Abstract
Although the hazardous failure of lithium-ion batteries is rare, the fallout can be severe. The safety and reliability of lithium-ion batteries are more important now than ever because of their widespread adoption, yet our ability to predict failure through online and offline diagnostics is still very limited. Lithium-ion batteries are highly complex, nonlinear systems. To make matters worse, two cells of identical geometry, chemistry, and history might respond differently when exposed to identical mechanical, thermal, or electrical stimuli. This limits the value of classical deterministic modeling techniques. Applying a probabilistic approach allows for quantification of uncertainty to support decisions in design and control. Machine-learning algorithms are well suited for predicting nonlinear systems like lithium-ion cells, but training and validation of algorithms are challenging for safety applications because large amounts of failure data are needed. Even if the algorithms predict accurately, machine learning is typically agnostic to underlying physics and thus presents limited value in informing researchers and engineers on design opportunities to improve the cells’ performance. There is much interest within the battery research community in tackling these challenges, and this perspective aims to offer suggestions on promising avenues of investigation to achieve accurate predictions of the risk of cell failure while gaining some physical insights into the predicted behaviors.
Original language | American English |
---|---|
Pages (from-to) | 316-329 |
Number of pages | 14 |
Journal | Joule |
Volume | 5 |
Issue number | 2 |
DOIs | |
State | Published - 17 Feb 2021 |
Bibliographical note
Publisher Copyright:© 2020
NREL Publication Number
- NREL/JA-5700-76901
Keywords
- battery failure
- data-driven methods
- diagnostics
- lithium-ion batteries
- machine learning
- modeling
- safety