Interpretable Machine Learning Framework for Battery State of Health Estimation Using Graphene Sensors
The repository includes code that can be used to perform experiments with models for Soh estimation and the dataset (raw and processed) for such analyses. Our dataset includes the temperature readings from the thermoucuples (of batteries and ambient temperature), from the original sensing approach based on the DZP Technologies' GR92-001 graphene ink sensor and SoH. The dataset can be used to reproduce our results and for other future works.
It is a part of the supplementary materials for the following paper:
Interpretable Machine Learning Framework for Battery State of Health Estimation Using Graphene Sensors (Katarzyna Filusa, Obinna Nwadiutob, Zlatka Stoevac, Joanna Domanskaa, Fideline Tchuenbou-Magaiab).
Abstract: Enabling accurate monitoring of lithium-ion batteries is vital for the safety and performance of devices in consumer electronics and large-scale applications. Conventional battery management systems often struggle to detect health degradation at the individual cell level, as they rely heavily on data such as voltage profiles that are typically unavailable during normal operation and introduce considerable hardware complexity. To address these deficiencies, we propose to use data from our percolative graphene ink-based sensor combined with lightweight machine learning methods for battery-level surveillance. We introduce a novel framework leveraging explainable artificial intelligence for estimating the State of Health (SoH) of batteries. It uses polynomial features derived from the sensor resistance measurements as input variables, enabling a physics-informed interpretation. Explainability techniques are leveraged to provide insights into the model’s decision-making process and ensure transparent SoH estimation. Our approach achieves reliable SoH prediction without the need for extensive historical datasets and specialist laboratory-acquired data. Thus, it demonstrates great potential for practical and scalable deployment in real-world use cases.
The repository consists of the following elements:
- process_data.py - script that processes the data by extracting the polynomial coeeficients (the degree can be set)
- resistance_ml.py - script that trains, tests and presents the results of the ml models and xai methods
- DATA - the folder consists of raw and processed data