Diagnosis of Abnormal Overcharge Internal Pressure in Lithium Battery Based on Physics‐Informed Neural Network's Residual Distribution

Abstract

Changes in the internal pressure of lithium batteries often reflect the state of the batteries. Rapid diagnosis of abnormal internal pressure is importance for battery safety. This article proposes a battery overcharge internal pressure abnormality diagnosis method based on the detection of safety vent strain. First, this method establishes a multiphysical model between the strain of the battery's safety vent, surface temperature rise, and state of charge (SOC) under normal internal pressure. Second, a physics‐informed neural network is used to enhance the multiphysical model's accuracy. Third, the residual sequence generated from actual measurements and estimated values, along with an improved quartile range, is used to diagnose internal pressure abnormalities. Battery overcharge experiment results show that this method can diagnose internal pressure abnormalities within 260 s after overcharging and alarm. At this time, the battery has slight bulging, and the SOC is 102.1%. By conducting charging experiments under different conditions, the low false alarm rate of this method is only 0.575%. The proposed method can identify overcharging abnormalities before the lithium battery safety vent is opened, which greatly ensures the safe operation of the battery system.