Estimating the State of Health (SOH) of Lithium-ion batteries is critical for ensuring their reliability and extending their operational lifespan. This poster presents a methodology to estimate SOH without requiring additional degradation experiments, leveraging the open NASA battery dataset to validate the approach. The study focuses on unsupervised domain adaptation to transfer trained knowledge effectively to new datasets. The core data used is the incremental capacity (IC) curve derived from full charging cycles, which provides a rich representation of battery behavior.

One proposed strategy involves transferring charging cycles closer to laboratory conditions using a physical model trained on a limited field-data aged cells. The physical model is built according to equivalent circuit approaches to track the voltage of the cells during cycling based on the limited field-data information. The predicted performance under simulated laboratory conditions is then fed into a data-driven model to estimate SOH. Additionally, the study explores an alternative approach using feature-based unsupervised domain adaptation to identify optimal configurations for accurate and robust SOH estimation when transferring knowledge from laboratory environment to the field data. By comparing these techniques and benchmarking them against suitable baselines, the research aims to develop a versatile framework that bridges gaps between lab-based and real-world battery datasets, advancing the state-of-the-art in battery health estimation.