Answers are grounded in this paper and cite their sources.
Reliable remaining useful life estimation is critical for safe and cost-effective battery management, enabling timely maintenance, warranty planning, and preventing unexpected failures in electrified transportation and energy storage systems. However, accurate reliable remaining useful life estimation of lithium-ion batteries is challenging due to distribution shifts across datasets and operating conditions. This work examines whether time-series foundation models can deliver capacity forecasting. We study TimesFM under six evaluation pipelines that span within-dataset and cross-dataset transfer, with and without source/target adaptation. Experiments on four public benchmarks use the first 30% of each target life as history and recursively predict the remaining 70%. Standard accuracy metrics are reported and compared with strong baselines. We further conduct a sensitivity analysis for TimesFM by varying target prefix ratios (30%, 20%, and 10%). Results show that TimesFM captures long-horizon degradation trends; source-adapted transfer yields the best performance when source and target are from the same dataset, while cross-dataset transfer degrades accuracy. Incorporating target-prefix fine-tuning yields a modest but consistent drop, and the sensitivity analysis preserves the same trend across split ratios.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: