Battery test systems provide the experimental foundation that makes EKF and UKF battery state estimation reliable. They generate controlled current, voltage, temperature, and rest-period data under dynamic operating conditions, while parameter-identification workflows convert those measurements into an equivalent-circuit model suitable for state estimation. The EKF then linearizes that model locally, whereas the UKF propagates representative sigma points through the nonlinear model without explicitly calculating Jacobians.
Core takeaway: Filter performance depends heavily on model quality and measurement quality. Battery test systems create the excitation data, and parameter identification converts that data into the electrical model, initial states, noise assumptions, and validation evidence required by EKF and UKF algorithms.
Why Battery Testing Is the Starting Point
Dynamic profiles reveal battery behavior
Static charge and discharge tests are insufficient for estimating how a battery behaves during real operation. Battery test systems apply dynamic complex-pulse profiles that reproduce changing load conditions, such as acceleration, braking, idling, and regenerative charging.
These profiles produce voltage and current responses that expose the battery’s ohmic resistance, polarization behavior, transient dynamics, and relaxation characteristics.
Accurate measurements determine estimator quality
An EKF or UKF can only be as reliable as the data entering it. High-precision test systems measure current, terminal voltage, temperature, and timing with sufficient accuracy to distinguish genuine battery dynamics from measurement noise.
This is especially important during high-rate pulses, where rapid current changes can cause abrupt voltage fluctuations and make state tracking more difficult.
Multiple operating conditions improve robustness
Battery parameters vary with SOC, temperature, current rate, aging, and operating direction. Testing across these conditions helps identify whether a single fixed model is adequate or whether the estimator requires parameter maps or adaptive updates.
For BMS development, this reduces the risk of validating an algorithm only under laboratory conditions that do not represent field operation.
How Parameter Identification Builds the Filter Model
Equivalent-circuit models translate measurements into states
A common workflow fits measured current-voltage behavior to an equivalent circuit model, often containing an open-circuit-voltage relationship and one or more resistance-capacitance branches.
The model represents internal quantities such as SOC and polarization voltage as state variables. These states are not directly measurable during normal operation, so the filter estimates them from the applied current and measured terminal voltage.
Pulse responses identify electrical parameters
A dynamic pulse causes an immediate voltage change followed by slower transient behavior. The immediate response helps identify ohmic resistance, while the slower response supports identification of polarization resistances and time constants.
Repeated tests across SOC and temperature conditions can produce parameter sets or lookup tables that better reflect the battery’s nonlinear behavior.
Taylor expansion supports EKF formulation
For an EKF, the nonlinear battery model is locally approximated around the current state estimate using a Taylor-series expansion. This produces the state-transition and observation Jacobian matrices needed by the filter.
The resulting model allows the EKF to predict the next SOC and internal dynamic states, compare the predicted voltage with the measured voltage, and correct the estimate using the Kalman gain.
The model must include suitable initial conditions
Parameter identification is not limited to resistance and capacitance values. The workflow also helps establish:
- Initial SOC and polarization states
- Open-circuit-voltage versus SOC behavior
- Process-noise assumptions
- Measurement-noise assumptions
- Sampling intervals and model discretization
Poor choices in these areas can produce instability or persistent estimation bias even when the filter equations are implemented correctly.
How the Identified Model Supports the EKF
Prediction uses the battery’s dynamic equations
At each sampling step, the EKF uses the identified state-space model and measured current to predict the battery’s next internal state.
For example, it may predict SOC evolution from current integration while simultaneously predicting the voltage contribution from polarization states.
Measurement feedback corrects prediction error
The filter calculates a predicted terminal voltage and compares it with the measured voltage. The difference, commonly called the innovation or residual, is used with the observation matrix and Kalman gain to correct the predicted state.
This feedback limits the long-term drift associated with pure coulomb counting.
Jacobians connect state errors to voltage errors
The EKF requires derivatives that describe how changes in the internal states affect the predicted voltage and state transition. These derivatives are obtained from the identified battery model.
If the model is strongly nonlinear or poorly parameterized, the local linear approximation may become inaccurate, reducing estimation quality.
Test data validates convergence
The same test system used to identify the model can compare estimated SOC against a reference generated from controlled laboratory measurements. Validation should include dynamic pulses, rest periods, different SOC ranges, and relevant temperatures.
This reveals tracking delays, oscillations, bias, and sensitivity to current or voltage noise before the algorithm is deployed in a BMS.
How the Identified Model Supports the UKF
UKF avoids explicit Jacobian calculations
The UKF uses an unscented transformation rather than directly linearizing the model with Jacobians. It selects a set of sigma points around the current state estimate and propagates each point through the nonlinear battery equations.
The transformed points are then recombined to estimate the predicted state mean and covariance.
Nonlinear behavior is represented more directly
Because each sigma point passes through the nonlinear model, the UKF can capture state-distribution changes that a local EKF linearization may miss.
This is useful when voltage-SOC relationships, polarization dynamics, or parameter dependencies create substantial nonlinearity.
UKF still depends on model and data quality
Avoiding Jacobians does not eliminate the need for accurate parameter identification. The UKF still requires a credible state-transition model, observation model, parameter set, initial covariance, and noise characterization.
A sophisticated filter cannot compensate for incorrect resistance values, inaccurate voltage-SOC data, or test measurements contaminated by systematic error.
Testing determines whether UKF complexity is justified
Battery test data can be used to compare EKF and UKF performance under the same operating profiles. If nonlinear effects are modest, the EKF may deliver sufficient accuracy with lower computational demand.
If the battery exhibits strong nonlinear behavior or the EKF shows significant linearization error, the UKF may justify its greater computational cost.
How the Complete Workflow Fits Together
Step 1: Design an informative test profile
The test profile should contain enough current variation to excite the battery dynamics. Compound pulse sequences are useful because they expose both fast voltage responses and slower relaxation behavior.
High-rate testing is valuable for stressing the estimator, but it should be performed within safe cell operating limits.
Step 2: Capture synchronized electrical and thermal data
The battery test system records current, voltage, temperature, and time under controlled conditions. Synchronization matters because even small timing mismatches can distort the apparent dynamic response used for identification.
Testing across environmental and operating conditions supports a model that is more representative of actual BMS use.
Step 3: Identify and validate model parameters
Researchers fit the measured response to the selected equivalent-circuit model and verify the fit against data that was not used for parameter estimation.
This separation helps determine whether the model captures general battery behavior rather than merely reproducing one specific pulse sequence.
Step 4: Implement the estimator
The identified model is discretized and embedded in the EKF or UKF. The algorithm is configured with initial state estimates, covariance matrices, process noise, and voltage-measurement noise.
For EKF implementations, the required Jacobians must be derived or generated consistently with the model. For UKF implementations, sigma-point scaling and covariance handling must be configured appropriately.
Step 5: Compare estimates against reference behavior
The estimator is evaluated using measured voltage and current from dynamic profiles. Researchers examine SOC error, convergence speed, transient tracking, stability, and behavior during rest.
This process connects algorithm performance to specific physical conditions rather than relying only on a nominal simulation.
Improving EKF Behavior During Aggressive Pulses
Why conventional EKF performance can degrade
During high-rate charge-discharge pulses, sudden voltage changes can cause SOC tracking delays and estimate oscillations. The EKF’s local linearization and fixed gain behavior may not respond adequately to rapidly changing conditions.
The problem can be amplified by polarization effects, inaccurate model parameters, or mismatched noise assumptions.
Adaptive gain can accelerate correction
One reported approach scales the conventional Kalman gain using an adaptive factor:
[ K'_k = \lambda K_k ]
A larger gain during active charge-discharge periods makes the estimator respond more strongly to measurement residuals. During rest periods, returning the factor to 1 can reduce overcorrection caused by polarization-related voltage behavior.
Adaptive gain is a design choice, not a universal rule
The supplementary test results describe gain factors between 40 and 60 for particular high-rate pulse experiments, with SOC error reduced from above 15% to within 3.5%. These values should not be treated as universal settings.
The appropriate gain depends on the cell, model, sampling rate, pulse profile, sensor quality, and noise assumptions. It must be tuned and validated experimentally.
Understanding the Trade-offs
EKF: efficient but locally approximate
The EKF is generally attractive for embedded BMS applications because it is computationally efficient and fits naturally into a recursive state-estimation process.
Its limitation is the reliance on local linearization. When the battery model is strongly nonlinear or the operating point changes rapidly, the approximation can introduce estimation error.
UKF: better nonlinear handling but more computation
The UKF avoids explicit Jacobian derivation and can provide more accurate estimates under strong nonlinear conditions.
Its trade-off is increased computational and implementation complexity because multiple sigma points must be propagated through the battery model at every update.
High-rate testing improves realism but increases difficulty
Aggressive pulse profiles reveal weaknesses that gentle tests may hide. However, they also increase sensitivity to sensor bandwidth, thermal changes, voltage transients, parameter variation, and model mismatch.
Testing must therefore balance realistic stress conditions with safe operation and adequate measurement quality.
Overfitting can produce misleading confidence
A model that fits one pulse profile extremely well may perform poorly under a different temperature, SOC range, aging condition, or current rate.
Validation against independent profiles is essential before concluding that an EKF or UKF is robust.
Filtering cannot replace sound BMS design
State estimation is only one part of a BMS. The final system must also manage thermal behavior, cell balancing, charge and discharge limits, fault diagnosis, communications, and safe operating boundaries.
An accurate SOC estimate does not compensate for poor protection logic or inadequate pack-level monitoring.
Making the Right Choice for Your Goal
Use the testing and identification workflow to select the estimator based on the battery’s behavior and the BMS’s implementation constraints.
- If your primary focus is efficient embedded implementation: Start with a well-identified equivalent-circuit model and an EKF, then validate its linearization accuracy across the intended operating range.
- If your primary focus is nonlinear estimation accuracy: Use the same high-quality identified model with a UKF and compare its improvement against its additional computational cost.
- If your primary focus is high-rate pulse tracking: Include aggressive dynamic profiles and evaluate adaptive-gain EKF strategies only after tuning them against measured data.
- If your primary focus is production robustness: Test across SOC, temperature, current rate, rest periods, and aging conditions rather than validating on a single nominal profile.
- If your primary focus is model credibility: Separate parameter-identification data from validation data and examine both voltage residuals and SOC-estimation error.
Reliable EKF and UKF battery estimation begins not with the filter equations, but with representative tests, accurate measurements, and a model that faithfully captures the battery’s dynamics.
Summary Table:
| Aspect | EKF | UKF |
|---|---|---|
| Model Approximation | Local linearization using Taylor series (Jacobians) | Sigma-point propagation through nonlinear model, no Jacobians |
| Computational Load | Lower | Higher (multiple sigma points) |
| Nonlinear Handling | May degrade under strong nonlinearity | Better captures nonlinear behavior |
| Data Requirements | Same: requires accurate battery model, initial states, noise assumptions | Same: requires accurate battery model, initial states, noise assumptions |
| Best Use Case | Efficient embedded BMS with mild nonlinearity | High nonlinearity or when EKF linearization errors are significant |
Ready to enhance your battery state estimation? KINTEK provides comprehensive laboratory equipment for battery R&D, including advanced battery test systems and precision tools. Our solutions support dynamic profiling, parameter identification, and validation for EKF/UKF algorithms—helping you achieve accurate SOC estimation and robust BMS development. Contact us today to learn how our equipment can accelerate your research and improve your state-of-charge algorithms.