Data sufficiency for transferable lithium-ion battery periodical SOH estimation under resource constraints

Battery state-of-health (SOH) estimation is vital for the safety of energy storage systems, yet full-life-cycle data curation remains resource intensive. Here, we present an empirical examination of data sufficiency (DS) to identify the data amount needed for SOH estimation algorithms with anticipated predictability and transferability. DS is defined as a linear function of normalized predictability and transferability of physical features. The effectiveness of DS is validated on 7 datasets (involving 310 batteries over 300,000 cycles), encompassing 6 materials (i.e., LFP, NCM811, NCM333, NCM523, NCA, and LCO) and 7 transfer scenarios (i.e., temperature, charging rate, discharging rate, and cutoff voltage). Results show that, on average, no more than 8% of lifetime data can achieve a median mean absolute percentage error (MAPE) of 1% under the investigated transfer scenarios, with the calculated DS aligning with post hoc DS. This work suggests the careful evaluation of DS for building data-driven battery algorithms before massive, expensive, and time-consuming data curation.