Software / AI

Availability of Open Data Repositories and Synthetic Data Generation Accelerate AI/ML Driven Applications The availability of open datasets / repositories and efforts into standardization are accelerating several AI / ML applications of battery research as the demand for energy storage accelerates. In addition to lab data, increased diversity of available field data is crucial for advancing battery management studies in real-world environments.

Institutional Repositories

The “Lab-to-Fieldˮ Gap & Synthetic Data Generation

Battery Archive Battery DataHub Open Datasets Oxford Battery Degradation Dataset CALCE Experimental Test Data NREL Battery Failure Databank Data Standardization & Tools

The limited availability of large-scale, high-quality field data is a primary bottleneck for BMS development. This table outlines some latest developments in synthetic data generation that aim to address this: Architecture

Methodology & Innovation

Lab-to-Field Impact

TimeGAN / RCGAN

Uses Recurrent Conditional GANs to generate multivariate time-series conditioned on auxiliary variables (e.g., Temperature, SOC.

Synthesizes scarce "edge case" data (e.g., extreme cold profiles) to robustify models against unseen field conditions.

CPIGAN

Embeds "Five Basic Physics Constraints" (e.g., Monotonicity, Energy Conservation) directly into the generator's loss function.

Prevents "physical hallucinations," ensuring generated degradation paths obey thermodynamic laws.

LLMs SDForger)

"Textualizes" voltage signals, treating time-series forecasting as a natural language token prediction task.

Captures long-range temporal dependencies and enables multimodal analysis (integrating maintenance logs with voltage data).

BattINFO LF Energy Battery Data Alliance

Example AI / ML Applications PINNs for Battery Diagnostics: Physics-Informed Neural Networks PINNs are nearly 1,000 times faster than traditional physics models, making them ideal for real-time diagnostics. Mamba Model: a more efficient alternative to Transformers, for early Remaining Useful Life RUL) prediction (trained on CALCE CS2 dataset).

Ampworks Source: BattGenie, PINNs, NREL, Liang Y, Zhao S., Energies 2024, Liu H, Li C et al., Nat Commun, 2025, Mohanty, Prasanta & Jena et al., IEEE Transactions on Industry Applications, 2025, Myisha A. Chowdhury, Gift Modekwe, and Qiugang Lu, 2025, Xiong, Jiawei & Fink et al., Mechanical Systems and Signal Processing, 2023, Cécile Rousseau, Tobia Boschi, et al. NeurIPS, 2025

Extracted from the PDF; charts and images may read imperfectly.