Publication Date
Spring 2025
Degree Type
Master's Project
Degree Name
Master of Science in Computer Science (MSCS)
Department
Computer Science
First Advisor
Saptarshi Sengupta
Second Advisor
Navrati Saxena
Third Advisor
Ching-Seh Wu
Keywords
Li-ion Battery Prognostics, State of Health Estimation, Remaining Useful Life, Transformers for RUL, LLM in Prognostics, LLM in Predictive Maintenance
Abstract
In this thesis, we present a comparative analysis of the use of transformerbased and Large Language Model (LLM) models for State of Health (SoH) and Remaining Useful Life (RUL) prediction of lithium-ion batteries. With electric cars and renewable energy systems based on batteries at the forefront, the need to predict degradation accurately in order to enhance the performance and reduce maintenance costs has become imperative. Most traditional prediction methods lag the complex and non-linear characteristics of degradation in batteries, and hence the usage of sophisticated methods becomes a necessity. The research employs the CALCE dataset, which includes long-horizon cycling data for four lithium-ion batteries (CS2 35, CS2 36, CS2 37, CS2 38), to develop and evaluate two new methodologies. First, a transformer-based model is used, combining the capabilities of autoencoder features, position encoding, and multihead attentions to successfully identify temporal patterns related to battery aging. Second, a new approach is explored that takes advantage of pretraining of large language models (LLMs) via zero-shot prediction and fine-tuning methods. Visual inspection shows that both approaches skillfully capture the characteristic three-phase degradation curve of lithium-ion batteries: an initial stabilization period, a phase of gradual degradation, and a final accelerated decrease. The models successfully identify key turning points where rapid degradation begins, thus providing valuable information for maintenance policies and replacement schedules. Comparative evaluation indicates that, while transformers have more consistent behavior across different batteries, LLMs provide better uncertainty quantification and better generalization to unforeseen degradation patterns. The research highlights the need for choosing specific models to match specific application needs, with transformers performing better in some situations. This work contributes to the field by establishing a framework for advanced battery health prediction that can significantly reduce the time and resources required for degradation testing while improving the accuracy of RUL estimation in practical battery management systems.
Recommended Citation
MAHAGAONKAR, CHINMAY NILESH, "Transformer Integration, Fine-tuning and Zero-Shot Learning for State of Health Estimation in Li-ion Batteries using Large Language Models" (2025). Master's Projects. 1524.
DOI: https://doi.org/10.31979/etd.ed74-jfrt
https://scholarworks.sjsu.edu/etd_projects/1524
