Abstract:

The second life use of lithium-ion batteries has gained significant attention in recent years, driven by the potential to repurpose cells from electric vehicles for less demanding applications. A critical aspect of this repurposing is accurately estimating the Remaining Useful Life (RUL) of the batteries. Traditional techniques often rely on data from the battery's first life, which may not be available in practical scenarios. To address this issue, we propose a data-driven method for RUL estimation that does not depend on first-life information. Our approach considers a realistic scenario where an aged battery cell, lacking previous usage data, is evaluated for second life use through a limited number of test cycles. We compute features such as incremental capacity curves, and other health indicators from the measured voltage and current waveforms of the used cell. These features are automatically processed by deep learning algorithms, including Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. This methodology achieves an average error of only 62 cycles for cells with a lifespan of up to 1200 cycles and a RUL error of less than 10% for deeply aged batteries. These results outperform state-of-the-art algorithms that utilize data from the cell's entire lifespan, demonstrating the efficacy and robustness of this approach.