Unlocking the potential of unlabeled data: Self-supervised machine learning for battery aging diagnosis with real-world field data

Unlocking Unlabelled Battery Field Data

How can we unlock the full potential of vast, unlabeled battery field data?

Despite significant advancements achieved by data-driven methods, diagnosis accuracy remains constrained by the high costs of check-up tests and the scarcity of labeled data.

In this paper experimental verification is performed on the battery degradation dataset from 20 commercial EVs, collected over more than two years. The proposed framework enables accurate and robust aging diagnosis, requiring only the labeled data from two EVs.

overview of the application framework of the proposed battery degradation diagnosis method in field operation with unlabeled data. (b) Visualized frameworks of self-supervised, semi-supervised, and supervised machine learning
 overview of the application framework of the proposed battery degradation diagnosis method in field operation with unlabeled data. (b) Visualized frameworks of self-supervised, semi-supervised, and supervised machine learning

This is a high level overview of the paper: Qiao Wang, Min Ye, Sehriban Celik, Zhongwei Deng, Bin Li, Dirk Uwe Sauer, Weihan Li, Unlocking the potential of unlabeled data: Self-supervised machine learning for battery aging diagnosis with real-world field data, Journal of Energy Chemistry, Volume 99, 2024, Pages 681-691, ISSN 2095-4956

Why it Matters

Ensuring the health and safety of lithium-ion batteries in EVs is essential for long-term efficiency and reliability. However, current diagnostic methods are limited by the high costs of labeled data and testing procedures. This work focuses on unlocking the potential of unlabeled data—making diagnostics more affordable and scalable in real-world applications.

Contribution

  • Developed a self-supervised machine learning approach that improves diagnostic accuracy by 74.54% (best case) and 60.50% (worst case) compared to supervised methods.
  • Uses labeled data from just two EVs to provide accurate battery aging estimates, significantly reducing costs.
  • Validated over two years of data from 20 commercial EVs, addressing real-world challenges like cell inconsistencies, charging uncertainties, and field conditions.
  • Leverages random short charging sequences within a specific voltage window using incremental capacity (IC) analysis, ensuring practical applicability with physical interpretation.
comparative analysis of the proposed framework and benchmarks for unlocking battery data
Comparative analysis of the proposed framework and benchmarks. (a–c) Comparison results of capacity estimation. (d–f) Comparison results of MAXE, RMSE, and MAE in the best performance. (g–i) Comparison results of MAXE, RMSE, and MAE in the worst performance.

This framework shows that unlabeled data can be a powerful, low-cost resource for ensuring the health and safety of lithium-ion batteries, revolutionizing diagnostics for large-scale EV operations.

Unlocking the potential of unlabeled data: Self-supervised machine learning for battery aging diagnosis with real-world field data

Qiao Wang, Min Ye, Sehriban Celik, Zhongwei Deng, Bin Li, Dirk Uwe Sauer, Weihan Li, Unlocking the potential of unlabeled data: Self-supervised machine learning for battery aging diagnosis with real-world field data, Journal of Energy Chemistry, Volume 99, 2024, Pages 681-691, ISSN 2095-4956

Leave a Comment