Find and Use Public Battery Datasets for Your Project

Why this happens

Battery ageing takes months of controlled cycling, which few students can run. Universities and research labs have published cycling and ageing data for exactly this reason, and using it lets you work on estimation and prediction methods with data that has proper test conditions behind it.

What you need

Step by step

1. 1

   Pick datasets that match your question

   Decide what you want to study: capacity fade over cycles, internal resistance growth, temperature effects or drive-cycle behaviour. Commonly cited public sources include the NASA prognostics battery data, the University of Maryland CALCE battery data, the Oxford battery degradation data and large fast-charging cycle-life datasets published by university groups. Search for these by name through your college library or search engine and download only from the publishing institution.

2. 2

   Read the documentation before the data

   Each dataset comes with notes on cell chemistry, format, nominal capacity, temperature, charge and discharge protocols and file structure. Write these down in a table. Without them you cannot interpret the numbers, and mixing cells tested at different temperatures or rates is a common source of misleading conclusions in student reports.

3. 3

   Check the licence and citation request

   Most datasets ask you to cite a specific paper or data record. Note the required citation and any licence terms in your project folder now, so your report credits the source properly. Some datasets are free for research and teaching only, which matters if you later want to use your work commercially.

4. 4

   Load and inspect the files

   Many datasets come as MAT, CSV or HDF5 files. Use pandas or SciPy to load one cell first. Print the column names, plot voltage, current and temperature against time for a single cycle, and confirm the plots look like a normal charge and discharge. Fix unit conversions now, such as milliamp-hours versus amp-hours or seconds versus hours.

5. 5

   Extract per-cycle features

   For each cycle compute discharge capacity by integrating current over time, plus features such as average temperature, end-of-charge voltage and an internal resistance estimate if the data includes pulses. Store one row per cycle in a tidy table. Plot capacity against cycle number for each cell; this capacity-fade curve is the backbone of most state-of-health projects.

6. 6

   Split data honestly for modelling

   If you train a model, split by cell rather than by cycle, so the test cells are completely unseen during training. Splitting cycles of the same cell between training and testing gives flattering results that do not hold up. Report the split clearly, along with simple baselines such as a straight-line fit, so readers can judge whether the model adds value.

7. 7

   Document limits and conclusions

   Lab cells are cycled under controlled conditions that differ from an Indian scooter in summer traffic. State the chemistry, temperature and protocol limits of your conclusions, and avoid claiming your model will predict a real vehicle's battery life. Keep your processing script, the cleaned table and the citation together so the work can be checked.

How to tell it worked

You have a clean per-cycle table, capacity-fade plots for several cells, a documented test-condition summary and correct citations ready for your report.

Questions