Battery model data checklist

1. Define the prediction you need

Record the cell identifier, chemistry, nominal capacity, sample history and the intended temperature, state-of-charge (SOC) and current ranges. Describe the output you need: terminal-voltage response, a drive-cycle comparison, or another explicit engineering question.

An equivalent-circuit model (ECM) and a physics-based model such as SPMe do not have identical data requirements. Record which parameters are measured, taken from literature, fixed by assumption, or estimated. A good voltage fit alone does not establish that each physical parameter is uniquely identified.

2. Collect complementary measurements

Capacity and reference conditions

Include measured capacity, test temperature, current, voltage limits and the procedure used to establish the SOC reference. Keep measured capacity distinct from a datasheet rating.

OCV or slow-rate reference data

Label how the curve was obtained: rested open-circuit measurements or low-current charge/discharge. Include direction, temperature and rest criteria. Slow-rate terminal voltage is an approximation to equilibrium OCV, not automatically the same measurement.

HPPC or other pulse data

Retain current, voltage, time and available temperature channels before, during and after pulses. Record SOC, pulse magnitude and duration, charge/discharge direction and rest periods. HPPC voltage response is commonly used to estimate ECM parameters across operating conditions.

Independent validation profiles

Reserve measured responses to a separate drive or duty cycle that reflects the intended application. Record the initial conditions and the same units and sign conventions as the fitting data.

Choose the test protocol with the lab and model owner. This checklist does not specify safe cell operating limits, pulse sizes or a universal sampling rate.

3. Check the files before fitting

• Time: identify the time column and units; flag resets, duplicate timestamps and gaps.
• Current: state whether positive means charge or discharge. Check the stated convention against the test steps.
• Units: distinguish A from mA, V from mV, seconds from hours, and Ah from mAh.
• Temperature: distinguish measured cell temperature from chamber setpoint; record sensor location when known.
• Coverage: list the temperatures, SOC intervals and loads actually measured. Flag missing or truncated runs rather than silently treating them as complete.
• Provenance: preserve raw exports and document every conversion, filter, exclusion and resampling step.

4. Keep validation independent

Name the fitting files and held-out validation files before tuning. If you repeatedly tune against the validation set, reserve another untouched check before reporting generalization.

Report the error metric, units, evaluated time window, initial conditions and results by temperature and profile. Include failed runs and excluded intervals. A single average can hide poor performance in an operating region that matters to the application.

Record the model and parameter version alongside the results. Predictions outside tested conditions need additional evidence; a small voltage error is not a safety certification.

5. Send a reviewable handoff

The downloadable Markdown template gives you a place to record the prediction task, cell context, file inventory, preprocessing, validation split and open questions. It contains no sample measurements or claimed results. Leave unknowns explicit so the receiving team can resolve them.

Need measurements collected? Explore AmpLabs cell testing. Already have data? Explore cell simulation and model fitting. AmpLabs is in private beta; request access to discuss your workflow.