Remaining useful life (RUL) is the estimated time, energy throughput, or cycle count a battery has left before it no longer meets the requirements of its application. RUL is related to state of health (SoH), which describes the battery’s current condition relative to a reference or new battery. It is different from state of charge (SoC), which describes how much charge is available at a particular moment.
A battery reaches end of life when it falls below the minimum viable performance (MVP) defined for its use. That limit is application-specific. A battery used for emergency backup may require more reserve than one used in a low-consequence device. Capacity is usually the leading health indicator because capacity loss reduces operating time, but voltage, resistance, temperature, duty cycle, and safety behavior also matter. An 80% capacity threshold is widely used as a convention for some lithium-ion studies and products, but it is not a universal end-of-life rule. Some vehicle starter-battery applications may continue to crank an engine after substantial capacity loss, while still having reduced reserve and higher failure risk.
Most batteries degrade progressively, making trend tracking useful. Abnormal behavior can also result from increased internal resistance, electrical shorts, physical damage, or lithium-ion dendrite growth. Those failure modes may not follow a smooth capacity-fade curve. A practical RUL estimate therefore combines measured or inferred capacity with the application’s performance threshold, operating risk, and the amount of reserve needed for unexpected demand.
Battery Testing Methods for Estimating Remaining Useful Life
Testing methods vary with battery chemistry, construction, operating profile, available instrumentation, and the consequences of failure. The essential requirement is to obtain either direct capacity measurements or credible capacity estimates and track them over time. A single test can describe present condition; a sequence of comparable tests is much more useful for estimating degradation and RUL.
Capacity is the amount of charge a battery can deliver under specified test conditions. As capacity falls, a device may run for less time between charges or a backup system may provide less autonomy. Capacity fade is often gradual, but it is not guaranteed to be linear or predictable throughout life. Temperature, charging voltage, discharge rate, depth of discharge, storage conditions, and cell imbalance can change the apparent trend.
A voltage reading, internal-resistance value, or temperature measurement alone does not directly reveal usable capacity. These measurements can provide valuable diagnostic clues, but their meaning depends on chemistry, temperature, load, rest time, measurement method, and battery age. Some BMSs measure resistance-related parameters directly; others infer them from current and voltage behavior or from a model.
A useful five-method framework is:
Data-driven analysis. Historical operating data can include timestamps, current and voltage, load patterns, charge and discharge intervals, temperature, ambient conditions, cycle count, and other environmental stressors. Algorithms can identify degradation trends and anomalies across many batteries. Machine-learning methods, including neural networks, may estimate SoH or RUL when they are trained and validated on data representative of the battery type and duty cycle. They are not automatically reliable simply because a large dataset is available. Changes in chemistry, thermal management, usage, or measurement quality can reduce accuracy.
Electrochemical impedance spectroscopy. EIS applies a small alternating signal over a range of frequencies and measures the electrical response. The resulting frequency response can be examined with tools such as a Nyquist plot and related equivalent-circuit parameters. These features may correlate with charge-transfer behavior, electrolyte condition, resistance, and other aging mechanisms. EIS can support SoH estimation, sometimes with a data-driven model, but interpretation is sensitive to temperature, state of charge, contact conditions, frequency range, cell variability, and the quality of the reference database. Results from a cell do not automatically transfer to a module or pack, where balancing, interconnections, and cell-to-cell variation add complexity.
BMS-derived data. A conventional BMS can record cycle count, depth of discharge, voltage, current, temperature, and coulomb-counting information. It may also detect anomalies such as unusual voltage divergence or unexpected resistance changes. This information supports health assessment, but RUL estimation is limited when the BMS does not know the battery’s current usable capacity. Coulomb counting accumulates measurement and integration errors, particularly when the battery is rarely brought to a known full or empty reference. A BMS can therefore indicate stress and abnormal behavior without providing a dependable capacity measurement by itself.
SMBus Full Charge Capacity readings. Smart batteries may report Full Charge Capacity (FCC), a digital estimate based largely on coulomb counting during field operation and charging. FCC data is convenient because it can be read without a separate capacity test. Its accuracy depends on the battery’s internal fuel-gauge model, current and voltage measurement, operating history, temperature, and calibration state. Irregular use can prevent the gauge from receiving the reference conditions needed to update its estimate. Calibration or a controlled learning cycle may improve accuracy, but it does not eliminate all uncertainty.
Full-cycle analyzer testing. A battery analyzer can charge and discharge a battery under controlled conditions and measure delivered capacity. This is a more direct method than inferring capacity from ordinary operating data and can be practical for portable batteries, service preparation, acceptance testing, or investigation of suspected degradation. The result is still dependent on the selected current, cutoff limits, temperature, rest periods, and safety procedure. A laboratory-style capacity result should not be treated as identical to capacity under a high-power or low-temperature application.
Direct capacity testing and electrochemical testing measure battery behavior under defined conditions. Operating-data methods infer condition from normal use. The indirect methods are valuable because they can avoid taking equipment out of service, but they require careful calibration, comparable measurements, and a model that reflects the application.
RUL is calculated against a chosen end-of-life threshold, not against an abstract definition of failure. The frequently used 80% capacity point can be a useful comparison value, but the correct threshold may be higher or lower. A communications backup battery, medical device, or safety system may need substantial reserve before its nominal capacity limit is reached. A vehicle starter battery may remain capable of starting an engine after considerable capacity loss, yet its ability to tolerate cold starts, repeated starts, or accessory loads may be reduced. Thresholds should reflect required runtime, peak power, environmental conditions, failure consequences, and the time needed to obtain a replacement.
Reserve capacity is especially important where demand is uncertain. An operator may choose to replace a battery before the measured capacity reaches the minimum needed for normal operation because an unexpected load, cold temperature, delayed maintenance visit, or measurement error could consume the remaining margin. The resulting RUL is a risk-management estimate as much as a degradation estimate.
Cloud Analytics for Battery Capacity and Fleet RUL Forecasting
Connected chargers, battery analyzers, monitors, and rapid testers can accumulate performance records that are difficult to obtain from isolated manual tests. A fleet system may associate each battery with timestamps, location or equipment identity, charge and discharge events, temperature, alarms, test results, and maintenance actions. Over time, this creates a history from which degradation trends and outliers can be compared.
For an SMBus smart battery, a connected charger may read the reported FCC and transmit it with other diagnostic information. The reading is useful but should be interpreted in context. Random or shallow use can cause coulomb-counting drift, and a fuel gauge may require calibration or suitable full and empty reference events before its estimate becomes representative.
Non-smart batteries require a different estimation process. One approach is to estimate the residual SoC at the beginning of a charging interval, then count the charge added until the battery reaches a defined full-charge condition. In simplified form:
estimated usable capacity = residual charge at the start + charge added during charging
For lithium-ion systems, an estimator may combine a battery model, voltage and current measurements, and an Extended Kalman Filter. The filter uses measurements to correct the model’s evolving estimate of SoC and related parameters. It is not a substitute for suitable measurements or calibration. Its performance depends on the model, sensor accuracy, chemistry, temperature range, and charging regime.
The measurement interval matters. A sufficiently long operating or charging interval provides more information than a short topping charge. A topping charge may add too little energy to distinguish residual-charge uncertainty from actual capacity, especially when the starting SoC is not known accurately. A longer runway, controlled charging event, or full operating cycle generally provides a stronger basis for estimating usable capacity, subject to the battery’s chemistry and charging limits.
Cloud systems can apply filtering, quality checks, and anomaly detection before using a record in an RUL model. They may reject data collected during sensor faults, unusual temperatures, interrupted charging, or incomplete cycles. Fleet-level models can also identify whether one battery is degrading faster than similar batteries under comparable conditions. This supports:
- visibility of battery condition across distributed equipment;
- trend analysis by chemistry, model, location, or duty cycle;
- condition-based maintenance rather than replacement only by calendar age;
- earlier investigation of abnormal batteries;
- planned replacement and inventory management; and
- prioritization of equipment where failure would have the greatest consequence.
A diagnostic system can support both smart and conventional batteries if it has adequate measurements and a suitable estimation method. A simple health indicator or status light should not be assumed to represent SoH consistently across different battery types. The meaning of a reported health value depends on the battery’s reference capacity, algorithm, test history, and selected end-of-life definition.
Cloud forecasting also introduces limitations. Connectivity gaps can remove important sections of a battery’s history. Poorly calibrated sensors, inconsistent test conditions, incomplete identifiers, and changes in operating duty can make apparently precise predictions unreliable. Chemistry and pack design affect degradation behavior, so a model trained on one population may not generalize to another. Data governance, privacy, access control, and cybersecurity are additional considerations when operational battery data is transmitted to external systems.
A fleet forecast should therefore report uncertainty and confidence conditions rather than present one date as certain. A useful system identifies the observations supporting the estimate, the applicable threshold, and the conditions that would cause the forecast to change.
Using Battery RUL Estimates for Reliability and Maintenance
Battery monitoring is most valuable when it changes a maintenance decision. Historical capacity and operating data can show whether a battery is degrading normally, approaching an application-specific limit, or behaving abnormally. Connected diagnostics can then support reliability assurance and condition-based maintenance across a fleet rather than relying only on age or an occasional manual test.
Batteries with operational consequences should not simply be installed and ignored. A battery may appear functional during routine operation while having insufficient reserve for a longer outage, a cold start, a high-power event, or an unexpected delay. Periodic testing, BMS records, connected monitoring, or a combination of these methods can reveal that margin before it becomes a service failure.
RUL estimates can support scheduled replacement before the battery reaches the minimum viable performance for its application. They can also support risk management by ranking batteries according to predicted margin, uncertainty, and consequence of failure. A battery with a modest predicted RUL but low operational consequence may be treated differently from one with a similar estimate supporting a safety-critical load.
This approach is consistent with reliability-centered maintenance: maintenance actions are selected according to function, failure consequence, condition, and risk. Battery degradation is not identical to mechanical wear, however. Electrical, electrochemical, thermal, and control-system effects interact, and some abnormal failure modes can develop faster than a historical capacity trend suggests.
Commercial RUL forecasts should therefore be treated as estimates, not guarantees. Their reliability depends on battery chemistry, duty cycle, temperature history, measurement quality, calibration, model validation, data continuity, and the chosen end-of-life threshold. Used with those qualifications, capacity trends and connected diagnostics can make battery fleets more observable and allow replacement or investigation before performance becomes unacceptable.
References
- Battery University | BU-901b: How to Measure the Remaining Useful…. (n.d.). http://www.batteryuniversity.com/article/bu-901b-how-to-measure-the-remaining-useful-life-of-a-battery
- Remaining Useful Life - Battery Design. (n.d.). https://www.batterydesign.net/remaining-useful-life
- Advances in Battery SOH Estimation | PDF | Machine Learning | Electric Vehicle. (n.d.). https://www.scribd.com/document/900093070/Evaluation-of-Advances-in-Battery-Health-Predictio
- Review of State Estimation and Remaining Useful Life Prediction Methods for Lithium–Ion Batteries. (n.d.). https://www.mdpi.com/2071-1050/15/6/5014
- Preprint_Diagnosis and remaining useful life prediction of end-of-life battery systems for second-life applications.pdf. (n.d.). https://publica.fraunhofer.de/bitstreams/6324cfc5-9dbf-4a98-8bb7-42518e503356/download
- SOC estimation for a lithium-ion pouch cell using machine learning under different load profiles. (n.d.). https://pmc.ncbi.nlm.nih.gov/articles/PMC12103532
- Overview of Li Ion Cell State of Health Estimation Methods - Batch 2 | Decibels Lab. (n.d.). https://www.youtube.com/watch?v=iJi1BTXNzGE
- Lithium-Ion Battery State of Health Estimation and Remaining Useful Life Prediction: Practical Techniques for Varied Operating Conditions. (n.d.). https://uwindsor.scholaris.ca/items/376c318b-c1f8-44e7-babc-5fc8b3f02378
- Research on the Remaining Useful Life Prediction Method of Energy Storage Battery Based on Multimodel Integration. (n.d.). https://pmc.ncbi.nlm.nih.gov/articles/PMC11447950
- AI-Driven Prognostics for State of Health Prediction in Li- .... (n.d.). https://arxiv.org/pdf/2504.05728