Fig 1.
Systematic schematic diagram of the overall architecture.
Fig 2.
LIB model.
Fig 3.
Simplified FD model for battery pack.
Fig 4.
2DCNN structure.
Fig 5.
Convolution operation process.
Fig 6.
LBFD processprocess. ((Icons in the picture are sourced from:
Table 1.
The experimental basic environmental parameters.
Table 2.
Performance comparison with state-of-the-art methods.
Table 3.
Quantitative comparison between feeding raw data directly into a CNN and the method proposed in this study.
Table 4.
Operation performance and signal decomposition effect.
Table 5.
Comparison of computational efficiency and memory consumption.
Fig 7.
Method training loss testing.
Fig 8.
Loss function and accuracy of each training set.
Fig 9.
Consistency of extracting data feature information.
Fig 10.
Accuracy of fault feature identification.
Fig 11.
Runtime analysis.
Fig 12.
Analysis of the proportion of misidentified content.
Fig 13.
Comparison of confusion matrix of Test Set A.
Fig 14.
Visualization process of LBFD.
Table 6.
Statistical analysis of model attention distribution for different fault types based on Grad-CAM.
Table 7.
Performance breakdown by fault category on the combined test set.