Thermal Runaway Stage Identification of Lithium-Ion Batteries Based on Principal Component Analysis

Abstract

To address the issues of high computational complexity and redundancy caused by high-dimensional data in multi-source sensing data fusion for battery thermal runaway identification, this paper proposes a dimensionality reduction fusion method based on Principal Component Analysis (PCA). By conducting thermal runaway experiments on lithium-ion batteries and synchronously acquiring multi-physical field signals such as temperature and gas concentration, a high-dimensional monitoring matrix was constructed. The proposed multi-source data fusion method retains 90.521% of the information while reducing ten-dimensional sensing data to three dimensions, achieving feature simplification and physical correlation analysis. Based on the fused multi-source sensing features and extracted principal components, a comprehensive score curve was constructed to quantify the thermal runaway evolution process by identifying key feature points. A multi-stage logical criterion fusion method was proposed: the triple joint criteria using features F1, F4, and F6 for gas accumulation stage identification achieved an average deviation of 10.73%; the criteria F2, F5, and F7 for pressure relief stage detection reached 91.67% zero-deviation accuracy; and the dual criteria F3 and F8 enabled precise thermal runaway stage identification with an average deviation of 12.05%.

Keywords: Lithium-ion batteries, Thermal runaway, PCA, Data fusion, Feature extraction, Multi-stage detection