查看文章

Early prediction of battery lifetime would aid in their development, manufacture, and optimization. Furthermore, for high-rate applications such as fast charging, first-principles electrochemical models fail to capture the dynamics. In this work, we develop machine learning models to predict the final cycle life using the first 20-100 cycles for a dataset consisting of commercial lithium-ion batteries cycled to failure during fast charging. We develop a novel visualization of voltage data, identifying a linear trend that is useful for early prediction. Our best models achieve around 12% mean percent error for cycle numbers from 40 to 100. This work demonstrates high prediction accuracy at low cycle number, accelerating battery high-throughput screening applications.