Fig. 3: Effectiveness of safety fault detection. | Nature Communications

Fig. 3: Effectiveness of safety fault detection.

From: Model-constrained deep learning for online fault diagnosis in Li-ion batteries over stochastic conditions

Fig. 3

a, b present the t-SNE visualization of the inputs and outputs of the proposed network. The axes represent the two-dimensional space into which the data is compressed. Blue dots represent normal moment samples, while red dots represent fault moment samples. The left side of c–f display the time-series data of battery voltage and temperature variations, as well as the output results of the model comprehensive index for four safety-faulted vehicles, namely, ISC, EA, TR, and EL, respectively. The output results of threshold 1, threshold 2, and threshold 3 correspond to sequences of 3σ, 4.5σ, and 6σ, respectively. The red point indicates the triggering threshold fault point. The four types of safety faults are recognized in advance to varying degrees. The right side of c–f show the real-time probability estimation results for four types of battery safety faults. Fill with blue when the probability is greater than 60%, and fill with green when the probability is less than or equal to 60%. Source data are provided as a Source Data file.

Back to article page