Battery Prognostics ML

A Python research scaffold for battery state-of-health (SOH) estimation, remaining useful life (RUL) prediction, impedance feature analysis, and physics-informed machine learning.

The repository is being developed as a practical starting point for reproducible experiments. The current release provides data-validation utilities, feature-processing helpers, baseline model components, and evaluation functions. It does not claim validated performance on a particular battery dataset.

Scope

• Inspect cycle-level battery data and check required columns.
• Derive simple degradation and impedance summaries from user-provided data.
• Establish baseline sequence-model components for RUL experiments.
• Evaluate predictions with standard regression metrics.
• Keep data, notebooks, source code, and reported results separate for reproducibility.

Repository layout

battery-prognostics-ml/
├── data/                 # Data instructions; datasets are not included
├── notebooks/            # Starter notebooks for analysis and modelling
├── src/                  # Reusable Python modules
├── results/              # Notes on storing experiment outputs
├── requirements.txt
└── README.md

Setup

Python 3.10 or newer is recommended.

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\\Scripts\\activate
python -m pip install --upgrade pip
pip install -r requirements.txt

Data

Place data you are permitted to use in a local data directory and follow the schema described in data/README.md. Raw datasets are intentionally not committed. Do not upload private, licensed, or restricted research data without permission.

The utilities expect tabular cycle-level data. Dataset-specific column names and end-of-life definitions should be stated in each experiment rather than assumed to be universal.

Start here

1. Run notebooks/01_data_exploration.ipynb to inspect a local CSV file.
2. Use notebooks/02_feature_engineering.ipynb to review basic feature-processing examples.
3. Use notebooks/03_rul_prediction_lstm.ipynb to see the structure of a baseline sequence-model experiment.

The notebooks contain starter workflows, not published experimental results. Update paths and column names to match your data before running them.

Methodological notes

• Split data by battery when evaluating cross-battery generalization; random row splits can leak information between train and test sets.
• Define RUL and end-of-life consistently with the dataset protocol and document any post-EOL observations.
• Fit preprocessing steps on training data only.
• Report the evaluation protocol, sample counts, metrics, and uncertainty where appropriate.
• A physics-informed loss must encode a stated physical relationship and be tested against a suitable baseline. A class name alone does not make a model physics-informed.
• Do not report benchmark claims until the experiment has been run and independently checked.

Current status

This is an initial research toolkit. The data loaders and feature utilities provide basic reusable functionality; model components are starting points that require dataset-specific training, validation, and tuning. No performance claims are made in this repository at this stage.

License

Released under the MIT License. See LICENSE.