Lee et al., Batteries 9(5), 264 (2023), Figure 2. CC BY 4.0.

What the study explores

Place the model in a system

The proposal connects generalized and personalized DNNs, offline training, and ECU, edge, and cloud roles.

Examine output plausibility

Outlier detection, plausibility checks, and a BMS Luenberger-observer reference path are used to address estimation output bounds.

Distinguish experiments from implementation

The paper reports experiments across 65 DNN configurations on NASA cell-test data. Embedded implementation and broader pack- and vehicle-level validation were future work.

Lee et al., Batteries 9(5), 264 (2023), Figure 1. CC BY 4.0.

Questions for today’s AI safety work

The study provides a starting point for discussing AI function boundaries, allocation of safety requirements, monitoring limitations, and verification evidence. This is a retrospective connection to ISO/PAS 8800:2024, not a claim that the research demonstrates conformance.

The figure’s ASIL and safety-zone labels belong to the proposed design and do not establish an assessed implementation.

PUBLICATION

State-of-Health Estimation and Anomaly Detection in Li-Ion Batteries Based on a Novel Architecture with Machine Learning

Junghwan Lee et al. · 2023

https://doi.org/10.3390/batteries9050264

CC BY 4.0