Lifetime is a critical requirement for devices relying on lithium ion batteries. However, lifetime testing is a major bottleneck in battery development due to the large number of usage conditions, the large number of repetitions required due to cell-to-cell variation, and the long length of experiments.
Our work, aimed at optimizing battery fast charging, attempts to address both the long testing times and the large number of required experiments using machine learning techniques. First, we developed machine learning models to predict the final cycle life using data from the first 100 cycles. These models, which rely on transformations of voltage-capacity curves (Figure 1), allow for predictions using 10x less data with <10% error.
Figure 1
We then combined early prediction with optimal experimental design in a closed-loop system (Figure 2) to efficiency find the highest-fetime fast-charging protocols out of 224 total candidates.
Figure 2.
By reducing the number of cycles required per experiment (using early prediction) and the total number of experiments (using optimal experimental design), we can tackle expensive battery optimization challenges with a ~15x reduction in time.
Data-driven prediction of battery cycle life before capacity degradation
Closed-loop optimization of fast-charging protocols for batteries with machine learning
Be First to Comment