other diagnostics justified by the hypothesis.
5. Set post-mortem criteria in advance. Choose sampling points and control materials before seeing the final curves.
6. Report uncertainty and exceptions. Outliers, failed cells, fixture problems, and protocol deviations are part of the evidence.

Battery aging becomes more understandable when the study keeps mechanism, measurement, and prediction separate. Capacity tells you that performance changed. Diagnostics narrow the possible causes. Controlled post-mortem work can test those causes. Prediction models then have a cleaner, more traceable dataset from which to learn.

References

• Lithium ion battery degradation: what you need to know, Physical Chemistry Chemical Physics (2021).
• Differential Analysis of Galvanostatic Cycle Data from Li-Ion Batteries, Chemistry of Materials (2022).
• Identifying degradation patterns of lithium ion batteries from impedance spectroscopy using machine learning, Nature Communications (2020).
• Data-driven prediction of battery cycle life before capacity degradation, Nature Energy (2019).
• Battery lifetime prediction across diverse ageing conditions with inter-cell deep learning, Nature Machine Intelligence (2024).