Thermal Runaway Early Warning for Lithium-Ion Batteries Upon Strain Perspective

With the widespread application of lithium-ion batteries, the issue of thermal runaway has garnered increasing attention. Strain signals, due to their low latency and high sensitivity, hold significant potential for the early warning of thermal runaway and have been extensively studied. However, the inherent complexity and substantial individual variations of strain signals make it challenging to effectively identify early anomalies, often leading to delays in warnings or false alarms. This article addresses these challenges by proposing a novel fragment attention perception (FAP) neural network. The network incorporates an innovative heterogeneous attention mechanism that extracts individualized battery characteristics from historical data, enabling accurate strain signal predictions, which allows for proactive thermal runaway warnings. Experimental results under conditions of overcharge and overrate demonstrate that the proposed method can provide an early warning for thermal runaway more than 8 h in advance, achieving an accuracy rate exceeding 96%.